A concentrator operation data management method

Abstract

The present invention relates to the field of data processing, and specifically to a method for managing concentrator operation data, which ensures efficient and accurate management of data by real-time collection, preprocessing and anomaly detection of operation data from smart meters and sensors. By adopting an improved LOF (local outlier factor) anomaly detection algorithm, by dynamically adjusting the neighborhood size (k value), combining local density differences and multi-dimensional data fluctuation similarity, potential abnormal data points can be accurately identified. The abnormal data points finally identified will be marked and stored, and the system will trigger an alarm mechanism to promptly notify the operation and maintenance personnel to handle them. The present invention realizes efficient and accurate anomaly detection by improving the LOF algorithm, dynamically adjusting the neighborhood size k value, combining local density differences and multi-dimensional fluctuation similarity.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method for managing concentrator operation data. Background Art

[0002] The concentrator is designed for smart grid and smart metering systems. As a communication gateway, the device connects smart meters, sensors and other devices with the Advanced Metering Infrastructure (AMI) system, and undertakes important functions such as data collection, task execution, data transmission and storage. It can handle complex communication protocols (such as DLMS / COSEM, RTU Modbus, DNP3, etc.) and provide data storage, processing and transmission capabilities. It is used in the fields of remote monitoring, data collection, fault diagnosis, etc. of smart grids.

[0003] However, in its long-term operation, the communication stability and data reliability of the concentrator still face challenges. Therefore, the management, analysis and anomaly detection of the concentrator operation data have become key steps to ensure the reliability of the equipment. In smart grid and smart metering systems, the concentrator is responsible for collecting data from various devices, including multiple parameters such as power, voltage, temperature, humidity, etc. Since these devices are running all year round and are affected by environmental conditions and workloads, their performance and data fluctuations have great uncertainty. Therefore, the data collected by the concentrator may be abnormal, which may be caused by factors such as equipment failure, sensor error, communication interference or environmental fluctuations. If the abnormal data is mistaken for normal data and continues to be used, it may lead to wrong control decisions of the smart grid, thereby affecting power distribution and load management. Therefore, anomaly detection is particularly important in the management of concentrator operation data, which can effectively ensure the stable operation of the system and reduce potential failures and losses.

[0004] As an unsupervised anomaly detection algorithm, the LOF (Local Outlier Factor) algorithm is often used in the management of concentrator operation data. The basic idea of ​​the LOF algorithm is to detect anomalies by calculating the density difference of data points relative to their neighborhood. However, in the power system, the parameter fluctuation ranges of different devices under different operating conditions vary greatly. At this time, the fixed neighborhood size cannot flexibly adapt to the changes in data. When the device data fluctuates greatly, a smaller neighborhood size may cause local data to be misjudged as abnormal; when the device data is relatively stable, a larger neighborhood size may ignore anomalies in a small range, resulting in underreporting. Summary of the invention

[0005] In view of the problem that the above-mentioned LOF algorithm cannot flexibly adapt to data changes, the present invention proposes a concentrator operation data management method, including: collecting multiple parameters of terminal devices connected to the concentrator; the multiple parameters at any collection time constitute a parameter vector, and the multiple parameter vectors under the set collection duration constitute a data set; the data set is subjected to anomaly detection using the LOF algorithm to obtain the LOF value of each parameter vector, and in response to the existence of the LOF value being greater than the set LOF threshold, an alarm mechanism is triggered and the operation and maintenance personnel are notified; the LOF algorithm includes processing the initial neighborhood size to obtain an adaptive neighborhood size, the adaptive neighborhood size is negatively correlated with the product of the initial neighborhood size and the local fluctuation similarity, and the calculation method of the local fluctuation similarity is specifically: taking any parameter vector as the target data point of the LOF algorithm, and calculating the local density of the target data point; the local density is correlated with the first The difference between the local density of the remaining data points in the neighborhood constitutes the local density difference; the data points corresponding to the local density differences greater than the set threshold constitute the abnormal point set; the parameter time series of all abnormal points in the abnormal point set are obtained, and the relationship between any abnormal point and its The similarity between the parameter time series of the rest of the data points in the distance neighborhood yields the local fluctuation similarity.

