A charging station image construction and anomaly detection method based on multi-dimensional feature fusion

Abstract

The application discloses a charging station portrait construction and abnormality detection method based on multi-dimensional feature fusion, comprising the following steps: S1, receiving input data of the charging station and performing data preprocessing and multi-dimensional feature vector extraction; S2, performing standardization processing on the extracted multi-dimensional feature vector to eliminate the dimensional difference between different features; S3, adopting a hierarchical clustering algorithm to cluster and classify the charging stations; S4, adopting a contour coefficient to evaluate the clustering quality of the charging stations; S5, constructing a standard portrait template for each cluster; S6, constructing an isolated forest model to perform abnormality detection; S7, constructing a self-encoder network to perform auxiliary abnormality detection; S8, integrating the detection results of the isolated forest and the self-encoder, and adopting a weighted fusion strategy to perform integrated abnormality judgment. The method solves the problems of inaccurate site feature identification, lagging equipment fault response, lack of data support for operation decision, low capacity utilization and insufficient fine operation capability in the operation of the existing charging station.

Description

Technical Field

[0001] This application belongs to the field of charging load prediction technology for charging stations, specifically referring to a method for constructing charging station profiles and detecting anomalies based on multi-dimensional feature fusion. Background Technology

[0002] With the global energy transition and the advancement of national carbon neutrality goals, the electric vehicle industry has entered a stage of rapid development. As a crucial link in the electric vehicle industry chain, the construction and operation level of charging infrastructure directly impacts the promotion, popularization, and user experience of electric vehicles. In recent years, the number of charging piles in my country has continued to grow rapidly, exceeding 3 million public charging piles by the end of 2024, forming a charging network covering major cities nationwide. However, while charging stations are expanding rapidly, many problems in operation and management are gradually becoming apparent. On the one hand, there is a structural imbalance in the layout of charging stations. Provincial capitals and core commercial areas have high charging pile density, while development along highways, in urban and rural areas, and in residential communities is relatively lagging. Queues for charging occur in some hot spots, while charging piles in some areas remain idle for extended periods, with an overall utilization rate of less than 15%. On the other hand, charging station operators face difficulties such as long investment recovery periods, high operating costs, and weak profitability, making it difficult for traditional operating models to support the sustainable development of the industry.

[0003] There are several pain points in the current operation and management of charging stations that urgently need to be addressed. First, the identification of charging station characteristics is inaccurate. Operators struggle to accurately grasp the load characteristics and user needs of different regions and types of charging stations, leading to unreasonable resource allocation and a lack of scientific basis for capacity planning. Second, equipment failure response is lagging. "Zombie charging piles" account for 10% of the industry, and dumb charging piles account for over 30%. Low digitalization means traditional inspection methods cannot detect equipment anomalies in a timely manner, resulting in users frequently encountering charging piles that cannot charge normally, leading to low charging success rates and poor user experience. Third, operational decisions lack data support. Most charging station operations still rely on manual experience management, lacking systematic analysis of charging load patterns, user behavior characteristics, and the impact of the surrounding environment, resulting in insufficient refined operational capabilities. Fourth, maintenance costs remain high. Traditional air-cooled equipment has a lifespan of only 3 to 5 years, requiring operators to replace equipment before recouping their investment. Coupled with high manual inspection costs and untimely fault repairs, this further compresses operating profit margins. These problems hinder the improvement of charging station operational efficiency and the healthy development of the industry. Existing technologies have significant shortcomings in addressing the aforementioned problems. Traditional anomaly detection methods based on statistical thresholds rely solely on the mean and standard deviation of historical data to set fixed thresholds, failing to adapt to the dynamic changes and diverse characteristics of charging loads, resulting in high false alarm and false negative rates. Prediction methods based on single time-series models, such as LSTM and GRU, while capturing the dependencies in time series, ignore the spatial correlations between charging stations and the differences in various application scenarios, leading to limited prediction accuracy and weak generalization ability. Existing clustering analysis methods mostly use simple Euclidean distance metrics, which are difficult to accurately characterize the probability distribution characteristics of charging load curves and do not fully utilize the geographical location and surrounding environment information of charging stations, resulting in classification results that do not match actual application scenarios. Furthermore, most operation and management systems lack the ability to analyze the root causes of anomalies, only able to identify anomalies but unable to provide interpretable diagnostic results, making it difficult for maintenance personnel to quickly locate problems and take targeted measures. These technological limitations keep charging station operation and management in a rudimentary stage, unable to meet the industry's needs for refined and intelligent development. Summary of the Invention To overcome the shortcomings of existing technologies, this application provides a method for constructing charging station profiles and detecting anomalies based on multi-dimensional feature fusion, which solves the problems of inaccurate site feature identification, delayed equipment fault response, lack of data support for operational decisions, low capacity utilization, and insufficient refined operation capabilities in the operation of existing charging stations.

[0004] This invention provides a method for constructing charging station profiles and detecting anomalies based on multi-dimensional feature fusion, the method comprising: Step S1: Receive input data from the charging station and perform data preprocessing and multidimensional feature vector extraction; Step S2: Standardize the extracted multidimensional feature vectors to eliminate the differences in dimensions between different features; Step S3: Use a hierarchical clustering algorithm to cluster and classify the charging stations; Step S4: Use the profile coefficient to evaluate the clustering quality of charging stations; Step S5: Construct a standard profile template for each cluster; Step S6: Construct an isolated forest model for anomaly detection; Step S7: Construct an autoencoder network for auxiliary anomaly detection; Step S8: Integrate the detection results of isolated forest and autoencoder, and use a weighted fusion strategy to perform integrated anomaly determination.

