WO2020006841A1 - 用电量异常检测方法、装置、终端及计算机可读存储介质 - Google Patents

用电量异常检测方法、装置、终端及计算机可读存储介质 Download PDF

Info

Publication number
WO2020006841A1
WO2020006841A1 PCT/CN2018/102145 CN2018102145W WO2020006841A1 WO 2020006841 A1 WO2020006841 A1 WO 2020006841A1 CN 2018102145 W CN2018102145 W CN 2018102145W WO 2020006841 A1 WO2020006841 A1 WO 2020006841A1
Authority
WO
WIPO (PCT)
Prior art keywords
sequence
feature
average
power consumption
values
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2018/102145
Other languages
English (en)
French (fr)
Inventor
郑立颖
徐亮
肖京
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020006841A1 publication Critical patent/WO2020006841A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/18Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis

Definitions

  • the present application relates to the technical field of abnormality detection, and in particular, to a method for detecting abnormality in power consumption, an abnormality detecting device, an abnormality detecting terminal, and a computer-readable storage medium.
  • the method based on moving average will be greatly outlied Impact, resulting in a sequence after the outlier is also judged to be abnormal;
  • the disadvantage of starting from the perspective of distance difference is that the power consumption time series generally has a wide span and large data, the distance calculation is more complicated, and the time-consuming and efficient is not efficient. high. Therefore, the traditional power consumption abnormality detection method has low accuracy, is easy to be disturbed by abnormal values, has complicated calculations, and has low detection efficiency.
  • the main purpose of this application is to provide a method for detecting abnormality in power consumption, an abnormality detection device, an abnormality detection terminal, and a computer-readable storage medium. Interference, and the calculation is complicated, and the detection efficiency is low.
  • an embodiment of the present application provides a method for detecting abnormality in power consumption, where the method for detecting abnormality in power consumption includes:
  • the corresponding feature data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • the step of calculating the relative entropy between each feature data sequence and the average feature sequence according to a relative entropy algorithm includes:
  • the method further includes:
  • the average distribution curve and the abnormal distribution curve are respectively displayed in a preset coordinate system for user identification.
  • the method further includes:
  • the relative entropy of the abnormal distribution curve and the offset differences of the target time points are displayed in a preset coordinate system for user analysis.
  • the method further includes:
  • the step of collecting sequence characteristic values at each target time point from all power consumption time series at a preset time interval to generate a characteristic data sequence corresponding to each power consumption time series further includes :
  • a new power consumption time series is imported from a preset sample database in a loop, and a new characteristic data sequence is obtained according to the new power consumption time series, and the new statistics are collected. The number of sequences until the number of new sequences is not less than the first threshold.
  • a new power consumption time series is imported from a preset sample database in a loop, and new features are obtained according to the new power consumption time series.
  • Data sequence, counting the number of new sequences, after the step that the number of new sequences is not less than the first threshold further includes:
  • the steps of obtaining a new characteristic data sequence according to the new power consumption time series and counting the number of new values are performed cyclically until the number of new values is not less than the second threshold.
  • the present application also provides an abnormality detection device.
  • the abnormality detection device includes:
  • a collection module configured to collect sequence characteristic values at each target time point from all power consumption time series at a preset time interval to generate a characteristic data sequence corresponding to each power consumption time series;
  • a first calculation module configured to average the sequence feature values corresponding to each target time point in all the feature data sequences to obtain an average feature value at each target time point;
  • a first generating module configured to generate a corresponding average feature sequence from all the average feature values in time sequence
  • a second calculation module configured to calculate a relative entropy between each feature data sequence and an average feature sequence according to a relative entropy algorithm
  • a setting module is configured to set the corresponding characteristic data sequence corresponding to the relative entropy to an abnormal sequence if the relative entropy is greater than a preset value.
  • the present application further provides an anomaly detection terminal.
  • the anomaly detection terminal includes: a memory, a processor, a communication bus, and anomaly detection readable instructions stored on the memory.
  • the communication bus is used for For implementing the communication connection between the processor and the memory;
  • the processor is configured to execute the pre-authorization processing readable instruction to implement the following steps:
  • the corresponding feature data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • the present application further provides a readable storage medium, where the readable storage medium stores one or more readable instructions, and the one or more readable instructions may be replaced by one or more
  • the processor executes to:
  • the corresponding feature data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • This application collects sequence feature values at each target time point from all power consumption time series at preset time intervals to generate a feature data sequence corresponding to each power consumption time series; for all the feature data sequences The sequence feature values corresponding to each target time point in the calculation are averaged to obtain the average feature value at each target time point; the corresponding average feature sequence is generated from all the average feature values in chronological order; calculated according to the relative entropy algorithm The relative entropy between each feature data sequence and the average feature sequence; if the relative entropy is greater than a preset value, the corresponding feature data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • This application innovatively forms a distribution curve at the time point of each day of the time series, and determines whether the data is abnormal by judging the difference between the daily distribution curve and the average distribution curve.
  • the time point is used to screen the data to improve the reliability of the sample without a lot of calculations. It solves the technical problems of traditional power consumption abnormality detection methods with low accuracy, easy to be disturbed by outliers, complicated calculations, and low detection efficiency.
  • the accuracy of power consumption abnormality detection is improved, the interference of abnormal values is eliminated, the calculation complexity is reduced, and the detection efficiency is improved.
  • FIG. 1 is a schematic flowchart of a first embodiment of a method for detecting abnormal power consumption according to this application;
  • FIG. 2 is a schematic diagram of function modules of the abnormality detection device of the present application.
  • FIG. 3 is a schematic diagram of a device structure of a hardware operating environment involved in a method according to an embodiment of the present application.
  • the method for detecting abnormality in power consumption includes:
  • Step S10 Collect sequence characteristic values at each target time point from all power consumption time series at a preset time interval to generate a characteristic data sequence corresponding to each power consumption time series;
  • the power consumption time series refers to a series of power consumption values arranged in order of their occurrence time.
  • an abnormality detection sample is provided in the power consumption time series, and the system will perform data collection on it.
  • this embodiment collects data in the power consumption time series in a unified manner, and all the power consumption time series in the collection process are collected after a preset time interval.
  • This preset time interval can be adjusted according to actual business needs. For example, one data in the time series is collected every one hour to generate a 24-hour distribution data sequence; one data may be collected every half an hour to generate a 48-hour distribution data sequence, and so on. In this way, the system generates a characteristic data sequence corresponding to each power consumption time series.
  • the feature data sequence is a feature value at each target time point collected at a regular time interval
  • the feature data sequence in this embodiment wins the sequence feature value according to the order of the target time points.
  • each data point corresponds to a target time point and a sequence characteristic value.
  • the characteristic data sequence A includes a1 (0, 2), a2 (4, 2.5), a3 (8, 4), a4 (12, 2.5), a5 (16, 2), and a6 (20, 2).
  • Six data points Six data points. Through these six data points, the characteristic data sequence A can reflect the actual distribution of the day as a whole.
  • Step S20 Perform an average calculation on sequence feature values corresponding to each target time point in all the feature data sequences to obtain an average feature value at each target time point;
  • the system can perform average calculation on the characteristic values of the target time points in all the characteristic data sequences. That is, there are multiple common target time points in different feature data sequences.
  • the system will obtain the sequence feature values at each corresponding target time point and average the feature values at the same target time point. For example, assume that the power consumption sequence obtained by the system includes sequence A and sequence B, where:
  • Step S30 Generate a corresponding average feature sequence from all the average feature values in time sequence
  • the system can obtain the average feature values of the sequence feature values at each target time point. And these average feature values have their corresponding time point order, so the system can sort and transform each average feature value into a corresponding average feature sequence according to the time point order.
  • Step S40 Calculate the relative entropy between each feature data sequence and the average feature sequence according to the relative entropy algorithm
  • Relative average entropy is calculated with each feature data sequence.
  • Relative entropy also known as Kullback divergence, is a method to describe the difference between two probability distributions, P (true distribution) and Q (theoretical distribution).
  • P true distribution
  • Q theoretical distribution
  • the relative entropy itself represents the data dispersion of the feature data sequence and the average feature sequence.
  • the average feature sequence is closest to the theoretical normal distribution, and the relative entropy of the daily feature data sequence and the average feature sequence is greater. It means that the greater the data dispersion, the larger the deviation of the characteristic data sequence from the normal distribution. Therefore, the relative entropy itself can be used as an abnormality detection parameter of each characteristic data sequence.
  • the step S40 includes:
  • Step S41 extracting target sequence feature values with consistent values in each feature data sequence, and extracting target average feature values with consistent values in the average feature sequence;
  • Each feature data sequence may have the same feature value, or the feature values are different, but the feature values with a difference of about 0.1-0.9 are classified into the same class by the system setting, and the system will extract these feature values as each The target sequence feature value in the feature data sequence. In the same way, the system will extract the average eigenvalues of the same type in the average eigenvalue sequence as the target average eigenvalues.
  • Step S42 Count the first probability set of the target sequence feature value in the corresponding feature data sequence as Pi, and count the second probability set of the target average feature value in the average feature sequence.
  • Count the appearance probability of each target sequence feature value in the corresponding feature data sequence For example, there are 10 target sequence feature values in total, and the target sequence feature value of value A appears twice.
  • the first probability of the target sequence feature value is 0.2.
  • the system can obtain the second probability set Qi of the target average feature value in the average feature sequence.
  • Step S43 Substituting Pi and Qi into the following relative entropy algorithm formula for calculation to obtain the relative entropy D kl between each feature data sequence and the average feature sequence:
  • the relative entropy D kl between each feature data sequence and the average feature sequence is directly obtained.
  • the target sequence with the same value in the current feature data sequence A has a feature value of 1, 2, 3, and the first probability set when the values are 1, 2, 3 is (0.2, 0.4, 0.4); and in the current average feature sequence, The second probability set of the average eigenvalues of the targets with the same values is 1, 2, 3 is (0.4, 0.2, 0.4);
  • step S50 if the relative entropy is greater than a preset value, the corresponding characteristic data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • a preset value is set, and the preset value serves as a reference threshold value of relative entropy, and provides a determination criterion for determining whether a characteristic data sequence is an abnormal sequence.
  • the relative entropy represents the degree of dispersion of the overall distribution of the feature data sequence. If the relative entropy is greater than a preset value, it means that the data dispersion of the current feature data sequence and the average feature sequence is greater than the warning line, that is, the The sequence eigenvalues on the whole do not agree with the sequence eigenvalues of the average feature sequence close to the normal distribution. On the contrary, the overall distribution of the sequence eigenvalues in the characteristic data sequence shows a tendency to deviate from the normal distribution.
  • this kind of characteristic data sequence that deviates from the normal distribution is largely not a normal sequence. Therefore, this deviation trend needs a quantified index as a reference, and this reference is a preset value.
  • the preset value limits the data range of the relative entropy, which is less than or equal to the preset value, which indicates that the feature value of the feature data sequence is very close to the average feature value in the average feature sequence, and the distribution situation conforms to the true distribution.
  • the relative entropy is greater than a preset value, it means that the dispersion of the characteristic data sequence exceeds the warning line set by the system. At this time, the system will set the characteristic data sequence corresponding to the relative entropy as an abnormal sequence.
  • This application collects sequence feature values at each target time point from all power consumption time series at preset time intervals to generate a feature data sequence corresponding to each power consumption time series; for all the feature data sequences The sequence feature values corresponding to each target time point in the calculation are averaged to obtain the average feature value at each target time point; the corresponding average feature sequence is generated from all the average feature values in chronological order; calculated according to the relative entropy algorithm The relative entropy between each feature data sequence and the average feature sequence; if the relative entropy is greater than a preset value, the corresponding feature data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • This application innovatively forms a distribution curve at the time point of each day of the time series, and determines whether the data is abnormal by judging the difference between the daily distribution curve and the average distribution curve.