[0006] The concentrator operation data management method of the present invention provides a dynamic and adaptive anomaly detection method by improving the LOF algorithm and combining the analysis of local density differences and multi-dimensional fluctuation similarities. The traditional LOF algorithm has the limitation of a fixed neighborhood size k, which can easily lead to false alarms or missed alarms, especially when the equipment status and environmental conditions change greatly. The present invention adaptively adjusts the k value so that the neighborhood size is dynamically adjusted according to the local density difference and fluctuation similarity of the data points. The introduction of the adaptive neighborhood size enables the LOF algorithm to select a smaller neighborhood when the fluctuation is large, thereby enhancing the sensitivity to abnormal points; when the fluctuation is small, a larger neighborhood is selected to avoid oversensitivity, thereby reducing false alarms and missed alarms. Compared with the prior art, the present invention significantly improves the anomaly detection accuracy of the concentrator in a dynamic environment, and provides a more stable and reliable data management solution for smart grids and smart metering systems.

[0007] Furthermore, the method for calculating the adaptive neighborhood size is specifically as follows:

[0008] ;

[0009] in Represents anomaly point set Middle Data points Adaptive neighborhood size; represents the rounding function; represents the initial neighborhood size; represents the exponential function; Represents data points The local fluctuation similarity.

[0010] The adaptive neighborhood size calculation method further optimizes the performance of the LOF algorithm through dynamic adjustment based on local fluctuation similarity. This method uses an exponential function to dynamically adjust the neighborhood size according to the fluctuation similarity, thereby ensuring the consistency and accuracy of the algorithm under different abnormal conditions. Compared with the fixed neighborhood size of the prior art, the adaptive neighborhood size of the present invention is more flexible, can effectively cope with changes in equipment operation and environmental conditions, and improve the accuracy of anomaly detection.

[0011] Furthermore, the calculation method of the local fluctuation similarity is specifically as follows:

[0012] ;

[0013] in Represents anomaly point set Middle Data points The local fluctuation similarity of represents a counting function; Represents data points No. Distance neighborhood; Represents a parameter set; Represents the DTW distance calculation function; Represents data points Middle Parameter timing; Represents data points No. Distance Neighborhood Medium data point No. A sequence of parameters.

[0014] The present invention further enhances the recognition accuracy of abnormal data points by calculating the local fluctuation similarity. By comparing the multi-dimensional time series similarity between the abnormal point and other points in its neighborhood, it is possible to more accurately determine whether the fluctuation is caused by actual abnormal data. Compared with the prior art, this method not only considers the distance between the data point and the neighborhood, but also integrates the time series information of multi-dimensional parameters, effectively improving the system's fault tolerance to occasional failures.

[0015] Furthermore, the calculation method of the local density difference is specifically as follows:

[0016] ;

[0017] in Indicates Data points The local density difference of Indicates Data points No. Distance neighborhood; represents a counting function; Represents data points The local density of Represents data points The local density of .

[0018] The local density difference calculation method of the present invention makes the LOF algorithm more accurate in local density analysis. By calculating the local density difference, normal data and abnormal data can be better distinguished. Unlike the traditional LOF algorithm that only relies on local density calculation, the present invention effectively improves the accuracy of abnormality detection by accurately calculating the local density difference.

[0019] Furthermore, the local density is calculated as follows:

[0020] ;

[0021] in Represents data points The local density of Represents data points No. distance.

[0022] Furthermore, obtaining the parameter timing of all abnormal points in the abnormal point set also includes: for any parameter of any abnormal point, obtaining the parameter data of a set number of collection moments before and after it, all the parameter data constitute the parameter timing of the corresponding abnormal point; the parameter timing does not include the parameter data at the corresponding collection moment of the corresponding abnormal point.

[0023] The present invention can more accurately calculate the real fluctuation of the abnormal point by acquiring the time series data before and after the abnormal point and excluding the data at the current abnormal moment. This method effectively avoids misjudgment caused by occasional failures or data collection errors, and improves the stability and reliability of abnormality detection.

[0024] Furthermore, the set threshold is obtained by using a maximum inter-class variance method based on all local density differences.

[0025] Furthermore, it also includes preprocessing operations on the multiple parameters, specifically: using a bandpass filter to perform signal smoothing processing on the multiple parameters; using maximum and minimum value normalization to process the multiple parameters after smoothing processing to obtain standard data.