[0005] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, step S1, receiving input data from the charging station and performing data preprocessing and multi-dimensional feature extraction, includes: Step S11: Receive input data; wherein the input data includes the latitude and longitude coordinates of N charging stations, hourly charging records within the sensed time period, and a preprocessed distance prior matrix; Step S12: Clean the received input data, calculate the average and standard deviation of the historical charging load for each charging station, and mark the charging load at a certain moment as an outlier and remove it when it deviates from the average by more than three times the standard deviation; for missing charging data, a spatiotemporal interpolation method is used to fill in the missing data. Step S13: Extract time features from historical charging records. For each charging station, extract its 24-hour load curve and record the average charging load for each hour. Identify the time when the peak occurs, i.e., the hour with the largest load in 24 hours. Step S14: Calculate the load volatility, which is the standard deviation of the load series divided by the mean. Step S15: Calculate the capacity utilization rate, which is the cumulative sum of charging load at all times divided by the number of time steps and then divided by the installed capacity. Step S16: Extract user behavior features, including average single charging time and fast charging ratio. The fast charging ratio is equal to the number of fast charging times divided by the total number of charging times. Step S17: Extract POI features from the electronic map, call the API interface of the electronic map to obtain 20 types of POI data within a 500-meter radius of each charging station; for each charging station, count the number of each type of POI within a 500-meter radius and construct a 20-dimensional POI feature vector. Step S18: Calculate the POI diversity index using the Shannon entropy method. First, calculate the proportion of each type of POI to the total POIs, and then multiply the proportion by the negative of its logarithm and sum them up.

[0006] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, in step S2, standardizing the extracted multi-dimensional feature vectors to eliminate the differences in dimensions between different features, includes: The Z-score standardization method is used to calculate the mean and standard deviation of each feature vector across all charging stations. Then, the mean is subtracted from each feature value and the standard deviation is divided to make the standardized feature mean zero and the standard deviation one. A comprehensive feature matrix is ​​constructed by integrating time features, load features, user behavior features, and POI features; the number of rows in the comprehensive feature matrix is ​​N, which is the number of charging stations, and the number of columns is D, which is the total feature dimension.

[0007] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, step S3, which involves clustering and classifying charging stations using a hierarchical clustering algorithm, includes: A similarity measurement method between charging stations is defined, and the similarity of the 24-hour load curves is calculated using Jensen-Shannon divergence. That is, for the load curves of two charging stations, they are first normalized to a probability distribution, that is, the load value of each hour is divided by the total load of 24 hours; then the mixture distribution of the two probability distributions is calculated, that is, the average value of the hourly load. The comprehensive distance between charging stations is defined by combining comprehensive feature similarity and spatial prior constraints. The comprehensive distance consists of three parts: JS divergence with a weight of 60%, Euclidean distance of standardized feature vectors with a weight of 40%, and spatial constraint terms based on distance prior matrix with a weight of 20%. Hierarchical clustering uses the Ward connection criterion, selecting the two clusters with the smallest increase in intra-cluster variance after merging each time.

[0008] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, step S4, evaluating the clustering quality of charging stations using contour coefficients, includes: For each charging station, first calculate its average distance to other stations within the same cluster, called the intra-cluster distance; then calculate its average distance to all stations in its nearest neighbor cluster, called the inter-cluster distance; the profile coefficient is equal to the inter-cluster distance minus the intra-cluster distance, and then divided by the larger of the two values; the profile coefficient ranges from negative one to positive one, the closer the value is to positive one, the more consistent the station is with the height of its own cluster and the better separated from other clusters, the closer the value is to zero, the more the station is on the cluster boundary, and the negative value indicates that there may be an allocation error; the overall profile coefficient is the average of the profile coefficients of all charging stations; The cluster number K is iterated from 3 to 10, and the overall silhouette coefficient corresponding to each K value is calculated. The K value that maximizes the silhouette coefficient is selected as the optimal number of clusters. At the same time, the Davies-Bouldin index is used as an auxiliary evaluation. The smaller the index, the better the clustering effect.

[0009] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, step S5, constructing a standard profile template for each cluster, includes: Extract the load curve of the center of the pattern, which is the average of the 24-hour load curves of all charging stations in the cluster, to form a representative load curve for this type; Calculate the load fluctuation range, and calculate the standard deviation of the load of all stations in the cluster for each hour. The larger the standard deviation, the greater the load difference between different stations during the time period; the smaller the standard deviation, the more consistent the load of this type of station during the time period. Extract the typical POI feature distribution, calculate the average value of the feature vectors of all charging station POIs in the cluster, and form a 20-dimensional typical POI distribution vector.

[0010] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, step S6, constructing an isolated forest model for anomaly detection, includes: An isolated forest consists of no fewer than 100 isolated trees, each of which recursively divides the data by randomly selecting features and split points; Calculate the average path length of a data point across all isolated trees; Define a normalization factor that represents the theoretical average path length of a binary search tree containing n samples, calculated using the harmonic number; The anomaly score is equal to the negative exponent of two, where the exponent is the average path length divided by the normalization factor. The anomaly score ranges from zero to one. When the anomaly score is greater than 0.6, it is considered a general anomaly; greater than 0.7, it is considered a severe anomaly; and greater than 0.8, it is considered an extreme anomaly. For data from a charging station at a certain moment, a feature vector containing current load, historical 24-hour load sequence, and time features is constructed, and then input into an isolated forest to calculate the anomaly score.