  • the time point is used to screen the data to improve the reliability of the sample without a lot of calculations. It solves the technical problems of traditional power consumption abnormality detection methods with low accuracy, easy to be disturbed by outliers, complicated calculations, and low detection efficiency.
  • the accuracy of power consumption abnormality detection is improved, the interference of abnormal values is eliminated, the calculation complexity is reduced, and the detection efficiency is improved.
  • a second embodiment of the power consumption abnormality detection method of the present application is proposed.
  • the difference from the foregoing embodiment is that after step S50, the method further includes:
  • Step S60 Generate corresponding average distribution curves and abnormal distribution curves from the average feature sequence and the abnormal sequence, respectively;
  • the system in order to facilitate the user to visually check the difference in sequence feature values between the analyzed abnormal sequence and the average feature sequence, the system will convert the average feature sequence and the abnormal sequence into corresponding average distribution curves and Anomaly distribution curve. Since the average feature sequence and the sequence feature value in the anomaly sequence are respectively sorted in chronological order, the system can generate an average distribution curve and an abnormal distribution curve according to the target time point and the corresponding sequence feature value.
  • step S70 the average distribution curve and the abnormal distribution curve are respectively displayed in a preset coordinate system for user identification.
  • the system will construct a preset coordinate system, and display the average distribution curve and the abnormal distribution curve in the form of a visualization curve in the preset coordinate system.
  • the preset coordinate system in order to improve the recognition degree, the normal distribution curve and the abnormal distribution curve will be displayed by different marking forms, for example, the abnormal distribution curve is marked as a red curve, which indicates the power consumption corresponding to the abnormal distribution curve.
  • the characteristic data sequence in the time series is an anomalous sequence to facilitate user identification.
  • the relative entropy of the anomalous sequence and the average distribution curve is larger than the preset value, so the dispersion is large, and the difference can be directly observed in the preset coordinate system.
  • step S70 the method further includes:
  • Step A Calculate based on all sequence feature values on the anomalous distribution curve and average feature values at corresponding target time points on the average distribution curve to obtain an offset difference;
  • step B if a trigger operation based on the abnormal distribution curve is detected, the relative entropy of the abnormal distribution curve and the offset difference between the target time points are displayed in a preset coordinate system for user analysis.
  • the system When the user triggers the abnormal distribution curve (such as clicking to check the data in the abnormal distribution curve), the system will directly display the relative entropy of the abnormal distribution curve and the offset difference of each target time point in the preset coordinate system. In this way, according to the offset difference, the user can know the change trend of power consumption at different target points in time, and can analyze the cause of the abnormality by analyzing the change trend, such as a short circuit, a failure of the meter, and the like.
  • a fourth embodiment of the power consumption abnormality detection method of the present application is proposed.
  • the difference from the foregoing embodiment is that the method further includes:
  • Step S80 When the abnormal control ratio is received, obtain a target preset value mapped to the abnormal control ratio from a preset mapping table;
  • the abnormality judgment criteria for anomaly detection may change. For example, if the power is cut or cut off for a certain period of time, the power consumption will change significantly, and the system will judge that it is abnormal. In fact, this change is not caused by abnormalities, but caused by known and controllable reasons. Therefore, this abnormal situation needs to be eliminated. That is to say, this application can adjust the abnormality judgment standard according to actual business requirements. For example, if there is a need to limit power within the current month, the corresponding abnormal sequence will rise. In order to eliminate the abnormal sequence, the judgment criterion can be changed, and the determination of the judgment criterion is related to the preset value.
  • the abnormal control ratio refers to filtering all abnormal sequences according to the ratio.
  • the system stores a preset mapping table.
  • the system receives an abnormal control ratio input from the outside, and finds a target preset value mapped to the ratio in a preset mapping table.
  • the abnormality control ratio is to determine 5% of all the characteristic data sequences as abnormal sequences. Even if the characteristic data sequences are normal sequences, by raising the judgment standard, some normal sequences with large fluctuations are determined as abnormal sequences.
  • step S90 the current default preset value is adjusted to the target preset value to adjust the judgment criterion of the abnormal sequence.
  • the default preset value of the current system will no longer meet the current actual business needs, and the system will update the default preset value to the target preset value, thereby adjusting the judgment criterion of the abnormal sequence.
  • a fifth embodiment of the power consumption abnormality detection method of the present application is proposed.
  • the difference from the foregoing embodiment is that after step S10, the method further includes:
  • Step S100 Count the number of sequences of all the characteristic data sequences currently described.
  • step S110 if the number of sequences is less than the first threshold, a new power consumption time series is imported from a preset sample database cyclically, and a new characteristic data sequence is obtained according to the new power consumption time series. Count the number of new sequences until the number of new sequences is not less than the first threshold.
  • the system can set a first threshold, and the first threshold can be dynamically adjusted according to the actual business requirements.
  • the first threshold is the minimum threshold of the number of sequences. If the number of sequences is less than the first threshold, it indicates that the number of the current sequence is too small, which affects the accuracy of the judgment of the abnormal sequence.
  • the system needs to import a new power consumption time series from a preset sample database, and obtain the new feature data sequence by performing the steps in the first embodiment on the power consumption time series.
  • the system will continuously obtain new feature data sequences, count all the current feature data sequences, and then perform a judgment based on the first threshold until the number of sequences is not less than the first threshold.
  • sequence feature value at the target time point in the feature data sequence is a missing value (the system is collecting the time series Equipment failure or interference caused sequence feature values to be unsuccessfully collected), which means that the feature data sequence cannot play any role in anomaly detection and relative entropy calculation process, it will also increase the computational complexity and fail to provide the correct data source . Therefore, the characteristic data sequence will be deleted by the system as an invalid data sequence, which can reduce the computational complexity and avoid the introduction of invalid data, thereby providing the reliability of anomaly detection.
  • step S110 the method further includes:
  • Step S120 if the number of sequence feature values at any target time point in all the feature data sequences is less than the second threshold value, import a new power consumption time series from the preset sample database;
  • the sequence characteristic value is invalid data.
  • the monthly power consumption corresponding to the A power consumption time series is normal, but all the power consumption values include one or more invalid data, then These invalid data will cause the accuracy of the calculation results to shift.
  • the system In order to ensure the data referentiality of the sequence eigenvalues, the system usually specifies that the number of sequence eigenvalues must reach a reasonable value to ensure that the sample can be covered in a wide range and improve the accuracy of anomaly detection. Therefore, the system sets a second threshold, and the second threshold will be used as a reference threshold for the number of values. The system will count the number of sequence feature values at any target time point in all the feature data sequences. If the number of values is less than the second threshold, it means that the data sample size of the current sequence feature values does not meet the standard, and the detection accuracy of possible anomaly detection There is an influence, so it is necessary to increase the sequence feature value of the feature data sequence. At this time, the system will import the new power consumption time series from the preset sample database.
  • step S130 the steps of obtaining a new characteristic data sequence according to the new power consumption time series and counting the new number of values are performed cyclically until the new number of values is not less than the second threshold.
  • the system After obtaining a new power consumption time series, the system will collect data to obtain a new characteristic data series, and count the number of sequence characteristic values at any target time point until the number of values is not less than the second Threshold.
  • FIG. 3 is a schematic diagram of a device structure of a hardware operating environment involved in the method according to the embodiment of the present application.
  • the terminal may be a terminal device such as a PC or a portable computer.
  • the abnormality detection terminal may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection and communication between the processor 1001 and the memory 1005.
  • the memory 1005 may be a high-speed RAM memory or a non-volatile memory (for example, a magnetic disk memory).
  • the memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • the abnormality detection terminal may further include a user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like.
  • the user interface may include a display, an input unit such as a keyboard, and the optional user interface may further include a standard wired interface and a wireless interface.
  • the network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 which is a computer storage medium (the computer storage medium may be a non-volatile storage medium), may include an operating system, a network communication module, and anomaly detection readable instructions.
  • the operating system is a readable instruction for managing and controlling the hardware and software resources of the anomaly detection terminal, and supports the operation of the anomaly detection readable instructions and other software and / or readable instructions.
  • the network communication module is used to implement communication between components in the memory 1005 and to communicate with other hardware and software in the abnormality detection terminal.
  • the processor 1001 is configured to execute the anomaly detection readable instructions stored in the memory 1005, and implement the following steps:
  • the corresponding feature data sequence corresponding to the relative entropy is set as an abnormal sequence.
  • the abnormality detection device includes:
  • a collection module configured to collect sequence characteristic values at each target time point from all power consumption time series at a preset time interval to generate a characteristic data sequence corresponding to each power consumption time series;
  • a first calculation module configured to average the sequence feature values corresponding to each target time point in all the feature data sequences to obtain an average feature value at each target time point;
  • a first generating module configured to generate a corresponding average feature sequence from all the average feature values in time sequence
  • a second calculation module configured to calculate a relative entropy between each feature data sequence and an average feature sequence according to a relative entropy algorithm
  • a setting module is configured to: if the relative entropy is greater than a preset value, set the characteristic data sequence corresponding to the relative entropy to an abnormal sequence.
  • the second calculation module includes:
  • the second calculation module includes:
  • An extraction unit configured to extract target sequence feature values with consistent values in each of the feature data sequences, and extract target average feature values with consistent values in the average feature sequence;
  • a statistics unit configured to count a first probability set of the target sequence feature value in the corresponding feature data sequence, set it as Pi, and count a second probability of the target average feature value in the average feature sequence Set, set to Qi;
  • the calculation unit substitutes Pi and Qi into the following relative entropy algorithm formula for calculation to obtain the relative entropy D kl between each feature data sequence and the average feature sequence:
  • the abnormality detection device further includes:
  • a second generating module configured to generate the corresponding average distribution curve and the abnormal distribution curve respectively from the average feature sequence and the abnormal sequence
  • a first display module is configured to display the average distribution curve and the abnormal distribution curve respectively in a preset coordinate system for user identification.
  • the abnormality detection device further includes:
  • a calculation module configured to calculate based on all the sequence feature values on the abnormal distribution curve and the average feature value at a corresponding target time point on the average distribution curve to obtain an offset difference
  • a second display module configured to display, if a trigger operation based on the abnormal distribution curve is detected, the relative entropy of the abnormal distribution curve and the offset difference of each target time point in a preset coordinate system for User analysis.
  • the abnormality detection device further includes:
  • a receiving module configured to obtain a target preset value mapped to the abnormal control ratio from a preset mapping table when the abnormal control ratio is received;
  • An adjustment module is configured to adjust a current default preset value to the target preset value to adjust a judgment criterion of the abnormal sequence.
  • the abnormality detection device further includes:
  • a statistics module configured to count the number of sequences of all the characteristic data sequences currently described
  • a first judgment module configured to: if the number of sequences is less than a first threshold, import a new time series of power consumption from a preset sample database cyclically, and obtain a new one according to the new time series of power consumption The characteristic data sequence counts the number of new sequences until the number of new sequences is not less than the first threshold.
  • the abnormality detection device further includes:
  • An import module for importing a new time series of electricity consumption from the preset sample database if the number of sequence feature values at any target time point in all the feature data sequences is less than the second threshold;
  • the second judgment module is configured to perform the steps of obtaining a new characteristic data sequence according to the new power consumption time series and counting the new number of numbers until the new number of numbers is not less than the second threshold.
  • the application also provides a computer-readable storage medium, where the computer-readable storage medium stores one or more readable instructions, and the one or more readable instructions may also be executed by one or more processors. To implement steps of the method for detecting an abnormality in power consumption according to any one of the above.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Operations Research (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Algebra (AREA)
  • Evolutionary Biology (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