[0026] The technical effects of the present invention are:

[0027] The present invention provides a concentrator operation data management method based on an improved LOF algorithm, which effectively solves the false alarm and missed alarm problems that are prone to occur in traditional LOF algorithms when equipment and environmental conditions change by adaptively adjusting the neighborhood size and combining multi-dimensional fluctuation similarity analysis. By dynamically adjusting the neighborhood size, the algorithm can flexibly respond to data volatility, ensuring that abnormal data can still be efficiently identified under the diversity of equipment operating conditions and environmental changes. Compared with the prior art, the present invention improves the accuracy of anomaly detection while reducing misjudgments caused by equipment failures or communication interference, thereby improving the robustness and reliability of the system. In addition, through the abnormal data timing analysis and fault diagnosis functions, the present invention can provide timely and accurate decision support for operation and maintenance personnel, facilitate equipment maintenance and system optimization, and ensure the stable operation of smart grids and smart metering systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0029] Figure 1 It is a flow chart schematically showing a concentrator operation data management method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0031] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0032] Concentrator operation data management method embodiment:

[0033] like Figure 1 As shown, the concentrator operation data management method of the present invention includes:

[0034] S1. Collect concentrator operating parameters and perform data preprocessing.

[0035] The smart data concentrator is a high-performance gateway device dedicated to smart grids and smart metering systems. The device is responsible for connecting smart meters, sensors and other monitoring terminals to collect multiple operating parameters such as power, voltage, temperature, humidity, and communication quality. The data generated during the operation of the concentrator is not only an intuitive reflection of the health status of the device itself, but also a key indicator of the safety and stability of the entire smart grid operation. Therefore, the management of the concentrator operation data is also very important. It can not only ensure the normal operation of the real-time monitoring and remote management of the smart metering system; it can also provide detailed records of the health status of the equipment and the environmental conditions, providing data basis for fault diagnosis and preventive maintenance; and it can realize early identification and alarm of abnormal situations, ensuring that abnormal events can be quickly located and remedial measures can be taken when they occur.

[0036] The concentrator has multiple built-in communication interfaces, including PLC, RF, GPRS, 3G, DSL and fiber optic interfaces, and each interface establishes a direct data connection with the corresponding smart meter and sensor. The data acquisition module manages all access devices in a unified manner, and uses a built-in protocol parser to collect data of different communication protocols in a standardized manner. The specific process of the data acquisition process is as follows: the concentrator loads the device list at startup, initializes the communication link with each acquisition terminal, and verifies the link status; the data acquisition module starts the data acquisition task, and pulls all device data into the concentrator memory in a unified manner according to the predetermined period. In this embodiment, the predetermined period can be set to 30Hz; a unified timestamp is assigned to the collected raw data, and the timestamp is generated by the built-in high-precision real-time clock of the concentrator to ensure that all data recording times are consistent; the data acquisition results are stored in the form of data frames. In this embodiment, the data frame format is set to a standard binary structure, including device identifier, collected power, temperature, humidity, voltage, timestamp and communication status code.

[0037] After collecting the operating parameters, a bandpass filter can be used to smooth the signal for possible interference and noise. After all data records are collected, the concentrator uses the built-in clock to align the time to ensure that all data records have a unified time reference, and all data are standardized using the maximum and minimum standard method. The final standardized data format is defined as:

[0038] ;

[0039] in Indicates The data frame at the acquisition time, It indicates the unique identifier of the data collection terminal. It should be noted that the concentrator corresponds to multiple data collection terminals, such as smart meters and temperature measuring devices. Indicates the data collection timestamp; Indicates a parameter value set of the acquisition terminal. In this embodiment, the acquisition parameters can be set as: power, temperature, humidity and voltage; Indicates the communication status code. The data formatting operation converts the data returned by each communication interface into a unified data frame. , to facilitate subsequent anomaly detection and data management.

[0040] The pre-processed data is stored in the internal cache of the concentrator according to the collection cycle. The cache structure uses a circular queue to ensure that the latest data is always available for subsequent processing modules. During the storage process, the data is also written into the non-volatile memory (NAND Flash) to ensure data persistence and prevent data loss due to power outages or system abnormalities.

