A power grid power supply load outlier detection method and device

Abstract

The present application relates to the technical field of power supply detection, and particularly relates to a power supply load abnormal value detection method and device, comprising obtaining power supply load data, fitting the power supply load data into lower layer Gaussian distribution, then training a multi-level Gaussian distribution mixed optimization model, obtaining optimal weight, performing aggregation operation on the lower layer Gaussian distribution with the optimal weight to obtain upper layer Gaussian distribution, generating a reliability rating of the current processing power supply load time point data, determining whether the power supply load time point data is abnormal based on the reliability rating, providing decision support for power grid managers, and then judging the abnormal value; the present application can effectively identify the abnormal value existing in the power supply load data, so as to avoid possible economic or property loss.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid power supply detection, and particularly relates to a power grid power supply load abnormal value detection method and device. BACKGROUND

[0002] The operation of modern society cannot be separated from power supply, and the abnormality of power supply may cause economic loss and even personal safety damage. Predicting the power supply load of the power grid helps to reduce power safety hazards and carbon emissions. At present, many artificial intelligence methods for predicting power supply load have appeared. These methods use artificial intelligence technology to perform data mining on historical data of the power supply load of the power grid, and obtain a prediction value for reference by power grid managers.

[0003] However, due to factors such as possible sensor transient faults, external malicious intervention, staff tampering, and operator misoperation, the original data required for data mining by the artificial intelligence method may be abnormal, which may cause the results generated by the machine learning model constructed based on the artificial intelligence method to mislead the power grid managers, thereby causing economic loss or safety hazards. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a power grid power supply load abnormal value detection method and device to solve the problem that data may be abnormal without being detected.

[0005] To achieve the above purpose, the present application provides a power grid power supply load abnormal value detection method, comprising:

[0006] obtaining an abnormality detection data set;

[0007] extracting a power supply load time point data from the abnormality detection data set;

[0008] sending the power supply load time point data to a power grid power supply load database, and obtaining power grid power supply load data S all from the power grid power supply load database through an element generation function;

[0009] fitting the power grid power supply load data S all into a lower layer Gaussian distribution;

[0010] taking the lower layer Gaussian distribution as an input value to train a multi-level Gaussian distribution mixed optimization model, and obtaining an optimal weight;

[0011] performing an aggregation operation on the lower layer Gaussian distribution with the optimal weight to obtain an upper layer Gaussian distribution, and obtaining a mean value and a standard deviation of a random variable subject to the upper layer Gaussian distribution through the upper layer Gaussian distribution;

[0012] The credibility rating of the current processing power supply load time point data is generated by the mean and standard deviation of the random variable subject to the upper Gaussian distribution, and whether the power supply load time point data is abnormal is determined based on the credibility rating.

[0013] Optionally, the power supply load time point data is sent to the power grid power supply load database, and the power grid power supply load data S all is obtained from the power grid power supply load database by the element generation function.

[0014] The maximum confidence interval Gap max is obtained.

[0015] The power supply load time point data is sent to the power grid power supply load database, and the record time T current , the maximum confidence interval Gap max and the record time T current are input into the element generation function to generate an element generation value.

[0016] The DataSet(·) function is called, and the element generation value is input to obtain the power grid power supply load data S all .

[0017] Optionally, the element generation value:

[0018]

[0019] Wherein, T current , Type and Gap max are the record time, the set generation type and the maximum confidence interval respectively; U Type is the universal set composed of all possible values of the set generation type Type, and the universal set U Type includes five strings of 'Year', 'Mouth', 'Day', 'Hour', 'Minute'; S(·) is the element generation function; i is the first access pointer; getData(·) is the power supply load acquisition function, which is used to obtain the power grid power supply load value load current of the corresponding time before i Type time intervals at the current time T i .

[0020] Optionally, the power grid power supply load data S all :

[0021]

[0022] Wherein, S all is the power grid power supply load data; Type is the set generation type, and U TypeA universal set U Type including 'Year', 'Mouth', 'Day', 'Hour', 'Minute'; the DataSet(·) function is used to generate an element of the power supply load data S all for the specified set generation type; S(·) is an element generation function.

[0023] Optionally, the power supply load data S all is a set containing multiple elements, and the power supply load data S all is fitted to the lower-level Gaussian distribution, including:

[0024] An unvisited element in the power supply load data S all is read and fitted to obtain an independent Gaussian distribution;

[0025] Multiple sets of independent Gaussian distributions obtained by fitting are saved, and all independent Gaussian distributions are lower-level Gaussian distributions.