一种用电量异常检测方法、异常检测装置、异常检测终端及计算机可读存储介质,该用电量异常检测方法包括:按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成各用电量时间序列对应的特征数据序列(S10);对所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得各目标时间点上的平均特征值(S20);按时间顺序将所有平均特征值生成对应的平均特征序列(S30);根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵(S40);若相对熵大于预设值,则将该相对熵对应的特征数据序列设为异常序列(S50)。该方法提高了用电量异常检测的精确度,排除了异常值的干扰,降低计算复杂度,提升了检测效率。

Description

用电量异常检测方法、装置、终端及计算机可读存储介质
本申请要求于2018年07月06日提交中国专利局、申请号为201810744182.1、发明名称为“用电量异常检测方法、装置、终端及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及异常检测技术领域,尤其涉及一种用电量异常检测方法、异常检测装置、异常检测终端及计算机可读存储介质。
背景技术
目前对于用电量时间序列的异常检测大多从用电量时间序列角度(例如同比差异、环比差异、移动平均、ARIMA预测等)和距离差异角度(例如比较某一时间窗口内序列与其他序列的欧氏距离)入手,从用电量时间序列角度入手的缺点在于:1.同比差异环比差异考虑的点太片面,很可能造成误判;2.基于移动平均的方法会很大程度被异常值影响,造成异常值之后的一段序列也被判断为异常;从距离差异角度入手的缺点在于,用电量时间序列一般来说跨度较广且数据较大,距离计算较为复杂,耗时大效率不高。因此,传统的用电量异常检测方法精确度不高,容易被异常值干扰,且计算复杂,检测效率低下。
发明内容
本申请的主要目的在于提供一种用电量异常检测方法、异常检测装置、异常检测终端及计算机可读存储介质,旨在解决传统的用电量异常检测方法精确度不高,容易被异常值干扰,且计算复杂,检测效率低下的技术问题。
为实现上述目的,本申请实施例提供一种用电量异常检测方法, 所述用电量异常检测方法包括:
按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
按时间顺序将所述所有平均特征值生成对应的平均特征序列;
根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
可选地,所述根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵的步骤包括:
提取所述每个特征数据序列中数值一致的目标序列特征值,并提取所述平均特征序列中数值一致的目标平均特征值;
统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
Figure PCTCN2018102145-appb-000001
可选地,所述若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列的步骤之后还包括:
将所述平均特征序列和所述异常序列分别生成对应的平均分布曲线和异常分布曲线;
将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别。
可选地,所述将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别的步骤之后还包括:
基于所述异常分布曲线上的所有序列特征值,以及所述平均分布 曲线上对应目标时间点上的平均特征值进行计算,获得偏移差值;
若检测到基于所述异常分布曲线的触发操作,将该异常分布曲线的相对熵以及所述各目标时间点的偏移差值显示在预设坐标系中,以供用户分析。
可选地,所述方法还包括:
当接收到异常控制比例时,从预设映射表中获取与所述异常控制比例相映射的目标预设值;
将当前默认的预设值调整为所述目标预设值,以调整所述异常序列的判断标准。
可选地,所述按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列的步骤之后还包括:
统计当前所述所有特征数据序列的序列个数;
若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值。
可选地,所述若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值的步骤之后还包括:
若所有特征数据序列中任一目标时间点上的序列特征值的数值个数小于第二阈值,则从所述预设样本数据库中导入新的用电量时间序列;
循环执行根据新的用电量时间序列获取新的特征数据序列,并统计新的数值个数的步骤,直至新的数值个数不小于第二阈值。
本申请还提供一种异常检测装置,所述异常检测装置包括:
采集模块,用于按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
第一计算模块,用于对所述所有特征数据序列中各目标时间点对 应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
第一生成模块,用于按时间顺序将所述所有平均特征值生成对应的平均特征序列;
第二计算模块,用于根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
设置模块,用于若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
此外,为实现上述目的,本申请还提供一种异常检测终端,所述异常检测终端包括:存储器、处理器、通信总线以及存储在所述存储器上的异常检测可读指令,所述通信总线用于实现处理器与存储器间的通信连接;
所述处理器用于执行所述预授权处理可读指令,以实现以下步骤:
按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
按时间顺序将所述所有平均特征值生成对应的平均特征序列;
根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
此外,为实现上述目的,本申请还提供一种可读存储介质,所述可读存储介质存储有一个或者一个以上可读指令,所述一个或者一个以上可读指令可被一个或者一个以上的处理器执行以用于:
按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
按时间顺序将所述所有平均特征值生成对应的平均特征序列;
根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
本申请通过按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;按时间顺序将所述所有平均特征值生成对应的平均特征序列;根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。本申请创新性地将时间序列每一天的时间点构成分布曲线,通过将每天的分布曲线跟平均分布曲线的差异性进行判断,从而判定数据是否异常。利用时间点对数据进行筛选,提高样本可靠性,无需经过大量计算,解决了传统的用电量异常检测方法精确度不高,容易被异常值干扰,且计算复杂,检测效率低下的技术问题,提高了用电量异常检测的精确度,排除了异常值的干扰,降低计算复杂度,提升了检测效率。
附图说明
图1为本申请用电量异常检测方法第一实施例的流程示意图;
图2为本申请异常检测装置的功能模块示意图;
图3为本申请实施例方法涉及的硬件运行环境的设备结构示意图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供一种用电量异常检测方法,在用电量异常检测方法第一实施例中,参照图1,所述用电量异常检测方法包括:
步骤S10,按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
用电量时间序列是指将用电量数值按其发生的时间先后顺序排列而成的数列。本实施例中,用电量时间序列提供了异常检测样本,系统将对其进行数据采集。为保障对时间序列分析的步骤一致性,本实施例对用电量时间序列中的数据进行了统一采集,采集过程中所有的用电量时间序列都是经过预设时间间隔而采集的。这个预设时间间隔可根据实际业务需求进行调整。例如每隔一小时采集时间序列中的一个数据,生成24小时的分布数据序列;也可以每隔半小时采集一个数据,生成48小时的分布数据序列等等。这样系统便生成了各用电量时间序列对应的特征数据序列。
由于所述特征数据序列是以规范的时间间隔采集的各目标时间点上的特征值,因此本实施例中特征数据序列按照各目标时间点的先后顺序排列度赢得序列特征值。例如,系统按每四小时采集一个用电量数据所获取到的特征数据序列A中,每个数据点都对应有目标时间点以及序列特征值。例如,特征数据序列A中包括a1(0,2),a2(4,2.5),a3(8,4),a4(12,2.5),a5(16,2),a6(20,2)一共六个数据点,通过这六个数据点,特征数据序列A能够从整体上反映该天的实际分布情况。
步骤S20,对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
在获取到多个特征数据序列之后,系统可对所有特征数据序列中的目标时间点的特征值进行均值计算。即不同的特征数据序列中均有多个共同的目标时间点,系统将获取各个对应目标时间点上的序列特 征值,并将同一目标时间点的特征值进行平均计算。例如,假设系统获取到的用电量序列包括序列A和序列B,其中,
序列A在0时用电量为5,6时用电量为7,12时用电量为5,18时用电量为7;序列B在0时用电量为3,6时用电量为5,12时用电量为5,18时用电量为3。那么,序列A和序列B的0时平均用电量为(5+3)/2=4,6时平均用电量为(7+5)/2=6,12时平均用电量为(5+5)/2=5,18时平均用电量为(7+3)/2=5。