[0041] The overall process of data collection and preprocessing is as follows: first, the concentrator is started to initialize the device and communication interface; then the data collection task is started to pull all terminal data at a preset period. The original data is recorded as: ; Then give all data a timestamp ; Further apply digital filters to the raw data to obtain denoised data ; Continue to standardize the data and convert it into a unified format data frame ;Finally, the data frame is stored in the cache and written to the non-volatile memory.

[0042] S2. Implement anomaly detection of the data set based on the LOF algorithm; obtain local density differences based on the density of data points; obtain local fluctuation similarity based on the potential abnormal data points obtained in step S2.2 and calculate the adaptive neighborhood size.

[0043] In step S1, the concentrator performs pre-processing operations such as denoising, time synchronization, standardization, and formatting on the raw data collected from smart meters and sensors to ensure high quality and consistency of the data. However, there may be abnormalities in these data caused by equipment failure, environmental changes, or communication interference, which may seriously affect the normal operation of the system. Therefore, anomaly detection is required in the management of concentrator operation data to promptly discover and solve potential problems.

[0044] S2.1. Implement anomaly detection of data sets based on LOF algorithm.

[0045] In this embodiment, the LOF (Local Outlier Factor) algorithm can be used as the main algorithm for anomaly detection. The LOF algorithm is a density-based unsupervised anomaly detection method that is suitable for high-dimensional data and complex data sets. The core idea is to determine whether the data point is an outlier by calculating the local density of the data point and comparing it with the density difference of other data points in the neighborhood. If the local density of a data point is significantly lower than that of the points in its neighborhood, it is considered an outlier.

[0046] In this embodiment, the LOF algorithm implementation steps are specifically as follows:

[0047] First, the multiple consecutive data frames closest to the current time after preprocessing in step S1 form a data set , and there are: ,in Represents the total number of data sets, Indicates In this embodiment, since the acquisition frequency is 30 Hz, the timeliness of abnormality detection and the amount of data can be comprehensively considered. The empirical value is set to 300, and each data frame contains multiple parameters;

[0048] Then the multiple parameters contained in each data frame are taken as parameter vectors, that is, each data frame corresponds to a parameter vector, and the Data frames at the acquisition time The corresponding parameter vector is recorded as , all the parameter vectors constitute a parameter set ,have: ,in represents the total number of parameter vectors, Indicates Data frames at the acquisition time The corresponding parameter vector is used for convenience. It can also be recorded as data points.

[0049] Further calculation parameter set Middle Data points No. Distance Neighborhood and the neighborhood distance , neighborhood size It can be set according to the characteristics of the device or the data distribution. In this implementation, it can be set to an empirical value of 5;

[0050] Then the local reachable density of each data point is calculated. The local reachable density reflects the density of the data point in its neighborhood. Data points , calculate its LOF value and record it as The LOF value reflects the data point The degree of local outlier relative to its neighborhood. If The value of is significantly greater than 1, indicating that Data points The local density in its neighborhood is significantly lower than other points in the neighborhood, which may be an outlier point;

[0051] Finally, set a LOF threshold If the LOF value of a data point exceeds the threshold, the point is determined to be an outlier. The selection of the LOF threshold can be determined by historical data or by cross-validation. In this embodiment, it can be set to an empirical value of 1.2. , then the data point is an abnormal point. At this point, the LOF algorithm outputs the abnormality detection results of each data point, specifically the LOF value of each data point, and the mark of whether it is an abnormal point. If there is an abnormal point, an alarm is triggered and the alarm information is sent to the remote monitoring system through the communication interface of the concentrator so that the operation and maintenance personnel can handle it in time.

[0052] S2.2. Obtain local density differences based on the density of data points.

[0053] In the above LOF algorithm, The value is usually fixed, which means that no matter the data set How the device, environment, or device status changes in the However, in actual applications, the parameter characteristics of the equipment and environmental conditions may change, which makes the fixed The value may not effectively reflect the normal behavior of the device, resulting in too high a false positive rate or a false negative rate. If the value is too small, the algorithm may be too sensitive and easily identify small fluctuations in device data as anomalies. If the value is too large, the algorithm may be too lenient and fail to capture local abnormal behavior. The LOF algorithm of the value has certain defects in the scenario of this embodiment. Subsequently, certain adaptive adjustments are made to the value so that the LOF algorithm can more accurately detect abnormal data in the operation data management of the concentrator.