[0011] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, step S7, constructing an autoencoder network for auxiliary anomaly detection, includes: An autoencoder consists of two parts: an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder reconstructs the original dimension from the low-dimensional representation. The training objective of an autoencoder is to minimize the error between the input and the reconstructed output, i.e., the mean squared error. The training process of an autoencoder uses normal charging data, enabling the network to learn to reconstruct the feature patterns of normal samples. For a new sample, calculate its reconstruction error, which is the square of the Euclidean distance between the input vector and the reconstructed vector. The anomaly threshold is set as the 95th percentile of the reconstruction error in the training set, meaning that 95% of the samples in the training set have reconstruction errors less than this threshold. When the reconstruction error of a new sample exceeds the threshold, it is judged as an anomaly.

[0012] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, in step S8, the integrated anomaly determination is performed using a weighted fusion strategy based on the detection results of the integrated isolated forest and autoencoder, including: The final outlier score is equal to the isolated forest outlier score multiplied by 0.6 plus the autoencoder normalized reconstruction error multiplied by 0.4; where the autoencoder normalized reconstruction error is equal to the actual reconstruction error divided by the outlier threshold. The sample is considered abnormal when the final anomaly score is greater than 0.65.

[0013] Furthermore, according to the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application, the method further includes: Step S9: Conduct experimental verification and comparative analysis.

[0014] The beneficial effects of this invention are as follows: The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this application integrates the time characteristics, load characteristics, user behavior characteristics, and surrounding environment characteristics of charging stations to establish a precise charging station classification system and profile template library, enabling differentiated management of different types of charging stations. Simultaneously, it requires the construction of an efficient anomaly detection model that can not only identify various anomaly modes such as equipment failure, load surges, and data anomalies in real time, but also perform root cause analysis, providing interpretable support for operation and maintenance decisions. Through the organic combination of charging station profiling and anomaly detection technologies, it can support various operational decisions such as capacity configuration optimization, precise marketing strategies, dynamic pricing mechanisms, and fault early warning, significantly improving the capacity utilization rate of charging stations, reducing operating costs, and improving user experience, thus promoting the transformation of charging station operation from extensive management to refined, intelligent, and data-driven management. This is not only a practical need for charging station operators to enhance their competitiveness and profitability, but also an important technological guarantee for the healthy and sustainable development of the electric vehicle industry. Attached Figure Description

[0015] The technical solution and other beneficial effects of this application will become apparent from the following detailed description of specific embodiments in conjunction with the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this embodiment.

[0017] Figure 2 The table shows the experimental verification results of the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this embodiment. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] In the description of this application, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] The following disclosure provides many different embodiments or examples for implementing different structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. In addition, various specific examples of processes and materials are provided in this application, but those skilled in the art will recognize the application of other processes and / or the use of other materials.

[0021] Example 1: The embodiments of this application will now be further described in conjunction with the accompanying drawings and specific implementation details.

[0022] Figure 1 This is a flowchart illustrating the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in this embodiment.

[0023] like Figure 1 As shown in the figure, this embodiment provides a method for constructing a charging station profile and detecting anomalies based on multi-dimensional feature fusion. The method includes: Step S1: Receive input data from the charging station and perform data preprocessing and multidimensional feature vector extraction; Step S2: Standardize the extracted multidimensional feature vectors to eliminate the differences in dimensions between different features; Step S3: Use a hierarchical clustering algorithm to cluster and classify the charging stations; Step S4: Use the profile coefficient to evaluate the clustering quality of charging stations; Step S5: Construct a standard profile template for each cluster; Step S6: Construct an isolated forest model for anomaly detection; Step S7: Construct an autoencoder network for auxiliary anomaly detection; Step S8: Integrate the detection results of isolated forest and autoencoder, and use a weighted fusion strategy to perform integrated anomaly determination.

[0024] Specifically, in step S1, receiving input data from the charging station and performing data preprocessing and multidimensional feature extraction, the following steps are included: Step S11: Receive input data; wherein the input data includes the latitude and longitude coordinates of N charging stations, hourly charging records within the sensed time period, and a preprocessed distance prior matrix; specifically, in this embodiment, the hourly charging records within the time period from June to August 2023 are used as an example for explanation.

[0025] Step S12: Clean the received input data, calculate the average and standard deviation of the historical charging load for each charging station, and mark the charging load at a certain moment as an outlier and remove it when it deviates from the average by more than three times the standard deviation; for missing charging data, a spatiotemporal interpolation method is used to fill in the missing data. Specifically, data cleaning is performed first, and outliers are identified using the three-standard-deviation criterion. For the charging load of charging station i at time t... Calculate its mean and standard deviation An exception is marked as such when the following conditions are met:

[0026] In this embodiment, a spatiotemporal weighted interpolation method is used for missing data. For the missing value of station i at time t, the interpolation calculation formula is:

[0027] in The historical average load of site i during the same period. For the spatial neighborhood of site i, For distance-based spatial weights, This is the time weighting coefficient (taken as 0.7).

[0028] Step S13: Extract time features from historical charging records. For each charging station, extract its 24-hour load curve and record the average charging load for each hour. Identify the time when the peak occurs, i.e., the hour with the largest load in 24 hours. Specifically, for charging station i, extract its 24-hour load curve vector. ,in The average load over hours (h). Calculate the peak time. :

[0029] Step S14: Calculate the load volatility, which is the standard deviation of the load series divided by the mean. Calculate load volatility (coefficient of variation) :

[0030] in and These are the standard deviation and mean of the load series, respectively.