步骤S30,按时间顺序将所述所有平均特征值生成对应的平均特征序列;
通过对各目标时间点的特征值进行平均,系统可获得各目标时间点上的序列特征值的平均特征值。而这些平均特征值具有各自对应的时间点顺序,因此系统可根据时间点顺序将各个平均特征值整理转化为对应的平均特征序列。该平均特征序列反映了当前所有特征数据序列的整体分布情况。如根据以上所述,平均特征序列Z包括:z1=(0,4),z2=(6,6),z3=(12,5),z4=(18,5)。
步骤S40,根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
将平均特征序列与每个特征数据序列进行相对熵计算。相对熵又称KL散度(Kullback Leibler divergence),是描述两个概率分布P(真实分布)和Q(理论分布)差异的一种方法。当前平均特征序列和各特征数据序列中的序列值均是确定的,其分布情况也基于目标时间点,因此不同序列之间的离散程度可利用相对熵算法,计算每个特征数据序列与平均特征序列之间的相对熵差异。
本实施例中,相对熵本身代表了特征数据序列和平均特征序列的数据离散度,平均特征序列最接近于理论上的正常分布,而每天的特征数据序列与平均特征序列的相对熵越大,说明数据离散度越大,也就意味着特征数据序列与正常分布的偏移量越大。因此,相对熵本身可作为各个特征数据序列的异常检测参数。
具体地,所述步骤S40包括:
步骤S41,提取所述每个特征数据序列中数值一致的目标序列特 征值,并提取所述平均特征序列中数值一致的目标平均特征值;
每个特征数据序列中可能会存在相同的特征值,或者特征值不同,但通过系统设定将相差0.1-0.9左右的特征值归为同一类,系统将把这些特征值提取出来,作为每个特征数据序列中的目标序列特征值。同理,系统将把平均特征序列中数值一致的同一类的平均特征值提取出来,作为目标平均特征值。
步骤S42,统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
统计每个目标序列特征值在所属特征数据序列中的出现概率,例如目标序列特征值总共10个,数值为A的目标序列特征值出现了2次,则A目标序列特征值的第一概率为0.2,将A目标序列特征值的第一概率分别计算出来,获得第一概率集合Pi,同理,系统可获得目标平均特征值在平均特征序列中的第二概率集合Qi。
步骤S43,将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
Figure PCTCN2018102145-appb-000002
利用第一概率集合Pi和第二概率集合Qi以及预设的相对熵算法公式,直接获得各特征数据序列和平均特征序列之间的相对熵D kl
为方便解释说明,以下将通过一个例子解释说明:
假设当前特征数据序列A中数值一致的目标序列特征值为1、2、3,且取值1、2、3时的第一概率集合为(0.2,0.4,0.4);而当前平均特征序列中数值一致的目标平均特征值为1、2、3的第二概率集合为(0.4,0.2,0.4);
将第一概率集合和第二概率集合代入到相对熵算法公式中:D(P∥Q)=0.2×log(0.2/0.4)+0.4×log(0.4/0.2)+0.4×log(0.4/0.4)=0.2×□0.69+0.4×0.69+0.4×0=0.138。即0.138为当前特征数据序列A与平均特征序列的相对熵。
步骤S50,若所述相对熵大于预设值,则将该相对熵所述对应的 特征数据序列设为异常序列。
本实施例设置了一个预设值,该预设值作为相对熵的参考门限值,为确定特征数据序列是否为异常序列提供一个判定标准。在本实施例中,相对熵代表了特征数据序列整体分布的离散程度,若相对熵大于预设值,说明当前特征数据序列和平均特征序列的数据离散度大于警戒线,即特征数据序列中的序列特征值整体上并不与接近正常分布的平均特征序列的序列特征值趋同于一致,相反,特征数据序列中的序列特征值整体分布呈现了背离正常分布的趋势。
而这种背离正常分布的特征数据序列很大程度上都不是正常序列,因此,这种背离趋势需要有一个量化指标做参考,而这个参考即为预设值。预设值限定了相对熵的数据范围,小于或等于该预设值,说明特征数据序列的特征值与平均特征序列中的平均特征值非常接近,分布情况符合真实分布。而当相对熵大于预设值时,说明特征数据序列的离散度超过系统设定的警戒线,此时,系统将把该相对熵对应的特征数据序列设为异常序列。
本申请通过按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;按时间顺序将所述所有平均特征值生成对应的平均特征序列;根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。本申请创新性地将时间序列每一天的时间点构成分布曲线,通过将每天的分布曲线跟平均分布曲线的差异性进行判断,从而判定数据是否异常。利用时间点对数据进行筛选,提高样本可靠性,无需经过大量计算,解决了传统的用电量异常检测方法精确度不高,容易被异常值干扰,且计算复杂,检测效率低下的技术问题,提高了用电量异常检测的精确度,排除了异常值的干扰,降低计算复杂度,提升了检测效率。
进一步地,在本申请用电量异常检测方法第一实施例的基础上, 提出本申请用电量异常检测方法第二实施例,与前述实施例的区别在于,所述步骤S50之后还包括:
步骤S60,将所述平均特征序列和所述异常序列分别生成对应的平均分布曲线和异常分布曲线;
本实施例中,为方便用户直观地查看分析异常序列和平均特征序列在序列特征值上的差异,系统在确定异常序列之后,将把平均特征序列和异常序列分别转化为对应的平均分布曲线和异常分布曲线。由于平均特征序列和异常序列中的序列特征值都是以时间顺序分别排序的,因此按时间先后顺序系统可根据目标时间点以及各自对应的序列特征值生成平均分布曲线和异常分布曲线。
步骤S70,将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别。
将各序列可视化为曲线的意义在于,用户可以直观地观测并识别哪些用电量时间序列偏离了正常分布情形。
系统将构建预设坐标系,并把平均分布曲线和异常分布曲线以可视化曲线的形式显示在预设坐标系中。在所述预设坐标系中,为提高辨识度,正常分布曲线和异常分布曲线将通过不同的标记形式进行显示,例如将异常分布曲线标记为红色曲线,表明该异常分布曲线对应的用电量时间序列中的特征数据序列为异常序列,以方便用户识别。异常序列与平均分布曲线的相对熵大于预设值,因此离散度较大,在预设坐标系中可直接观测到该差异性。
进一步地,在本申请用电量异常检测方法第二实施例的基础上,提出本申请用电量异常检测方法第三实施例,与前述实施例的区别在于,所述步骤S70之后还包括:
步骤A,基于所述异常分布曲线上的所有序列特征值,以及所述平均分布曲线上对应目标时间点上的平均特征值进行计算,获得偏移差值;
进一步地,在预设坐标系中显示异常分布曲线和平均分布曲线可方便观看,但用户可能需要精确地分析结果。例如房东需要对某个租户的异常用电量进行细致分析,或者用户需要调查异常用电量中的具 体细节,为方便用户进行比对分析,系统将基于异常分布曲线中各目标时间点上的序列特征值与平均分布曲线上对应的平均特征值进行差值计算,以得到每个目标时间点上异常分布曲线与平均分布曲线的偏移差值。例如异常分布曲线上5时的序列特征值为10,而平均分布曲线上5时的平均特征值为5,则偏移差值为10-5=5。
步骤B,若检测到基于所述异常分布曲线的触发操作,将该异常分布曲线的相对熵以及所述各目标时间点的偏移差值显示在预设坐标系中,以供用户分析。
当用户触发异常分布曲线(如点击查探异常分布曲线中的数据)时,系统将把该异常分布曲线的相对熵以及各目标时间点的偏移差值直接显示在预设坐标系中。这样根据偏移差值,用户可得知不同目标时间点上用电量的变化趋势,并可通过变化趋势分析判断产生该异常的原因,例如电路短路,电表仪失灵等等。
进一步地,在本申请用电量异常检测方法第三实施例的基础上,提出本申请用电量异常检测方法第四实施例,与前述实施例的区别在于,所述方法还包括:
步骤S80,当接收到异常控制比例时,从预设映射表中获取与所述异常控制比例相映射的目标预设值;
现实场景中,异常检测的异常判断标准可能会发生改变。例如某段时间内限电或者断电,那么用电量将发生大幅度变化,系统将判断这是异常,但实际上这种变化并非是因为异常,而是由可知可控的原因导致的,因此需要将该异常情况进行排除。也就是说,本申请可根据实际业务需求调整异常判断标准。例如,当前1个月内需要限电,那么相对应的异常序列将会上升,为将异常序列排除,可对判断标准进行更改,判断标准的确定与预设值相关。而异常控制比例是指对所有异常序列按比例进行过滤。
本实施例中,系统保存有预设映射表。系统接收到外界输入的异常控制比例,在预设映射表中找到与该比例相映射的目标预设值。例如异常控制比例是将所有特征数据序列中5%的序列确定为异常序列,即使特征数据序列为正常序列,通过提高判断标准,将部分波动 较大的正常序列确定为异常序列。
步骤S90,将当前默认的预设值调整为所述目标预设值,以调整所述异常序列的判断标准。
由上可知,当前系统默认的预设值将不再适应当前的实际业务需求,系统将把默认的预设值更新调整为目标预设值,从而调整异常序列的判断标准。
进一步地,在本申请用电量异常检测方法第一实施例的基础上,提出本申请用电量异常检测方法第五实施例,与前述实施例的区别在于,所述步骤S10之后还包括:
步骤S100,统计当前所述所有特征数据序列的序列个数;
在现实情况中,可能出现特征数据序列较少,导致影响异常序列的判断精度。例如,当前只有3月5日、4月5号和6月5号的特征数据序列进入异常检测,且该3月5日为异常序列,但这不能说明每个3月5日的特征数据序列都是异常序列。因此不同月份5日的特征数据序列需要达到足够的样本量。故,系统将统计当前所有特征数据序列的序列个数。
步骤S110,若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值。