[0054] In this embodiment, the operating data characteristics of the equipment are highly dynamic and diverse, especially in terms of fluctuations in parameters such as voltage, power, temperature, and humidity. In order to cope with this diversity, by calculating the Data points The density difference in its local neighborhood is used to evaluate the characteristic fluctuation of the point. First, the local density of the data point is obtained, specifically:

[0055] ;

[0056] in Represents data points The local density of Represents data points No. Distance. Distance means the difference between data points Foreign Points and data points The distance between the data points No. When the distance is larger, the number of points around it is smaller, and the local density Smaller; on the contrary, when the data point No. The smaller the distance, the more points there are around it, and the local density Further, calculate the data points The local density difference with other points in its neighborhood is:

[0057] ;

[0058] in Indicates Data points The local density difference of Indicates Data points No. Distance neighborhood, that is nearest neighbors; Represents a counting function, used to count statistical points No. The total number of data points in the distance neighborhood; Represents data points The local density of Represents data points The local density of .

[0059] when The larger the value, the Data points The local density of the other data points in its neighborhood is quite different, and the data volatility is strong. Data points The local density difference Larger, you should choose a smaller neighborhood size , in order to improve the ability to detect local anomalies; otherwise, The smaller the time, the more Data points The local density difference between the other data points in its neighborhood is small, and the volatility of the data is weak. Data points The local density difference If it is smaller, it means that the device is running more stably, and a larger neighborhood size should be selected. To avoid oversensitivity.

[0060] The local density difference calculation method of the present invention makes the LOF algorithm more accurate in local density analysis. By calculating the local density difference, normal data and abnormal data can be better distinguished. Unlike the traditional LOF algorithm that only relies on local density calculation, the present invention effectively improves the accuracy of abnormality detection by accurately calculating the local density difference.

[0061] S2.3. Based on the potential abnormal data points obtained in step S2.2, local fluctuation similarity is obtained and the adaptive neighborhood size is calculated.

[0062] In the above step S2.2, although some data points with strong data volatility can be screened out through local density differences, which can be considered as potential abnormal data points, this does not mean that they are necessarily caused by actual parameter abnormalities. It may also be caused by fluctuation abnormalities caused by occasional failures of the acquisition equipment. At this time, although the local density difference between it and other data points in the neighborhood is large, it may also be a normal data point. If the neighborhood size is adjusted only based on the local density difference, This will not only reduce the accuracy of subsequent anomaly detection, but also increase the amount of data required for calculation. Therefore, the neighborhood size cannot be adjusted immediately after calculating the local density difference. .

[0063] In order to avoid the above situation, in this embodiment, multiple time series are constructed based on multiple dimensions of the data point, and similarity calculation is performed with multiple time series of other data points in the neighborhood where the data point is located to obtain local fluctuation similarity, and then determine the real abnormal data, so as to obtain the final adaptive neighborhood size. The specific method is:

[0064] First, all local density differences obtained in step S2.2 are recorded as data sets ,have:

[0065] ;

[0066] in Represents a dataset consisting of all local density differences; Indicates Data points The local density difference of Indicates the total number of data points. The inter-class variance method can be used to obtain the segmentation threshold, and the data points corresponding to the local density differences greater than the segmentation threshold are recorded as potential abnormal data points. At the same time, all potential abnormal data points are recorded as point sets. , so far we have obtained a set of potential abnormal data points , which can also be recorded as an outlier set.

[0067] For a set of potentially abnormal data points Middle Data points , and its corresponding parameters are: In this embodiment, the authenticity of the abnormality of the potential abnormal data points is evaluated based on the continuous data of each acquisition terminal, so only the operating parameters of the acquisition terminal are retained, that is, , for each terminal operation parameter of a potential abnormal data point, obtain the relevant data of the 5 acquisition moments before and after it to obtain the time sequence of the terminal operation data. Exemplary explanation: There is a potential abnormal data point , which contains the following operating parameters: , with its power Take as an example, the relevant data of the five acquisition moments before and after are obtained to form the power time series, which are: It should be noted that the time series here does not include the operating parameters corresponding to the collection time of the potential abnormal data point. If the operating parameters of the collection time itself are abnormal due to an occasional failure of the collection terminal, by eliminating the relevant data of the collection time and comparing the relevant data of the adjacent collection time, it can help the subsequent operations to better eliminate the above-mentioned occasional abnormal situation. Here, the above power time series is recorded as , and correspondingly obtain the timing of other operating parameters, including temperature timing , humidity time series And voltage timing .