[0026] Optionally, the lower-level Gaussian distribution includes:

[0027]

[0028] wherein f(load, Type) is the lower-level Gaussian distribution; load is the power supply load value; Type is the set generation type, U Type is a universal set of all possible values of the set generation type Type, and the universal set U Type includes 'Year', 'Mouth', 'Day', 'Hour', 'Minute'; exp(·) is the exponential function of the natural base e; μ(Type) and σ(Type) are the mean and standard deviation of the random variable subject to the lower-level Gaussian distribution; length(·) is an element quantity calculation function; j is a second access pointer; DataSet(Type) is an element of the power supply load data S all constructed according to the set generation type Type.

[0029] Optionally, the lower-level Gaussian distribution is taken as an input value to train a multi-level Gaussian distribution mixture optimization model to obtain optimal weights, including:

[0030] The population is randomly initialized, and the initial Gaussian interval error of each individual in the initial population is calculated with the target Gaussian interval error as the evaluation index;

[0031] The differential evolution algorithm is invoked to optimize the weight distribution of the lower-level Gaussian distribution, obtain the optimized population, and calculate the optimized Gaussian interval error of each individual in the optimized population.

[0032] If the difference between the target Gaussian interval error and the initial target Gaussian interval error is less than a threshold or the maximum number of iterations is reached, the calculation stops, and the individual obtained by the last differential evolution algorithm is the optimal individual I. best Otherwise, replace the initial population with the optimized population and return to the step of calculating the initial target Gaussian interval error for each individual in the initial population, using the target Gaussian interval error as the evaluation index.

[0033] For the optimal individual I best Decoding yields the optimal weights of the lower-level Gaussian distribution;

[0034] The optimal individual I best :

[0035]

[0036] Among them, I best For the optimal individual, w best ('Year'), w best ('Mouth'), w best ('Day'), w best ('Hour') and w best ('Minute') represents the optimal weights for the five independent Gaussian distributions f(load,'Year'), f(load,'Mouth'), f(load,'Day'), f(load,'Hour'), ​​and f(load,'Minute'), respectively.

[0037] Optionally, the upper-layer Gaussian distribution is:

[0038]

[0039] Where F(load) is the upper-level Gaussian distribution; Type is the set generation type, U Type The set U is the universal set consisting of all possible values ​​of type Type. Type Includes five strings: 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'; exp(·) is the exponential function of the natural base e; π is the value of pi; μ(Type) and σ(Type) are the mean and standard deviation of a random variable following a lower-level Gaussian distribution, respectively; μ and σ are the mean and standard deviation of a random variable following a higher-level Gaussian distribution, respectively; w best (Type) represents the optimal weights of the lower-level Gaussian distribution; wbest ('Year'), w best ('Mouth'), w best ('Day'), w best ('Hour') and w best ('Minute') are the optimal weights of five groups of independent Gaussian distributions f(load,'Year'), f(load,'Mouth'), f(load,'Day'), f(load,'Hour') and f(load,'Minute') respectively; f(load,Type) is the lower-level Gaussian distribution.

[0040] Optionally, the credibility rating is:

[0041]

[0042] wherein, {load,T current} is the current processing power supply load point data, including power supply load value load and its recording time T current ; Level({load,T current}) is the credibility rating of the current processing power supply load point data; the value of the credibility rating Level({load,T current}) is 'high','middle', 'low' or 'abnormal', which respectively represents the confidence degree of the current processing power supply load point data as high credibility, medium credibility, low credibility or significant abnormality; μ and σ are respectively the mean and standard deviation of the random variable subject to the upper-level Gaussian distribution.

[0043] A power supply load outlier detection device,

[0044] An acquisition module acquires an anomaly detection data set;

[0045] An extraction module extracts a power supply load point data from the anomaly detection data set;

[0046] A processing module sends the power supply load point data to a power supply load database, and acquires power supply load data S all from the power supply load database through an element generation function;

[0047] A fitting module fits the power supply load data S all into a lower-level Gaussian distribution;

[0048] A training module takes the lower-level Gaussian distribution as an input value to train a multi-level Gaussian distribution hybrid optimization model, and obtains optimal weights;

[0049] The polymeric Gaussian module performs a polymeric operation on the optimal weight to the lower layer Gaussian distribution to obtain an upper layer Gaussian distribution, and obtains the mean and standard deviation of the random variable subject to the upper layer Gaussian distribution through the upper layer Gaussian distribution;

[0050] The anomaly evaluation module generates the reliability rating of the power supply load time point data of the current processing through the mean and standard deviation of the random variable subject to the upper layer Gaussian distribution, and determines whether the power supply load time point data is abnormal based on the reliability rating.