根据实际业务需求,系统可设定第一阈值,该第一阈值可根据实际业务需求而动态调整。第一阈值即为序列个数的最低门限值,若序列个数小于第一阈值,说明当前序列个数过少,对异常序列的判断精确度造成影响。系统需从预设的样本数据库中导入新的用电量时间序列,并通过对用电量时间序列执行第一实施例中的步骤,获取到新的特征数据序列。
在本实施例中,系统将不断获取到新的特征数据序列,并将当前所有特征数据序列进行序列个数统计,再进行基于第一阈值的判断,直至序列个数不小于第一阈值。通过上述步骤,即可保障特征数据序列能够提供足够的数据样本,从而提高异常序列的判断精确度。
进一步地,正常情况下,特征数据序列中各目标时间点上都有对应的序列特征值,但若是该特征数据序列中处于目标时间点上的序列特征值为缺失值(系统在收集该时间序列时发生设备故障或干扰导致序列特征值未成功被采集到),意味着该特征数据序列无法为异常检测以及相对熵计算过程起到任何效用,还会增加计算复杂度,无法提供正确的数据源。因此该特征数据序列将作为无效数据序列被系统删除,既可以减轻计算复杂度,又可以避免引入无效数据,从而提供异常检测的可靠性。
进一步地,在本申请用电量异常检测方法第五实施例的基础上,提出本申请用电量异常检测方法第六实施例,与前述实施例的区别在于,所述步骤S110之后还包括:
步骤S120,若所有特征数据序列中任一目标时间点上的序列特征值的数值个数小于第二阈值,则从所述预设样本数据库中导入新的用电量时间序列;
本实施例中,若发生特征数据序列中的某个目标时间点的序列特征值为空值或缺失值时,将导致当前特征数据序列的序列个数减少了1个。即序列特征值为无效数据,例如用电量统计序列中,A用电量时间序列对应的该月用电量是正常的,但是所有用电量数值中包括了一个或多个无效数据,那么这些无效数据对于计算结果会造成计算结果的精度偏移。
而系统为保障序列特征值的数据参考性,通常会指定序列特征值的数值个数必须达到一个合理数值,以确保能够大范围覆盖样本,提高异常检测的精确性。因此系统设定了一个第二阈值,所述第二阈值将作为数值个数的参考门限值。系统将统计所有特征数据序列中任一目标时间点的序列特征值的数值个数,若数值个数小于第二阈值,说明当前序列特征值的数据样本量不达标,可能异常检测的检测精确度存在影响,因此需要增加特征数据序列的序列特征值。此时系统将从预设样本数据库中导入新的用电量时间序列。
步骤S130,循环执行根据新的用电量时间序列获取新的特征数据序列,并统计新的数值个数的步骤,直至新的数值个数不小于第二 阈值。
获取到新的用电量时间序列之后,系统将对其进行数据采集,得到新的特征数据序列,并统计任一目标时间点上序列特征值的数值个数,直至数值个数不小于第二阈值。
以下将通过举例进行解释说明,当前特征数据序列总共有5个,对应的各目标时间点上的序列特征值的数值个数也为5个,假设系统设定的第二阈值是6,则数值个数小于第二阈值,此时需要补充新的用电量时间序列,系统从预设样本数据库中导入新的用电量时间序列。根据第二阈值和数值个数,系统导入的新的用电量时间序列的样本数量为1个,对新的用电量时间序列进行数据采集,再统计各个目标时间点上序列特征值的数值个数为6,最后进行数值个数比对。最后数值个数(6)大于或等于6(第二阈值),则本实施例执行结束。
参照图3,图3是本申请实施例方法涉及的硬件运行环境的设备结构示意图。本申请实施例终端可以是PC,便携计算机等终端设备。如图3所示,该异常检测终端可以包括:处理器1001,例如CPU,存储器1005,通信总线1002。其中,通信总线1002用于实现处理器1001和存储器1005之间的连接通信。存储器1005可以是高速RAM存储器,也可以是稳定的存储器(non-volatile memory),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
可选地,该异常检测终端还可以包括用户接口、网络接口、摄像头、RF(Radio Frequency,射频)电路,传感器、音频电路、WiFi模块等等。用户接口可以包括显示屏(Display)、输入单元比如键盘(Keyboard),可选用户接口还可以包括标准的有线接口、无线接口。网络接口可选的可以包括标准的有线接口、无线接口(如WI-FI接口)。本领域技术人员可以理解,图3中示出的异常检测终端结构并不构成对异常检测终端的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图3所示,作为一种计算机存储介质(计算机存储介质可以为非易失性存储介质)的存储器1005中可以包括操作系统、网络通信 模块以及异常检测可读指令。操作系统是管理和控制异常检测终端硬件和软件资源的可读指令,支持异常检测可读指令以及其它软件和/或可读指令的运行。网络通信模块用于实现存储器1005内部各组件之间的通信,以及与异常检测终端中其它硬件和软件之间通信。
在图3所示的异常检测终端中,处理器1001用于执行存储器1005中存储的异常检测可读指令,实现以下步骤:
按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
按时间顺序将所述所有平均特征值生成对应的平均特征序列;
根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
本申请异常检测终端的具体实施方式与上述用电量异常检测方法各实施例基本相同,在此不再赘述。
参照图2,本申请提供了一种异常检测装置,所述异常检测装置包括:
采集模块,用于按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
第一计算模块,用于对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
第一生成模块,用于按时间顺序将所述所有平均特征值生成对应的平均特征序列;
第二计算模块,用于根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
设置模块,用于若所述相对熵大于预设值,则将该相对熵所述对 应的特征数据序列设为异常序列。
进一步地,所述第二计算模块包括:
可选地,所述第二计算模块包括:
提取单元,用于提取所述每个特征数据序列中数值一致的目标序列特征值,并提取所述平均特征序列中数值一致的目标平均特征值;
统计单元,用于统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
计算单元将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
Figure PCTCN2018102145-appb-000003
进一步地,所述异常检测装置还包括:
第二生成模块,用于将所述平均特征序列和所述异常序列分别生成对应的平均分布曲线和异常分布曲线;
第一显示模块,用于将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别。
进一步地,所述异常检测装置还包括:
计算模块,用于基于所述异常分布曲线上的所有序列特征值,以及所述平均分布曲线上对应目标时间点上的平均特征值进行计算,获得偏移差值;
第二显示模块,用于若检测到基于所述异常分布曲线的触发操作,将该异常分布曲线的相对熵以及所述各目标时间点的偏移差值显示在预设坐标系中,以供用户分析。
进一步地,所述异常检测装置还包括:
接收模块,用于当接收到异常控制比例时,从预设映射表中获取与所述异常控制比例相映射的目标预设值;
调整模块,用于将当前默认的预设值调整为所述目标预设值,以调整所述异常序列的判断标准。
进一步地,所述异常检测装置还包括:
统计模块,用于统计当前所述所有特征数据序列的序列个数;
第一判断模块,用于若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值。
进一步地,所述异常检测装置还包括:
导入模块,用于若所有特征数据序列中任一目标时间点上的序列特征值的数值个数小于第二阈值,则从所述预设样本数据库中导入新的用电量时间序列;
第二判断模块,用于循环执行根据新的用电量时间序列获取新的特征数据序列,并统计新的数值个数的步骤,直至新的数值个数不小于第二阈值。
本申请还提供了一种计算机可读存储介质,所述计算机可读存储介质存储有一个或者一个以上可读指令,所述一个或者一个以上可读指令还可被一个或者一个以上的处理器执行以用于实现上述任一项所述的用电量异常检测方法的步骤。
本申请可读存储介质具体实施方式与上述用电量异常检测方法各实施例基本相同,在此不再赘述。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种用电量异常检测方法,其特征在于,所述用电量异常检测方法包括:
    按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
    对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
    按时间顺序将所述所有平均特征值生成对应的平均特征序列;
    根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
    若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
  2. 如权利要求1所述的用电量异常检测方法,其特征在于,所述根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵的步骤包括:
    提取所述每个特征数据序列中数值一致的目标序列特征值,并提取所述平均特征序列中数值一致的目标平均特征值;
    统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
    将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
    Figure PCTCN2018102145-appb-100001
  3. 如权利要求1所述的用电量异常检测方法,其特征在于,所述若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列的步骤之后还包括:
    将所述平均特征序列和所述异常序列分别生成对应的平均分布曲线和异常分布曲线;
    将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别。
  4. 如权利要求3所述的用电量异常检测方法,其特征在于,所述将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别的步骤之后还包括:
    基于所述异常分布曲线上的所有序列特征值,以及所述平均分布曲线上对应目标时间点上的平均特征值进行计算,获得偏移差值;
    若检测到基于所述异常分布曲线的触发操作,将该异常分布曲线的相对熵以及所述各目标时间点的偏移差值显示在预设坐标系中,以供用户分析。
  5. 如权利要求1所述的用电量异常检测方法,其特征在于,所述方法还包括:
    当接收到异常控制比例时,从预设映射表中获取与所述异常控制比例相映射的目标预设值;
    将当前默认的预设值调整为所述目标预设值,以调整所述异常序列的判断标准。
  6. 如权利要求1所述的用电量异常检测方法,其特征在于,所述按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列的步骤之后还包括:
    统计当前所述所有特征数据序列的序列个数;
    若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值。
  7. 如权利要求6所述的用电量异常检测方法,其特征在于,
    所述若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值的步骤之后还包括:
    若所有特征数据序列中任一目标时间点上的序列特征值的数值 个数小于第二阈值,则从所述预设样本数据库中导入新的用电量时间序列;
    循环执行根据新的用电量时间序列获取新的特征数据序列,并统计新的数值个数的步骤,直至新的数值个数不小于第二阈值。
  8. 一种异常检测装置,其特征在于,所述异常检测装置包括:
    采集模块,用于按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
    第一计算模块,用于对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
    第一生成模块,用于按时间顺序将所述所有平均特征值生成对应的平均特征序列;
    第二计算模块,用于根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
    设置模块,用于若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
  9. 如权利要求8所述的异常检测装置,其特征在于,所述第二计算模块包括:
    提取单元,用于提取所述每个特征数据序列中数值一致的目标序列特征值,并提取所述平均特征序列中数值一致的目标平均特征值;
    统计单元,用于统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
    计算单元将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
    Figure PCTCN2018102145-appb-100002
  10. 如权利要求8所述的异常检测装置,其特征在于,所述异常检测装置还包括:
    第二生成模块,用于将所述平均特征序列和所述异常序列分别生成对应的平均分布曲线和异常分布曲线;
    第一显示模块,用于将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别。
  11. 如权利要求10所述的异常检测装置,其特征在于,所述异常检测装置还包括:
    计算模块,用于基于所述异常分布曲线上的所有序列特征值,以及所述平均分布曲线上对应目标时间点上的平均特征值进行计算,获得偏移差值;
    第二显示模块,用于若检测到基于所述异常分布曲线的触发操作,将该异常分布曲线的相对熵以及所述各目标时间点的偏移差值显示在预设坐标系中,以供用户分析。
  12. 如权利要求8所述的异常检测装置,其特征在于,所述异常检测装置还包括:
    接收模块,用于当接收到异常控制比例时,从预设映射表中获取与所述异常控制比例相映射的目标预设值;
    调整模块,用于将当前默认的预设值调整为所述目标预设值,以调整所述异常序列的判断标准。
  13. 如权利要求8所述的异常检测装置,其特征在于,所述异常检测装置还包括:
    统计模块,用于统计当前所述所有特征数据序列的序列个数;
    第一判断模块,用于若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值。
  14. 如权利要求13所述的异常检测装置,其特征在于,所述异常检测装置还包括:
    导入模块,用于若所有特征数据序列中任一目标时间点上的序列特征值的数值个数小于第二阈值,则从所述预设样本数据库中导入新的用电量时间序列;
    第二判断模块,用于循环执行根据新的用电量时间序列获取新的特征数据序列,并统计新的数值个数的步骤,直至新的数值个数不小于第二阈值。
  15. 一种异常检测终端,其特征在于,所述异常检测终端包括:存储器、处理器、通信总线以及存储在所述存储器上的异常检测可读指令,所述处理器用于执行所述异常检测可读指令,实现如下的步骤:
    按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
    对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
    按时间顺序将所述所有平均特征值生成对应的平均特征序列;
    根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
    若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
  16. 如权利要求15所述的异常检测终端,其特征在于,所述根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵的步骤包括:
    提取所述每个特征数据序列中数值一致的目标序列特征值,并提取所述平均特征序列中数值一致的目标平均特征值;
    统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
    将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
    Figure PCTCN2018102145-appb-100003
  17. 如权利要求15所述的异常检测终端,其特征在于,所述处理器用于执行所述异常检测可读指令,实现如下的步骤:
    将所述平均特征序列和所述异常序列分别生成对应的平均分布 曲线和异常分布曲线;
    将所述平均分布曲线和所述异常分布曲线分别显示在预设坐标系中,以供用户识别。
  18. 如权利要求15所述的异常检测终端,其特征在于,所述处理器用于执行所述异常检测可读指令,实现如下的步骤:
    统计当前所述所有特征数据序列的序列个数;
    若所述序列个数小于第一阈值,则循环从预设样本数据库中导入新的用电量时间序列,并根据所述新的用电量时间序列获取到新的特征数据序列,统计新的序列个数,直至新的序列个数不小于第一阈值;
    若所有特征数据序列中任一目标时间点上的序列特征值的数值个数小于第二阈值,则从所述预设样本数据库中导入新的用电量时间序列;
    循环执行根据新的用电量时间序列获取新的特征数据序列,并统计新的数值个数的步骤,直至新的数值个数不小于第二阈值。
  19. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有异常检测可读指令,所述异常检测可读指令被处理器执行时实现如以下的步骤:
    按预设时间间隔从所有用电量时间序列中采集各目标时间点上的序列特征值,以生成所述各用电量时间序列对应的特征数据序列;
    对所述所有特征数据序列中各目标时间点对应的序列特征值作均值计算,以获得所述各目标时间点上的平均特征值;
    按时间顺序将所述所有平均特征值生成对应的平均特征序列;
    根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵;
    若所述相对熵大于预设值,则将该相对熵所述对应的特征数据序列设为异常序列。
  20. 如权利要求19所述的计算机可读存储介质,其特征在于,所述根据相对熵算法计算每个特征数据序列和平均特征序列之间的相对熵的步骤包括:
    提取所述每个特征数据序列中数值一致的目标序列特征值,并提 取所述平均特征序列中数值一致的目标平均特征值;
    统计所述目标序列特征值在所述对应的特征数据序列中的第一概率集合,设为Pi,并统计所述目标平均特征值在所述平均特征序列中的第二概率集合,设为Qi;
    将Pi和Qi代入如下相对熵算法公式进行计算,以获得所述每个特征数据序列与所述平均特征序列之间的相对熵D kl
    Figure PCTCN2018102145-appb-100004
PCT/CN2018/102145 2018-07-06 2018-08-24 用电量异常检测方法、装置、终端及计算机可读存储介质 Ceased WO2020006841A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810744182.1 2018-07-06
CN201810744182.1A CN109726364B (zh) 2018-07-06 2018-07-06 用电量异常检测方法、装置、终端及计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2020006841A1 true WO2020006841A1 (zh) 2020-01-09