[0068] The present invention can more accurately calculate the real fluctuation of the abnormal point by acquiring the time series data before and after the abnormal point and excluding the data at the current abnormal moment. This method effectively avoids misjudgment caused by occasional failures or data collection errors, and improves the stability and reliability of abnormality detection.

[0069] So far, for the potential abnormal data point set Middle Data points , you can get its operating parameter timing , , as well as , and obtain the data point No. Distance Neighborhood The operating parameter timing of all other data points in the Distance Neighborhood The similarity of the operating parameter time series of other data points in the same data point is used to obtain the local fluctuation similarity of the data point. , the specific calculation method is:

[0070] ;

[0071] in Represents a set of potentially abnormal data points Middle Data points The local fluctuation similarity of represents a counting function; Represents data points No. Distance neighborhood; represents a set of operating parameters, in this embodiment, ; Represents the DTW distance calculation function; Indicates Data points Middle Operation parameter timing; Represents data points No. Distance Neighborhood Medium data point No. A sequence of operating parameters.

[0072] The present invention further enhances the recognition accuracy of abnormal data points by calculating the local fluctuation similarity. By comparing the multi-dimensional time series similarity between the abnormal point and other points in its neighborhood, it is possible to more accurately determine whether the fluctuation is caused by actual abnormal data. Compared with the prior art, this method not only considers the distance between the data point and the neighborhood, but also integrates the time series information of multi-dimensional parameters, effectively improving the system's fault tolerance to occasional failures.

[0073] So far, we have obtained data points Adaptive neighborhood size The calculation method is:

[0074] ;

[0075] in Represents a set of potentially abnormal data points Middle Data points Adaptive neighborhood size; Represents the rounding function, which is used to round off and retain the integer part; represents the initial fixed neighborhood size set in this embodiment; represents the exponential function; Represents data points The local fluctuation similarity of The larger the value, the more data points With its Distance Neighborhood If the gap between the multiple operating parameter sequences of the other data points in the dataset is large, the reason why the data point is classified as a potential abnormal data point set is more likely to be caused by the abnormality of the actual parameters. In this case, a smaller neighborhood size should be selected to improve the detection ability of local anomalies. should be smaller; otherwise, The smaller the time, the more data points With its Distance Neighborhood If the gap between the multiple operating parameter sequences of the other data points is large, the data point may be judged as a potential abnormal data point due to occasional failures such as the acquisition terminal. At this time, a larger neighborhood size should be selected to avoid oversensitivity. Should be bigger.

[0076] The adaptive neighborhood size calculation method further optimizes the performance of the LOF algorithm through dynamic adjustment based on local fluctuation similarity. This method uses an exponential function to dynamically adjust the neighborhood size according to the fluctuation similarity, thereby ensuring the consistency and accuracy of the algorithm under different abnormal conditions. Compared with the fixed neighborhood size of the prior art, the adaptive neighborhood size of the present invention is more flexible, can effectively cope with changes in equipment operation and environmental conditions, and improve the accuracy of anomaly detection.

[0077] Finally, for any data point , the calculation formula of LOF value is:

[0078] ;

[0079] in Represents data points LOF value; represents a set of potentially abnormal data points; Represents data points Improved adaptive neighborhood size After Distance neighborhood; Represents data points No. Distance Neighborhood Data points within represents the local reachable density calculation function; Represents data points The initial fixed neighborhood size corresponds to the Distance Neighborhood That is, when the data point does not belong to the abnormal data point set When , the neighborhood size is still the initial value , when the data point belongs to the abnormal data point set When , its neighborhood size is the improved adaptive neighborhood size .

[0080] S3. Complete anomaly detection in concentrator operation data management based on the LOF algorithm with improved adaptive neighborhood size.

[0081] In step S1, multiple operating parameters of the concentrator are collected and standardized to obtain a data frame. ; In step S2.1, multiple consecutive data frames closest to the current time constitute a data set , and converted into a parameter set ; Further based on the improved adaptive neighborhood size in step S2.3 The LOF algorithm has a parameter set Perform anomaly detection; finally, after calculating the LOF value each time and comparing it with the threshold α, the concentrator will identify all anomalies in real time.