[0031] Step S15: Calculate the capacity utilization rate, which is the cumulative sum of charging load at all times divided by the number of time steps and then divided by the installed capacity. Calculate capacity utilization :

[0032] in This represents the total number of time steps (2208 hours, which is the total time during the period from June to August 2023). Let i be the installed capacity of site i.

[0033] Step S16: Extract user behavior features, including average single charging time and fast charging ratio. The fast charging ratio is equal to the number of fast charging times divided by the total number of charging times. Calculate the average charging time per charge and fast charging ratio :

[0034] in and These represent the number of fast and slow charging cycles, respectively.

[0035] Step S17: Extract POI features from the electronic map by calling the API interface of the electronic map to obtain 20 types of POI data within a 500-meter radius of each charging station; for each charging station, count the number of each type of POI within a 500-meter radius and construct a 20-dimensional POI feature vector; in this embodiment, Gaode Map is used as an example for explanation.

[0036] Specifically, the system calls the Gaode Map API to obtain 20 types of POI data within a 500-meter radius of each charging station, including: residential areas, office buildings, shopping malls, catering services, healthcare, schools, transportation facilities, government agencies, financial services, leisure and entertainment, hotels, life services, car services, sports venues, tourist attractions, cultural venues, industrial parks, parking lots, public facilities, and other categories.

[0037] For charging station i, construct the POI feature vector. ,in This represents the number of POIs of type k within a 500-meter radius.

[0038] Step S18: Calculate the POI diversity index using the Shannon entropy method. First, calculate the proportion of each type of POI to the total POIs, and then multiply the proportion by the negative of its logarithm and sum them up.

[0039] Specifically, calculate the POI diversity index. (Shannon entropy):

[0040] This index reflects the diversity of facilities around charging stations; the higher the index, the more diverse the types of facilities.

[0041] In step S2, standardizing the extracted multidimensional feature vectors to eliminate differences in units between different features includes: The Z-score standardization method is used to calculate the mean and standard deviation of each feature vector across all charging stations. Then, the mean is subtracted from each feature value and the standard deviation is divided to make the standardized feature mean zero and the standard deviation one. Specifically, for feature vectors Z-score standardization is adopted:

[0042] in and denoted as the mean and standard deviation of the d-th feature across all stations.

[0043] By integrating time features, load features, user behavior features, and POI features, a comprehensive feature matrix is ​​constructed. , where N is the number of charging stations and D is the total dimension of the features (approximately 50-60 dimensions). This matrix serves as the input for subsequent cluster analysis.

[0044] In step S3, which involves using a hierarchical clustering algorithm to cluster and classify charging stations, the following steps are included: A similarity metric for charging stations is defined, using Jensen-Shannon divergence to calculate the similarity of 24-hour load curves. Specifically, for the load curves of two charging stations, they are first normalized to probability distributions, i.e., the load value for each hour is divided by the total load for 24 hours. Then, a mixture distribution of the two probability distributions is calculated, representing the average load for the corresponding hour. The Jensen-Shannon divergence is obtained by averaging the KL divergences between the two original distributions and the mixture distribution. The KL divergence is calculated by taking the logarithm of the probability value of the original distribution divided by the probability value of the mixture distribution for each hour, multiplying this by the probability value of the original distribution, and summing the results over 24 hours. The Jensen-Shannon divergence ranges from zero to one; a smaller value indicates greater similarity between the two load curves.

[0045] Specifically, for the 24-hour load curves of charging stations i and j, they are first normalized into probability distributions. and :

[0046] Calculate the mixture distribution :

[0047] The Jensen-Shannon divergence is defined as:

[0048] Where KL is the Kullback-Leibler divergence, calculated using the following formula:

[0049] The range of values ​​for JS divergence is (using the logarithm to the base 2), and the smaller the value, the more similar the two distributions are.

[0050] The comprehensive distance between charging stations is defined by combining comprehensive feature similarity and spatial prior constraints. The comprehensive distance consists of three parts: JS divergence with a weight of 60%, Euclidean distance of standardized feature vectors with a weight of 40%, and spatial constraint terms based on distance prior matrix with a weight of 20%. That is, by combining comprehensive feature similarity and spatial prior constraints, the comprehensive distance between charging stations i and j is defined as follows:

[0051] in Let i be the standardized feature vector of site i. These are the elements of the normalized distance prior matrix (the closer the distance, the larger the value). and These are the weighting coefficients, taken as 0.6 and 0.2 respectively.

[0052] Hierarchical clustering employs the Ward connectivity criterion, selecting the two clusters with the smallest intra-cluster variance increment after merging each time. The variance increment calculation considers the sample size of the two clusters and the distance between their cluster centers. The more samples a cluster has and the closer its cluster centers are, the smaller the variance increment after merging. Hierarchical clustering employs the Ward connection criterion, merging the two clusters that minimize the increase in intra-cluster variance at each step. For clusters... and The variance increment after merging is:

[0053] in For clusters The number of samples, It is the cluster center.