Family

ID=66293843

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/102145 Ceased WO2020006841A1 (zh) 2018-07-06 2018-08-24 用电量异常检测方法、装置、终端及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN109726364B (zh)
WO (1) WO2020006841A1 (zh)

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113312239A (zh) * 2021-06-11 2021-08-27 腾讯云计算(北京)有限责任公司 一种数据检测方法、装置、电子设备及介质
CN113917263A (zh) * 2021-10-09 2022-01-11 乐创达投资(广东)有限公司 一种基于预测数据的异常能耗二次监测的方法及系统
CN114167246A (zh) * 2021-10-20 2022-03-11 河南省计量科学研究院 防雷元件测试仪检定参数电压上升速率的测量方法及评估
CN114840370A (zh) * 2022-05-05 2022-08-02 深圳市广和通无线通信软件有限公司 通用串行总线测试方法、无线通信模组、设备及存储介质
CN114936675A (zh) * 2022-04-26 2022-08-23 国能朔黄铁路发展有限责任公司 一种故障预警方法、装置、存储介质及电子设备
CN115273207A (zh) * 2022-08-23 2022-11-01 北京远鉴信息技术有限公司 一种人脸图像质量的评估方法、装置及电子设备
CN116992388A (zh) * 2023-09-25 2023-11-03 广东德怡电子科技有限公司 基于数据分析的薄膜开关测试数据处理方法
CN117609881A (zh) * 2023-11-29 2024-02-27 阿童木(广州)智能科技有限公司 一种基于人工智能的金属重叠检测方法及系统
CN118229112A (zh) * 2024-04-07 2024-06-21 南京审计大学 一种基于人工智能的审计平台监管系统及方法
CN118337222A (zh) * 2024-06-14 2024-07-12 山东德源电力科技股份有限公司 一种具备智慧存储功能的集中器
CN118465413A (zh) * 2024-07-09 2024-08-09 江苏城工建设科技有限公司 一种用电器耗电异常检测装置和方法
CN118690545A (zh) * 2024-06-05 2024-09-24 河北耀讯新能源科技有限公司 一种全钒液流电池多物理场仿真数据的智能分析方法
CN118841107A (zh) * 2024-07-26 2024-10-25 东莞市创之润新材料有限公司 一种用于聚丙烯生产质量异常检测的数据优化方法
CN120238361A (zh) * 2025-04-17 2025-07-01 北京宇天恒瑞科技发展有限公司 一种网络攻击安全检测方法
CN120492817A (zh) * 2025-07-16 2025-08-15 长春工程学院 用于化工生产中的化工数据处理分析方法及系统
WO2026007436A1 (zh) * 2024-07-04 2026-01-08 广东电网有限责任公司 关于风光储能的用电特征数据分析方法、装置、系统、设备及介质

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113762685A (zh) * 2020-12-28 2021-12-07 京东城市(北京)数字科技有限公司 异常用户的检测方法、装置及电子设备
CN112765826B (zh) * 2021-01-27 2022-05-06 长沙理工大学 基于用电功率频率分布相对熵的室内大麻种植识别方法
CN114240102A (zh) * 2021-12-03 2022-03-25 广东电网有限责任公司江门供电局 一种线损异常数据识别方法、装置、电子设备及存储介质
CN114119613B (zh) * 2022-01-26 2022-04-22 山东慧丰花生食品股份有限公司 基于图像处理的花生选种方法
CN116933986B (zh) * 2023-09-19 2024-01-23 国网湖北省电力有限公司信息通信公司 一种基于深度学习的电力数据安全管理系统
CN119939127B (zh) * 2025-04-08 2025-06-20 江阴众和电力仪表有限公司 一种基于电能表的数据采集方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101645884A (zh) * 2009-08-26 2010-02-10 西安理工大学 基于相对熵理论的多测度网络异常检测方法
CN103441982A (zh) * 2013-06-24 2013-12-11 杭州师范大学 一种基于相对熵的入侵报警分析方法
CN103514398A (zh) * 2013-10-18 2014-01-15 中国科学院信息工程研究所 一种实时在线日志检测方法及系统
CN106357434A (zh) * 2016-08-30 2017-01-25 国家电网公司 一种基于熵分析的智能电网通信网络流量异常检测方法