[0082] If a data point is identified as an abnormal data point, it will be marked as "abnormal" in the system, and its data record, timestamp, device information and other related data will be saved in the log for subsequent analysis. At the same time, after identifying an abnormal data point, the system will immediately trigger the alarm mechanism and generate detailed alarm information. The alarm information will be sent to the remote monitoring platform through the communication interface of the concentrator to notify the operation and maintenance personnel, so as to manually check the abnormal data and help the operation and maintenance personnel evaluate the system operation status and provide data support for subsequent optimization.

[0083] The concentrator operation data management method of the present invention provides a dynamic and adaptive anomaly detection method by improving the LOF algorithm and combining the analysis of local density differences and multi-dimensional fluctuation similarities. The traditional LOF algorithm has the limitation of a fixed neighborhood size k, which can easily lead to false alarms or missed alarms, especially when the equipment status and environmental conditions change greatly. The present invention adaptively adjusts the k value so that the neighborhood size is dynamically adjusted according to the local density difference and fluctuation similarity of the data points. The introduction of the adaptive neighborhood size enables the LOF algorithm to select a smaller neighborhood when the fluctuation is large, thereby enhancing the sensitivity to abnormal points; when the fluctuation is small, a larger neighborhood is selected to avoid oversensitivity, thereby reducing false alarms and missed alarms. Compared with the prior art, the present invention significantly improves the anomaly detection accuracy of the concentrator in a dynamic environment, and provides a more stable and reliable data management solution for smart grids and smart metering systems.

[0084] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0085] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A concentrator operation data management method, characterized in that: The method comprises: Collect multiple parameters of the terminal device connected to the concentrator; the multiple parameters at any collection time constitute a parameter vector, and the multiple parameter vectors under the set collection duration constitute a data set; Performing anomaly detection on the data set using the LOF algorithm to obtain the LOF value of each parameter vector, and in response to the LOF value being greater than a set LOF threshold, triggering an alarm mechanism and notifying an operation and maintenance personnel; The LOF algorithm includes processing the initial neighborhood size to obtain an adaptive neighborhood size, the method is: ; Represents anomaly point set Middle Data points Adaptive neighborhood size; represents the rounding function; represents the initial neighborhood size; represents the exponential function; Represents data points The local fluctuation similarity of The calculation method of the local fluctuation similarity is specifically as follows: Any parameter vector is used as the target data point of the LOF algorithm, and the local density of the target data point is calculated; the local density and the first The mean of the difference in local density from the rest of the data points in the neighborhood constitutes the local density difference; All data points corresponding to local density differences greater than the set threshold constitute an outlier set; Obtain the parameter time series of all abnormal points in the abnormal point set, and calculate the time series between any abnormal point and its The similarity between the parameter time series of the other data points in the distance neighborhood is the local fluctuation similarity; the relationship is: ; represents a counting function; Represents data points No. Distance neighborhood; Represents a parameter set; Represents the DTW distance calculation function; Represents data points Middle Parameter timing; Represents data points No. Distance Neighborhood Medium data point No. A sequence of parameters.

2. A concentrator operation data management method according to claim 1, characterized in that: The calculation method of the local density difference is specifically as follows: ； in Indicates Data points The local density difference of Indicates Data points No. Distance neighborhood; represents a counting function; Represents data points The local density of Represents data points The local density of .

3. A concentrator operation data management method according to claim 2, characterized in that: The calculation method of the local density is specifically as follows: ； in Represents data points The local density of Represents data points No. distance.

4. A concentrator operation data management method according to claim 1, characterized in that: Obtaining the parameter time series of all abnormal points in the abnormal point set includes: For any parameter of any abnormal point, obtain the parameter data of a set number of acquisition moments before and after it, and all parameter data constitute the parameter time series of the corresponding abnormal point; The parameter time series does not include parameter data at a corresponding acquisition time of a corresponding abnormal point.

5. A concentrator operation data management method according to claim 1, characterized in that: The set threshold is obtained by using the maximum inter-class variance method based on all local density differences.

6. A concentrator operation data management method according to claim 1, characterized in that: It also includes preprocessing the multiple parameters, specifically: Performing signal smoothing processing on the multiple parameters using a bandpass filter; The multiple parameters after smoothing are processed using maximum and minimum value normalization to obtain standard data.

Images