[0051] The power supply load anomaly value detection method provided by the application fits the power supply load data into a plurality of independent Gaussian distributions according to different periods, so as to identify a plurality of periodic correlations existing in the power consumption level; the differential evolution algorithm is used to train the multi-level Gaussian distribution mixed optimization model, so that the upper layer Gaussian distribution obtained through the polymeric operation is more in line with the real power consumption situation; the reliability rating of the power supply load time point data is given according to the degree of fit between the value of the power supply load at a certain moment and the generated upper layer Gaussian distribution, so as to provide decision support for the power grid management personnel; the application can effectively identify the anomaly value existing in the power supply load data of the power grid, so as to avoid the possible economic or property loss. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only illustrate the embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0053] Figure 1 The flowchart of the embodiment of the application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages of the application more clear, the following will further illustrate the application in detail by combining with specific embodiments and referring to the drawings.

[0055] It should be noted that the technical terms or scientific terms used in the embodiments of the present application should be understood as the general meanings understood by those skilled in the art in the field of the present application, unless otherwise defined. The terms "first", "second", and the like used in the present application do not represent any order, number, or importance, but are only used to distinguish different components. The terms "include", "contain", and the like mean that the elements or objects before the terms cover the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and the like are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "upper", "lower", "left", "right", and the like are only used to represent relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships can also change accordingly.

[0056] As shown in Figure 1 A power grid power load outlier detection method, comprising the following steps:

[0057] S100: obtaining an anomaly detection data set;

[0058] S200: extracting a power load time point data from the anomaly detection data set;

[0059] S300: sending the power load time point data to a power grid power load database, and obtaining power grid power load data S all from the power grid power load database through an element generation function;

[0060] S400: fitting the power grid power load data S all into a lower layer Gaussian distribution;

[0061] S500: taking the lower layer Gaussian distribution as an input value to train a multi-level Gaussian distribution mixture optimization model to obtain an optimal weight;

[0062] S600: performing an aggregation operation on the lower layer Gaussian distribution with the optimal weight to obtain an upper layer Gaussian distribution, and obtaining a mean and a standard deviation of a random variable subject to the upper layer Gaussian distribution through the upper layer Gaussian distribution;

[0063] S700: generating a credibility rating of the power load time point data currently processed through the mean and the standard deviation of the random variable subject to the upper layer Gaussian distribution, and determining whether the power load time point data is abnormal based on the credibility rating.

[0064] At least one of the abnormality detection data sets in step S200 contains power supply load point data to be executed for abnormality detection operation, and the subsequent steps can be executed together for all power supply load point data to be executed for abnormality detection operation, and then the power supply load point data is detected together whether it is abnormal, or the detection of whether the power supply load point data is abnormal is performed separately.

[0065] The power supply load data S all According to different periods, a plurality of independent Gaussian distributions are fitted to identify a plurality of periodic correlations of the power consumption level; a differential evolution algorithm is used to train a multi-level Gaussian distribution mixed optimization model, so that the upper Gaussian distribution obtained through the aggregation operation is more consistent with the real power consumption; the fitting degree of the power supply load value of the power grid at a certain moment and the generated upper Gaussian distribution is used to give the credibility rating of the power supply load point data, and decision support is provided for the power grid managers; The present application can effectively identify the abnormal values in the power supply load data of the power grid, so as to avoid possible economic or property losses.

[0066] A power supply load abnormal value detection method, comprising the following steps:

[0067] S100: obtaining an abnormality detection data set;

[0068] S200: extracting a power supply load point data from the abnormality detection data set. The power supply load point data includes power supply load value load and its record time T current . The power supply load is the total load of the power plant to the power supply, which is equal to the power consumption load plus the line loss load at the same time. The obtained power supply load point data is input as an input value into step S300;

[0069] The power supply load value load is a random variable subject to Gaussian distribution.