[0054] In step S4, evaluating the clustering quality of charging stations using the profile coefficient, the following is included: For each charging station, first calculate its average distance to other stations within the same cluster, called the intra-cluster distance; then calculate its average distance to all stations in its nearest neighbor cluster, called the inter-cluster distance; the profile coefficient is equal to the inter-cluster distance minus the intra-cluster distance, and then divided by the larger of the two values; the profile coefficient ranges from negative one to positive one, the closer the value is to positive one, the more consistent the station is with the height of its own cluster and the better separated from other clusters, the closer the value is to zero, the more the station is on the cluster boundary, and the negative value indicates that there may be an allocation error; the overall profile coefficient is the average of the profile coefficients of all charging stations; Specifically, for charging station i, which belongs to cluster Calculate the average intra-cluster distance :

[0055] Calculate the average distance from station i to its nearest neighbor cluster. :

[0056] Profile coefficient Defined as:

[0057] The silhouette coefficient ranges from [-1, 1], and the closer the value is to 1, the better the clustering effect.

[0058] The cluster number K is iterated from 3 to 10, and the overall silhouette coefficient corresponding to each K value is calculated. The K value that maximizes the silhouette coefficient is selected as the optimal number of clusters. At the same time, the Davies-Bouldin index is used as an auxiliary evaluation. The smaller the index, the better the clustering effect.

[0059] Specifically, the overall silhouette coefficient is the average of all samples:

[0060] Iterate through the number of clusters K from 3 to 10, and select the K value that maximizes the silhouette coefficient as the optimal number of clusters.

[0061] In step S5, a standard profile template is constructed for each cluster k. The template contains the following: Extract the load curve of the center of the pattern, which is the average of the 24-hour load curves of all charging stations in the cluster, to form a representative load curve for this type; That is, the load curve at the center of the model The average load curve for all sites within the cluster is calculated as follows:

[0062] Calculate the load fluctuation range, and calculate the standard deviation of the load of all stations in the cluster for each hour. The larger the standard deviation, the greater the load difference between different stations during the time period; the smaller the standard deviation, the more consistent the load of this type of station during the time period. That is, the load fluctuation range, and the standard deviation of each time period is calculated. :

[0063] Extract the typical POI feature distribution, calculate the average value of the feature vectors of all charging station POIs in the cluster, and form a 20-dimensional typical POI distribution vector.

[0064] Typical POI characteristic distribution Calculate the mean vector of POI features within the cluster:

[0065] This feature vector characterizes the typical surrounding environment of this type of charging station. For example, the number of residential points of interest (POIs) around a residential area type station is significantly higher than that of other types.

[0066] In step S6, constructing an isolated forest model for anomaly detection, the following is included: An isolated forest consists of no fewer than 100 isolated trees, each of which recursively divides the data by randomly selecting features and split points; Calculate the average path length of a data point across all isolated trees; Specifically, in this embodiment, an isolation forest model is constructed for anomaly detection. The isolation forest consists of T isolated trees, and each tree recursively partitions the data by randomly selecting features and split points.

[0067] For a data point x, the path length in the isolated tree t is denoted as . This refers to the number of edges traversed from the root node to a leaf node. Calculate the average path length of x across all isolated trees. :

[0068] Define a normalization factor that represents the theoretical average path length of a binary search tree containing n samples, calculated using the harmonic number; That is, define the normalization factor. , where represents the average path length of a binary search tree containing n samples:

[0069] in For harmonic numbers, (Euler's constant).

[0070] The anomaly score is equal to the negative exponent of two, where the exponent is the average path length divided by the normalization factor. The anomaly score ranges from zero to one. When the anomaly score is greater than 0.6, it is considered a general anomaly; greater than 0.7, it is considered a severe anomaly; and greater than 0.8, it is considered an extreme anomaly. That is, calculating abnormal scores :

[0071] The range of abnormal scores is [value missing]. When [value missing] It was judged as abnormal at that time. This is a serious abnormality. This is an extreme anomaly.

[0072] For data from a charging station at a certain moment, a feature vector containing current load, historical 24-hour load sequence, and time features is constructed, and then input into an isolated forest to calculate the anomaly score.

[0073] That is, for charging station i at time t, construct a feature vector. Including current load, historical 24-hour load sequence, and time characteristics, the anomaly score is calculated by inputting into the isolated forest. .

[0074] In step S7, constructing an autoencoder network for auxiliary anomaly detection, the following is included: An autoencoder consists of two parts: an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder reconstructs the original dimension from the low-dimensional representation. That is, a self-encoder contains an encoder. and decoder The input x is compressed into a low-dimensional representation z and then reconstructed:

[0075] The training objective of an autoencoder is to minimize the error between the input and the reconstructed output, i.e., the mean squared error. The training process of an autoencoder uses normal charging data, enabling the network to learn to reconstruct the feature patterns of normal samples. The training objective is to minimize the reconstruction error.

[0076] For a new sample, calculate its reconstruction error, which is the square of the Euclidean distance between the input vector and the reconstructed vector. The anomaly threshold is set as the 95th percentile of the reconstruction error in the training set, meaning that 95% of the samples in the training set have reconstruction errors less than this threshold. When the reconstruction error of a new sample exceeds the threshold, it is judged as an anomaly.

[0077] In this embodiment, for a new sample x, its reconstruction error is calculated. :

[0078] Set an abnormal threshold The 95th percentile of the reconstruction error for the training set. It is judged as abnormal at that time.