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5186322B2 (ja) * 2008-09-26 2013-04-17 インターナショナル・ビジネス・マシーンズ・コーポレーション 時系列データ解析システム、方法及びプログラム
US8874763B2 (en) * 2010-11-05 2014-10-28 At&T Intellectual Property I, L.P. Methods, devices and computer program products for actionable alerting of malevolent network addresses based on generalized traffic anomaly analysis of IP address aggregates
CN102271091B (zh) * 2011-09-06 2013-09-25 电子科技大学 一种网络异常事件分类方法
US8688620B2 (en) * 2011-09-23 2014-04-01 Hewlett-Packard Development Company, L.P. Anomaly detection in data centers
US9141914B2 (en) * 2011-10-31 2015-09-22 Hewlett-Packard Development Company, L.P. System and method for ranking anomalies
CN103281293A (zh) * 2013-03-22 2013-09-04 南京江宁台湾农民创业园发展有限公司 一种基于多维分层相对熵的网络流量异常检测方法
CN103823980A (zh) * 2014-02-26 2014-05-28 浙江水利水电学院 一种基于熵理论的大坝监控异常数据的检测方法
CN105630885B (zh) * 2015-12-18 2019-05-28 国网福建省电力有限公司泉州供电公司 一种用电异常检测方法及系统
CN106101121B (zh) * 2016-06-30 2019-01-22 中国人民解放军防空兵学院 一种全网络流量异常抽取方法
CN107231348B (zh) * 2017-05-17 2020-07-28 桂林电子科技大学 一种基于相对熵理论的网络流量异常检测方法
CN107908555B (zh) * 2017-11-07 2020-01-14 中国平安人寿保险股份有限公司 一种sql脚本的异常检测方法及其终端

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101645884A (zh) * 2009-08-26 2010-02-10 西安理工大学 基于相对熵理论的多测度网络异常检测方法
CN103441982A (zh) * 2013-06-24 2013-12-11 杭州师范大学 一种基于相对熵的入侵报警分析方法
CN103514398A (zh) * 2013-10-18 2014-01-15 中国科学院信息工程研究所 一种实时在线日志检测方法及系统
CN106357434A (zh) * 2016-08-30 2017-01-25 国家电网公司 一种基于熵分析的智能电网通信网络流量异常检测方法

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113312239B (zh) * 2021-06-11 2024-03-15 腾讯云计算(北京)有限责任公司 一种数据检测方法、装置、电子设备及介质
CN113312239A (zh) * 2021-06-11 2021-08-27 腾讯云计算(北京)有限责任公司 一种数据检测方法、装置、电子设备及介质
CN113917263A (zh) * 2021-10-09 2022-01-11 乐创达投资(广东)有限公司 一种基于预测数据的异常能耗二次监测的方法及系统
CN113917263B (zh) * 2021-10-09 2022-05-24 乐创达投资(广东)有限公司 一种基于预测数据的异常能耗二次监测的方法及系统
CN114167246A (zh) * 2021-10-20 2022-03-11 河南省计量科学研究院 防雷元件测试仪检定参数电压上升速率的测量方法及评估
CN114167246B (zh) * 2021-10-20 2024-04-05 河南省计量测试科学研究院 防雷元件测试仪检定参数电压上升速率的测量方法及评估
CN114936675A (zh) * 2022-04-26 2022-08-23 国能朔黄铁路发展有限责任公司 一种故障预警方法、装置、存储介质及电子设备
CN114840370A (zh) * 2022-05-05 2022-08-02 深圳市广和通无线通信软件有限公司 通用串行总线测试方法、无线通信模组、设备及存储介质
CN115273207B (zh) * 2022-08-23 2025-11-07 北京远鉴信息技术有限公司 一种人脸图像质量的评估方法、装置及电子设备
CN115273207A (zh) * 2022-08-23 2022-11-01 北京远鉴信息技术有限公司 一种人脸图像质量的评估方法、装置及电子设备
CN116992388B (zh) * 2023-09-25 2023-12-12 广东德怡电子科技有限公司 基于数据分析的薄膜开关测试数据处理方法
CN116992388A (zh) * 2023-09-25 2023-11-03 广东德怡电子科技有限公司 基于数据分析的薄膜开关测试数据处理方法
CN117609881A (zh) * 2023-11-29 2024-02-27 阿童木(广州)智能科技有限公司 一种基于人工智能的金属重叠检测方法及系统
CN117609881B (zh) * 2023-11-29 2024-04-30 阿童木(广州)智能科技有限公司 一种基于人工智能的金属重叠检测方法及系统
CN118229112A (zh) * 2024-04-07 2024-06-21 南京审计大学 一种基于人工智能的审计平台监管系统及方法
CN118690545A (zh) * 2024-06-05 2024-09-24 河北耀讯新能源科技有限公司 一种全钒液流电池多物理场仿真数据的智能分析方法
CN118337222A (zh) * 2024-06-14 2024-07-12 山东德源电力科技股份有限公司 一种具备智慧存储功能的集中器
CN118337222B (zh) * 2024-06-14 2024-08-30 山东德源电力科技股份有限公司 一种具备智慧存储功能的集中器
WO2026007436A1 (zh) * 2024-07-04 2026-01-08 广东电网有限责任公司 关于风光储能的用电特征数据分析方法、装置、系统、设备及介质
CN118465413A (zh) * 2024-07-09 2024-08-09 江苏城工建设科技有限公司 一种用电器耗电异常检测装置和方法
CN118841107A (zh) * 2024-07-26 2024-10-25 东莞市创之润新材料有限公司 一种用于聚丙烯生产质量异常检测的数据优化方法
CN120238361A (zh) * 2025-04-17 2025-07-01 北京宇天恒瑞科技发展有限公司 一种网络攻击安全检测方法
CN120492817A (zh) * 2025-07-16 2025-08-15 长春工程学院 用于化工生产中的化工数据处理分析方法及系统

Also Published As

Publication number Publication date
CN109726364A (zh) 2019-05-07
CN109726364B (zh) 2023-01-10

Similar Documents

Publication Publication Date Title
WO2020006841A1 (zh) 用电量异常检测方法、装置、终端及计算机可读存储介质
US11080906B2 (en) Method for creating period profile for time-series data with recurrent patterns
WO2021073114A1 (zh) 基于统计的异常流量监测方法、装置、设备及存储介质
WO2018103453A1 (zh) 检测网络的方法和装置
US20190114244A1 (en) Correlation-Based Analytic For Time-Series Data
CN110399347A (zh) 告警日志压缩方法、装置及系统、存储介质
CN117349220B (zh) 一种基于pci总线的数据处理方法及系统
US20180095454A1 (en) Pre-processor and diagnosis device
CN117034043B (zh) 基于多能源物联网的智慧建筑综合能耗监测方法及系统
CN112948614B (zh) 图像处理方法、装置、电子设备及存储介质
CN117827568A (zh) 测试设备稳定性的监控方法、装置、电子设备和存储介质
CN114721912B (zh) 一种数据分析方法、装置、设备及介质
CN115392812B (zh) 一种异常根因定位方法、装置、设备及介质
CN110837582A (zh) 数据关联方法、装置、电子设备和计算机可读存储介质
CN117977717B (zh) 一种寒地风光热储能综合能源协同管理方法及系统
CN119646723B (zh) 一种集中器运行数据管理方法
CN120046086B (zh) 一种基于大数据的化工安全生产管理方法及系统
CN114356989A (zh) 审计异常数据检测方法及装置
CN118413437A (zh) 一种满足端到端安全的远程运维诊断系统及方法
JP7481976B2 (ja) 異常スコア算出装置、異常スコア算出方法およびプログラム
CN120596955B (zh) 一种低压电流互感器在线检测方法及系统
CN121479628A (zh) 基于多维用电特征的电能表分时负荷监测方法及系统
CN115729783A (zh) 故障风险监控方法、设备、存储介质及程序产品
CN114416418B (zh) 数据检测方法、装置、电子设备和存储介质
WO2018122889A1 (ja) 異常検出方法、システムおよびプログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18925584

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18925584

Country of ref document: EP

Kind code of ref document: A1