[0070] The record time T current , the current time value is represented in the following format:

[0071] T current =[Year(4),'-',Mouth(2),'-',Day(2),'-',Hour(2),'-',Minute(2)]

[0072] Wherein, T currentFor recording time, string format is used for storage. Year(·), Mouth(·), Day(·), Hour(·) and Minute(·) are current time year, month, day, hour and minute data acquisition functions, which will return a small Arabic numerals stored as a string type, the input value specifies the length of its return string, and 0 is added in front of the string when the number of digits is insufficient to meet the number of digits. In the above formula, the Year(4), Mouth(2), Day(2), Hour(2) and Minute(2) functions return the current time year, month, day, hour and minute data, respectively, and store these data in strings with lengths of 4, 2, 2, 2 and 2, respectively, and use the character '-' as an interval.

[0073] S300, send power supply load time point data to power grid power supply load database, obtain power grid power supply load data from power grid power supply load database through element generation function S all ; wherein the power grid power supply load database stores the long-term power grid power supply load data of a city, which is stored every minute in units of minutes.

[0074] The method comprises the following steps:

[0075] S301: obtain the variable maximum confidence interval Gap input by the user from the user interface max , the maximum confidence interval Gap max is obtained by referring to the experience of the staff.

[0076] S302: send power supply load time point data to power grid power supply load database, extract the recording time T in the power supply load time point data current , input the maximum confidence interval Gap max and the recording time T current into the element generation function to generate an element generation value:

[0077] The element generation value:

[0078]

[0079] Wherein, T current , Type and Gap max are the recording time, the set generation type and the maximum confidence interval, respectively. The set generation type Type is stored in the form of a string, and U Type is the universal set composed of all possible values of the set generation type Type, and the universal set U TypeThis includes five strings: 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'. S(·) is an element generation function used to generate an array containing at most the specified gaps, based on the specified set generation type and maximum confidence interval. max A set of n elements. i is the first access pointer, whose possible values ​​range from 1 to Gap. max The range includes all positive integers, including endpoint values. `getData()` is the power load acquisition function, used to obtain the current time T. current The grid power supply load value at the corresponding time before the i-th Type time interval. i .

[0080] S303: Call the DataSet() function, input the element values, and obtain the grid power supply load data S. all .

[0081]

[0082] Among them, S all The required grid power supply load data is a set containing 5 elements, with each element itself forming a separate set. Type specifies the set generation type, U... Type The set U is the universal set consisting of all possible values ​​of type Type. Type The dataset includes five strings: 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'. The DataSet() function is used to generate a data set of a specified type, constructing the power grid load data S. all One element. S(·) is the element-generating function, which takes three input values: T current Type and Gap max These are the recording time, set generation type, and maximum confidence interval, respectively. This function generates a set containing at most a gap, based on the specified set generation type and maximum confidence interval. max A collection of elements.

[0083] S400: Transfer grid power supply load data S all Each element in the data is independently fitted to an independent Gaussian distribution, and the power grid load data S all S is a set containing five elements. all Each element in the array is a separate element, and the number of elements does not exceed the maximum confidence interval Gap. max The set of data for power grid load S allextracts all the power grid supply load values load contained in each element and performs a fitting operation in the form of a Gaussian distribution, and calls the obtained five groups of independent Gaussian distributions as lower Gaussian distributions;

[0084] According to the following steps, the power grid supply load data S all is fitted into an independent Gaussian distribution respectively,

[0085] S401: The power grid supply load data S all is a set of five elements, and an unvisited element is found in the power grid supply load data S all and is input into step S402 as an input value. If there is no unvisited element in S all , go to step S403;

[0086] S402: Read an unvisited element in S all . Each element in the set S all is a set with an element number not exceeding the maximum confidence interval Gap max . For each element in the power grid supply load data S all , extract all the power grid supply load values load contained in each element and perform a fitting operation in the form of a Gaussian distribution according to the following formula:

[0087]

[0088] Wherein, f(load, Type) is a Gaussian distribution function obtained by fitting the historical power grid supply load sampling in the interval unit of Type, load is a random variable subject to Gaussian distribution, which means the power grid supply load value. Type is the set generation type, U Type is the universal set of all possible values of the set generation type Type, and the universal set U Type includes the five strings of ‘Year’, ‘Mouth’, ‘Day’, ‘Hour’, and ‘Minute’. exp(·) is the exponential function of natural base e, and when the input value is input, the function will return the calculation result of the power of natural base e. π is the value of the circular constant. μ(Type) and σ(Type) are respectively the mean and standard deviation of the random variable subject to lower Gaussian distribution obtained by fitting in the interval unit of Type. length(·) is a set element number calculation function, which is used to return the element number contained in the input set. j is the second access pointer. DataSet(Type) is the power grid supply load data S allone element. After completion, go to step S401;