[0079] In step S8, the ensemble anomaly determination is performed using a weighted fusion strategy based on the detection results of the integrated isolated forest and the autoencoder, including: The final outlier score is equal to the isolated forest outlier score multiplied by 0.6 plus the autoencoder normalized reconstruction error multiplied by 0.4; where the autoencoder normalized reconstruction error is equal to the actual reconstruction error divided by the outlier threshold. A sample is considered anomalous when the final anomaly score is greater than 0.65. This ensemble strategy combines the advantages of two methods: Isolation Forest excels at detecting significantly deviating point anomalies, while Autoencoder excels at detecting anomalies in temporal patterns. That is, integrating the detection results of isolated forests and autoencoders, and employing a weighted fusion strategy:

[0080] in , .when It is judged as abnormal at that time.

[0081] For detected anomalous samples, root cause analysis is performed to identify the primary causes of the anomalies. The DIFFI feature importance method is employed, which analyzes the segmentation paths of anomalous samples in isolated trees and calculates the contribution of each feature to the anomaly determination. Specifically, the calculation involves counting the number of times each feature is used for segmentation across all isolated trees, dividing by the tree depth, and then averaging the results. The three features with the highest importance are identified as the primary causes of the anomalies.

[0082] Specifically, in this embodiment, for detected anomalous samples, feature importance is calculated. The importance of feature d is calculated using the DIFFI (Depth-based Isolation Forest Feature Importance) method. :

[0083] in Let d be the number of times feature d is used for segmentation in tree t. Let t be the total depth of the tree. The top 3 features with the highest importance are the main causes of the anomalies.

[0084] In this embodiment, the charging station profile construction and anomaly detection method based on multi-dimensional feature fusion further includes: Step S9: Conduct experimental verification and comparative analysis.

[0085] Specifically, the method provided in this embodiment is verified by experiments on a real dataset of charging stations, which covers 92 days of hourly charging records, for experimental verification and comparative analysis.

[0086] The experimental results in this example are as follows: Figure 2 As shown.

[0087] The validation was performed on a dataset containing real charging stations, which covered 92 days of hourly charging records and was divided into training, validation and test sets in chronological order.

[0088] In the clustering effect evaluation experiment, the clustering quality was assessed by traversing the number of clusters from 3 to 10 and using the silhouette coefficient and Davies-Bouldin index. The experimental results showed that the optimal number of clusters was 6, at which point the silhouette coefficient reached 0.68 and the Davies-Bouldin index was 0.82, indicating good clustering quality. Six typical charging station types were identified: residential area night charging, commercial area all-day charging, transportation hub charging, industrial park weekday charging, mixed type, and low-load cultivation type. The characteristics of each type highly matched the actual application scenarios. The silhouette coefficient and Davies-Bouldin index are used to evaluate clustering quality. The formula for calculating the silhouette coefficient is:

[0089] in The average distance within the cluster. The silhouette coefficient is the average distance to the nearest neighbor cluster. Its value ranges from -1 to 1, with a larger value indicating better clustering. The Davies-Bouldin index is calculated as follows:

[0090] in The average distance within the cluster. The DBI value represents the distance between cluster centers; a smaller DBI value indicates better clustering. Traversing the cluster number K from 3 to 10, Table 1 shows that when the optimal cluster number K=6, the silhouette coefficient reaches 0.68 and the DBI index is 0.82, indicating good clustering quality. The six identified types are: residential area night-time charging, commercial area all-day charging, transportation hub charging, industrial park weekday charging, mixed charging, and low-load cultivation charging.

[0091] In the anomaly detection performance evaluation experiment, the anomaly detection performance was evaluated on a labeled test set containing 15,892 normal samples and 908 anomaly samples. Statistical thresholding, LOF, One-Class SVM, and LSTM autoencoder were selected as comparison methods, and performance was evaluated using four metrics: precision, recall, F1 score, and ROC-AUC. Experimental results show that the method provided in this embodiment achieves a precision of 95.1%, a recall of 92.5%, an F1 score of 93.8%, and an ROC-AUC of 0.96, outperforming the comparative methods in all evaluation metrics. Compared to the statistical thresholding method, precision is improved by 21.1%, and recall by 41.9%; compared to the second-best performing LSTM autoencoder, precision is improved by 5.4%, and recall by 8.1%, significantly outperforming existing methods. The experiments verify the effectiveness of core technologies such as the integration of isolated forest and autoencoder, POI feature fusion, and spatial prior constraints.

[0092] Specifically, four metrics are used for evaluation: precision, recall, F1 score, and ROC-AUC. The calculation formula is as follows:

[0093]

[0094] TP represents true positives, FP represents false positives, and FN represents false negatives. Statistical thresholding, LOF, One-Class SVM, and LSTM autoencoders were selected as comparative methods. As shown in Table 2, the method of this invention achieves a precision of 95.1%, a recall of 92.5%, an F1 score of 93.8%, and an ROC-AUC of 0.96, outperforming the comparative methods in all evaluation metrics. Compared to the statistical thresholding method, precision is improved by 21.1%, and recall by 41.9%; compared to the second-best performing LSTM autoencoder, precision is improved by 5.4%, and recall by 8.1%, significantly outperforming existing methods.