[0089] S403: save the power grid power load data S all The parameters μ(Type) and σ(Type) of the five groups of independent Gaussian distribution obtained by fitting the five elements in the table are as follows: μ(Type) and σ(Type) are the mean and standard deviation of the random variable subjected to the lower Gaussian distribution, which are fitted according to the Type as the interval unit.

[0090] S500: take the lower Gaussian distribution as the input value to train the multi-level Gaussian distribution mixed optimization model to obtain the optimal weight. Take the minimization of the target Gaussian interval error as the optimization objective, and process the lower Gaussian distribution by using the standard differential evolution algorithm to train the multi-level Gaussian distribution mixed optimization model. Repeat the training operation until the maximum function evaluation number set is reached;

[0091] The specific steps are as follows:

[0092] S501: set the evaluation number FE to 0, set the maximum function evaluation number FE to 5000, and set the population size N to 200. Randomly initialize the population, and calculate the target Gaussian interval error of each individual in the population under the initial state with the target Gaussian interval error as the evaluation index; max

[0093] S502: call the standard differential evolution algorithm to perform one iteration on the weight distribution of the lower Gaussian distribution, and call the population before and after the iteration as the parent population and the child population. Calculate the target Gaussian interval error of each individual in the child population;

[0094] In the step S502, the target Gaussian interval error of one individual in the population is calculated according to the following steps:

[0095] S502-1: calculate the target Gaussian interval error according to the following formula:

[0096]

[0097] where σ diff ​For the target Gaussian interval error, std(·) is a standard deviation calculation function, which is used to calculate the standard deviation of all elements in the input set; diff1, diff2 and diff3 are the absolute values of the difference between the expected value and the true value of the sample number of the upper Gaussian distribution of the random variable load falling in the interval [μ-σ, μ+σ], [μ-2σ, μ+2σ] and [μ-3σ, μ+3σ] respectively; load represents the power supply load value; length(·) is a set element number calculation function, which is used to return the number of elements contained in the input set; diffSet1, diffSet2 and diffSet3 represent the set of all sample points falling in the interval [μ-σ, μ+σ], [μ-2σ, μ+2σ] and [μ-3σ, μ+3σ] respectively; CP represents the total number of sample points, and the default value is 30, U CP is the full set of all sample points, where the first sample point can be represented as {load, T current , where T current represents the recording time; then, the recording time T current is moved forward by 1 minute to obtain a new sample point, which is incorporated into the set U CP , and the operation is repeated until the number of elements in the set reaches U CP ; poi1, poi2 and poi3 are all pointer variables, which are used to record the order of the accessed elements in their respective sets; μ and σ are the mean and standard deviation of the random variable subject to the upper Gaussian distribution obtained by performing the aggregation operation;

[0098] S502-2: increase the value of the evaluation number FE by 1; return the value of the target Gaussian interval error σ diff obtained.

[0099] S503: repeat step S502 until the maximum number of function evaluations 5000 is reached. At this time, the iteration is ended, and an optimal individual I best obtained after optimization of the population Population is returned. The optimal weight distribution of the five independent Gaussian distributions is obtained by decoding the optimal individual I best .

[0100]

[0101] where I best is the optimal individual, w best ('Year') is the optimal weight of the year, w best ('Mouth') is the optimal weight of the month, w best ('Day') is the optimal weight of the day, w best ('Hour') is the optimal weight of the hour, and w bestw ('Year'), w ('Mouth'), w ('Day'), w ('Hour') and w ('Minute') are the optimal weights of the five groups of independent Gaussian distributions f(load,'Year'), f(load,'Mouth'), f(load,'Day'), f(load,'Hour') and f(load,'Minute'), respectively.