[0095] In summary, the charging station profiling and anomaly detection method based on multi-dimensional feature fusion provided in this embodiment employs multi-dimensional feature fusion technology, integrating three types of input data: charging station latitude and longitude, historical charging records, and distance prior matrix. It extracts environmental information of 20 types of points of interest within a 500-meter radius of the charging station using Gaode POI features, and constructs a comprehensive feature matrix by combining time features, load features, and user behavior features. Based on the Jensen-Shannon divergence metric for load curve similarity, it introduces spatial prior constraints for hierarchical clustering, identifying different types of charging station profiles and constructing a reusable pattern feature library. An anomaly detection model integrating isolated forest and autoencoder is constructed, identifying various anomaly types such as equipment failure, load mutation, and data quality anomalies through a dual mechanism of path length and reconstruction error. Root cause analysis is performed using the DIFFI feature importance method, ultimately outputting a charging station profile report and anomaly diagnosis results, supporting refined operational decisions such as capacity configuration optimization, precise marketing strategies, dynamic pricing mechanisms, and fault early warning. It solves the problems of inaccurate site feature identification, delayed response to equipment failures, lack of data support for operational decisions, low capacity utilization, and insufficient refined operation capabilities in the operation of existing charging stations. It enables accurate profiling and classification of charging stations, real-time detection of abnormal behavior, optimization of operational strategies, and improvement of user experience, providing technical support for refined operation, intelligent management, and grid load optimization of charging stations.

[0096] This application presents a charging station profiling and anomaly detection method based on multi-dimensional feature fusion. By integrating the time characteristics, load characteristics, user behavior characteristics, and surrounding environmental characteristics of charging stations, it establishes a precise charging station classification system and profiling template library, enabling differentiated management of different types of charging stations. Simultaneously, it requires the construction of an efficient anomaly detection model that can not only identify various anomaly patterns such as equipment failures, load surges, and data anomalies in real time, but also perform root cause analysis, providing interpretable support for operation and maintenance decisions. Through the organic combination of charging station profiling and anomaly detection technologies, it can support various operational decisions such as capacity configuration optimization, precise marketing strategies, dynamic pricing mechanisms, and fault early warning, significantly improving the capacity utilization rate of charging stations, reducing operating costs, and improving user experience. This drives the transformation of charging station operation from extensive management to refined, intelligent, and data-driven management. This is not only a practical need for charging station operators to enhance their competitiveness and profitability, but also an important technological guarantee for the healthy and sustainable development of the electric vehicle industry.

[0097] Example 2: This embodiment also provides a terminal device for constructing charging station profiles and detecting anomalies based on multi-dimensional feature fusion. The terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the embodiments of the method described in Embodiment 1 of the present invention.

[0098] Furthermore, as an executable solution, the charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described structure of the charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion is merely an example and does not constitute a limitation on the charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion. It may include more or fewer components than described above, or combine certain components, or different components. For example, the charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion may also include input / output devices, network access devices, buses, etc., and this embodiment of the invention does not limit this.

[0099] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the charging station profiling and anomaly detection terminal device based on multi-dimensional feature fusion, connecting all parts of the device via various interfaces and lines.

[0100] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application required for a function; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0101] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the embodiments of the present invention.

[0102] If the module / unit integrating the charging station profile construction and anomaly detection terminal device based on multi-dimensional feature fusion is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), and a software distribution medium, etc.

[0103] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the present invention. Finally, it should be noted that in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0104] The foregoing has provided a detailed description of a charging station profile construction and anomaly detection method based on multi-dimensional feature fusion provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the technical solutions and core ideas of this application. Those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for constructing charging station profiles and detecting anomalies based on multi-dimensional feature fusion, characterized in that, The method includes: Step S1: Receive input data from the charging station and perform data preprocessing and multidimensional feature vector extraction; Step S2: Standardize the extracted multidimensional feature vectors to eliminate the differences in dimensions between different features; Step S3: Use a hierarchical clustering algorithm to cluster and classify the charging stations; Step S4: Use the profile coefficient to evaluate the clustering quality of charging stations; Step S5: Construct a standard profile template for each cluster; Step S6: Construct an isolated forest model for anomaly detection; Step S7: Construct an autoencoder network for auxiliary anomaly detection; Step S8: Integrate the detection results of isolated forest and autoencoder, and use a weighted fusion strategy to perform integrated anomaly determination.

2. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 1, characterized in that, In step S1, receiving input data from the charging station and performing data preprocessing and multidimensional feature extraction, the following are included: Step S11: Receive input data; wherein the input data includes the latitude and longitude coordinates of N charging stations, hourly charging records within the sensed time period, and a preprocessed distance prior matrix; Step S12: Clean the received input data, calculate the average and standard deviation of the historical charging load for each charging station, and mark the charging load at a certain moment as an outlier and remove it when it deviates from the average by more than three times the standard deviation; for missing charging data, a spatiotemporal interpolation method is used to fill in the missing data. Step S13: Extract time features from historical charging records. For each charging station, extract its 24-hour load curve and record the average charging load for each hour. Identify the time when the peak occurs, i.e., the hour with the largest load in 24 hours. Step S14: Calculate the load volatility, which is the standard deviation of the load series divided by the mean. Step S15: Calculate the capacity utilization rate, which is the cumulative sum of charging load at all times divided by the number of time steps and then divided by the installed capacity. Step S16: Extract user behavior features, including average single charging time and fast charging ratio. The fast charging ratio is equal to the number of fast charging times divided by the total number of charging times. Step S17: Extract POI features from the electronic map, call the API interface of the electronic map to obtain 20 types of POI data within a 500-meter radius of each charging station; for each charging station, count the number of each type of POI within a 500-meter radius and construct a 20-dimensional POI feature vector. Step S18: Calculate the POI diversity index using the Shannon entropy method. First, calculate the proportion of each type of POI to the total POIs, and then multiply the proportion by the negative of its logarithm and sum them up.

3. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 2, characterized in that, In step S2, standardizing the extracted multidimensional feature vectors to eliminate differences in units between different features includes: The Z-score standardization method is used to calculate the mean and standard deviation of each feature vector across all charging stations. Then, the mean is subtracted from each feature value and the standard deviation is divided to make the standardized feature mean zero and the standard deviation one. A comprehensive feature matrix is ​​constructed by integrating time features, load features, user behavior features, and POI features; the number of rows in the comprehensive feature matrix is ​​N, which is the number of charging stations, and the number of columns is D, which is the total feature dimension.

4. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 2, characterized in that, In step S3, which involves using a hierarchical clustering algorithm to cluster and classify charging stations, the following steps are included: A similarity measurement method between charging stations is defined, and the similarity of the 24-hour load curves is calculated using Jensen-Shannon divergence. That is, for the load curves of two charging stations, they are first normalized to a probability distribution, that is, the load value of each hour is divided by the total load of 24 hours; then the mixture distribution of the two probability distributions is calculated, that is, the average value of the hourly load. The comprehensive distance between charging stations is defined by combining comprehensive feature similarity and spatial prior constraints. The comprehensive distance consists of three parts: JS divergence with a weight of 60%, Euclidean distance of standardized feature vectors with a weight of 40%, and spatial constraint terms based on distance prior matrix with a weight of 20%. Hierarchical clustering uses the Ward connection criterion, selecting the two clusters with the smallest increase in intra-cluster variance after merging each time.

5. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 2, characterized in that, In step S4, evaluating the clustering quality of charging stations using the profile coefficient, the following is included: For each charging station, first calculate its average distance to other stations within the same cluster, called the intra-cluster distance; then calculate its average distance to all stations in its nearest neighbor cluster, called the inter-cluster distance; the profile coefficient is equal to the inter-cluster distance minus the intra-cluster distance, and then divided by the larger of the two values; the profile coefficient ranges from negative one to positive one, the closer the value is to positive one, the more consistent the station is with the height of its own cluster and the better separated from other clusters, the closer the value is to zero, the more the station is on the cluster boundary, and the negative value indicates that there may be an allocation error; the overall profile coefficient is the average of the profile coefficients of all charging stations; The cluster number K is iterated from 3 to 10, and the overall silhouette coefficient corresponding to each K value is calculated. The K value that maximizes the silhouette coefficient is selected as the optimal number of clusters. At the same time, the Davies-Bouldin index is used as an auxiliary evaluation. The smaller the index, the better the clustering effect.

6. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 5, characterized in that, In step S5, constructing a standard profile template for each cluster, the following is included: Extract the load curve of the center of the pattern, which is the average of the 24-hour load curves of all charging stations in the cluster, to form a representative load curve for this type; Calculate the load fluctuation range, and calculate the standard deviation of the load of all stations in the cluster for each hour. The larger the standard deviation, the greater the load difference between different stations during the time period; the smaller the standard deviation, the more consistent the load of this type of station during the time period. Extract the typical POI feature distribution, calculate the average value of the feature vectors of all charging station POIs in the cluster, and form a 20-dimensional typical POI distribution vector.

7. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 2, characterized in that, In step S6, constructing an isolated forest model for anomaly detection, the following is included: An isolated forest consists of no fewer than 100 isolated trees, each of which recursively divides the data by randomly selecting features and split points; Calculate the average path length of a data point across all isolated trees; Define a normalization factor that represents the theoretical average path length of a binary search tree containing n samples, calculated using the harmonic number; The anomaly score is equal to the negative exponent of two, where the exponent is the average path length divided by the normalization factor. The anomaly score ranges from zero to one. When the anomaly score is greater than 0.6, it is considered a general anomaly; greater than 0.7, it is considered a severe anomaly; and greater than 0.8, it is considered an extreme anomaly. For data from a charging station at a certain moment, a feature vector containing current load, historical 24-hour load sequence, and time features is constructed, and then input into an isolated forest to calculate the anomaly score.

8. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 7, characterized in that, In step S7, constructing an autoencoder network for auxiliary anomaly detection, the following is included: An autoencoder consists of two parts: an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder reconstructs the original dimension from the low-dimensional representation. The training objective of an autoencoder is to minimize the error between the input and the reconstructed output, i.e., the mean squared error. The training process of an autoencoder uses normal charging data, enabling the network to learn to reconstruct the feature patterns of normal samples. For a new sample, calculate its reconstruction error, which is the square of the Euclidean distance between the input vector and the reconstructed vector. The anomaly threshold is set as the 95th percentile of the reconstruction error in the training set, meaning that 95% of the samples in the training set have reconstruction errors less than this threshold. When the reconstruction error of a new sample exceeds the threshold, it is judged as an anomaly.

9. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 8, characterized in that, In step S8, the ensemble anomaly determination is performed using a weighted fusion strategy based on the detection results of the integrated isolated forest and the autoencoder, including: The final outlier score is equal to the isolated forest outlier score multiplied by 0.6 plus the autoencoder normalized reconstruction error multiplied by 0.4; where the autoencoder normalized reconstruction error is equal to the actual reconstruction error divided by the outlier threshold. The sample is considered abnormal when the final anomaly score is greater than 0.

65.

10. The charging station profile construction and anomaly detection method based on multi-dimensional feature fusion according to claim 1, characterized in that, The method further includes: Step S9: Conduct experimental verification and comparative analysis.

Images