[0102] 600: performing an aggregation operation on the lower-level Gaussian distributions with the optimal weights to obtain an upper-level Gaussian distribution, and obtaining the mean and the standard deviation of a random variable subject to the upper-level Gaussian distribution through the upper-level Gaussian distribution;

[0103] The upper-level Gaussian distribution is:

[0104]

[0105] In the above formula, F(load) is the upper-level Gaussian distribution obtained by performing the aggregation operation, load is a random variable subject to a Gaussian distribution, and represents the power grid power supply load value; Type is the set generation type, U Type is a universal set composed of all possible values of the set generation type Type, and the universal set U Type includes the five strings 'Year', 'Mouth', 'Day', 'Hour' and 'Minute'; exp(·) is an exponential function of the natural base e, and when the input value is input, the function returns the calculation result of the input power of the natural base e; π is the value of the circular constant; μ(Type) and σ(Type) are respectively the mean and the standard deviation of a random variable subject to a lower-level Gaussian distribution fitted according to Type as the interval unit; μ and σ are respectively the mean and the standard deviation of a random variable subject to an upper-level Gaussian distribution obtained by performing the aggregation operation; w best (Type) is the optimal weight of the lower-level Gaussian distribution; w best ('Year'), w best ('Mouth'), w best ('Day'), w best ('Hour') and w best ('Minute') are the optimal weights of the five groups of independent Gaussian distributions f(load,'Year'), f(load,'Mouth'), f(load,'Day'), f(load,'Hour') and f(load,'Minute'), respectively; f(load,Type) is a Gaussian distribution function obtained by sampling the historical power grid power supply load according to Type as the interval unit and fitting.

[0106] S700: generating the credibility level of the current processing power supply load time point data through the mean and standard deviation of the random variable subject to the upper-layer Gaussian distribution, and determining whether the power supply load time point data is abnormal based on the credibility level;

[0107] The credibility level:

[0108]

[0109] wherein, {load, T current} is the current processing power supply load time point data, including the power grid power supply load value load and its recording time T current ; Level({load, T current}) is the credibility level of the current processing power supply load time point data, whose value is stored in the form of a string; the value of the credibility level Level({load, T current}) is 'high','middle', 'low' or 'abnormal', which respectively represents that the confidence degree of the current processing power supply load time point data is high, medium, low or significantly abnormal; μ and σ are respectively the mean and standard deviation of the random variable subject to the upper-layer Gaussian distribution obtained by performing the aggregation operation.

[0110] A power grid power supply load outlier detection device, characterized in that,

[0111] A power grid power supply load outlier detection device,

[0112] An acquisition module acquires an outlier detection data set;

[0113] An extraction module extracts a power supply load time point data from the outlier detection data set;

[0114] A processing module sends the power supply load time point data to a power grid power supply load database, and obtains power grid power supply load data S all from the power grid power supply load database through an element generation function;

[0115] A fitting module fits the power grid power supply load data S all into a lower-layer Gaussian distribution;

[0116] A training module takes the lower-layer Gaussian distribution as an input value to train a multi-level Gaussian distribution mixed optimization model, and obtains an optimal weight;

[0117] An aggregated Gaussian module performs an aggregation operation on the lower-layer Gaussian distribution to obtain an upper-layer Gaussian distribution through the optimal weight, and obtains the mean and standard deviation of the random variable subject to the upper-layer Gaussian distribution through the upper-layer Gaussian distribution;

[0118] The abnormality evaluation module generates a reliability rating of the power supply load point data currently processed by a mean value and a standard deviation of a random variable subject to an upper-layer Gaussian distribution, and determines whether the power supply load point data is abnormal based on the reliability rating.

[0119] It should be understood by those of ordinary skill in the art that the above discussion of any of the embodiments is merely exemplary and is not intended to suggest that the scope of the present application is limited to these examples; the above embodiments or technical features among different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of the present application as described above, which are not provided in details for the sake of brevity.

[0120] Embodiments of the present application are intended to cover all such alternatives, modifications, and variations as fall within the broad scope of the appended claims. Accordingly, any one of the claims, or any element thereof, can be further modified or substituted in any manner which is, or is deemed to be, within the principles of the application.

Claims

1. A method for detecting abnormal values ​​in power grid load, characterized in that, include: Obtain the anomaly detection dataset; Extract a point-in-time data point of power supply load from the anomaly detection dataset; The power supply load data at specific times is sent to the power grid load database, and the power grid load data S is obtained from the power grid load database using an element generation function. all : Data on power grid load S all Fit to a lower-level Gaussian distribution: The power supply load data at specific times is sent to the power grid load database, and the power grid load data S is obtained from the power grid load database using an element generation function. all ,include: Get the maximum confidence interval ; The power supply load time data is sent to the power grid power supply load database, and the recorded time T is extracted from the power supply load time data. current The maximum confidence interval and recording time T current The input is passed to the element generation function to generate element values; Element generated value: ； in, , and These are the recording time, set generation type, and maximum confidence interval, respectively. Generate type for collection The universal set consisting of all possible values ​​of , the universal set It includes five strings: 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'; S(·) is the element generation function; i is the first access pointer; getData(·) is the power load acquisition function, used to obtain the current time. i The grid power supply load value at the corresponding time before the time interval. i ; Call the DataSet() function, input the elements to generate values, and obtain the power grid load data S. all ; The power grid load data S all : ； Among them, S all Data on power grid load; For collection generation type, Generate type for collection The universal set consisting of all possible values ​​of , the universal set The data sets include five strings: 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'. The DataSet() function is used to generate a data set of a specified type and construct the power grid load data S. all S is an element; S(·) is the element generating function; The lower-level Gaussian distribution is used as input to train a multi-level Gaussian distribution mixture optimization model to obtain the optimal weights. The optimal weights are aggregated with the lower Gaussian distribution to obtain the upper Gaussian distribution. The mean and standard deviation of the random variable that follows the upper Gaussian distribution are obtained through the upper Gaussian distribution. The reliability rating of the current power load time point data is generated by the mean and standard deviation of the random variable that follows the upper-level Gaussian distribution, and the abnormality of the power load time point data is determined based on the reliability rating. Among them, the power grid load data S all It is a collection containing multiple elements, namely, the grid power supply load data S all The fit is a lower-level Gaussian distribution, including: Read power grid load data S all An unvisited element is fitted to obtain an independent Gaussian distribution; Save multiple sets of independent Gaussian distributions obtained by each fitting, and all independent Gaussian distributions are the lower-level Gaussian distributions; The lower-level Gaussian distribution includes: ； in, The lower-level Gaussian distribution is represented; load is the power grid load value. For collection generation type, Generate type for collection The universal set consisting of all possible values ​​of , the universal set It includes the five strings 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'; exp(·) is the exponential function of the natural base e; and , respectively, are the mean and standard deviation of the random variable following a lower-level Gaussian distribution; length(·) is the function to calculate the number of elements in the set; j is the second access pointer; To generate types by collection The constructed power grid load data S all One of the elements; Using the lower-level Gaussian distribution as input, a multi-level Gaussian distribution mixture optimization model is trained to obtain the optimal weights, including: The population is randomly initialized, and the initial Gaussian interval error of each individual in the initial population is calculated using the target Gaussian interval error as the evaluation index. The differential evolution algorithm is invoked to optimize the weight distribution of the lower-level Gaussian distribution, obtain the optimized population, and calculate the optimized Gaussian interval error of each individual in the optimized population. If the difference between the target Gaussian interval error and the initial target Gaussian interval error is less than a threshold or the maximum number of iterations is reached, the calculation stops, and the individual obtained by the last differential evolution algorithm is the optimal individual I. best Otherwise, replace the initial population with the optimized population and return to the step of calculating the initial target Gaussian interval error for each individual in the initial population, using the target Gaussian interval error as the evaluation index. For the optimal individual I best Decoding yields the optimal weights of the lower-level Gaussian distribution; The optimal individual I best : ； Among them, I best For the optimal individual, , and Five independent Gaussian distributions , , , and The optimal weight; The upper-layer Gaussian distribution is as follows: ； in, It is a Gaussian distribution in the upper layer; For collection generation type, Generate type for collection The universal set consisting of all possible values ​​of , the universal set It includes the five strings 'Year', 'Mouth', 'Day', 'Hour', and 'Minute'; exp(·) is the exponential function of the natural base e; This is the value of pi; and These are the mean and standard deviation of a random variable that follows a lower-level Gaussian distribution, respectively. and These are the mean and standard deviation of a random variable that follows a higher-level Gaussian distribution, respectively. The optimal weights for the lower-level Gaussian distribution; , and Five independent Gaussian distributions , , , and The optimal weight; It is a lower-level Gaussian distribution; Credibility rating: ； in, This refers to the current point-in-time data of the power supply load being processed, including the grid power supply load value and its recording time. ; Reliability rating for the current point-in-time power load data being processed; Reliability rating The value is or The time intervals represent the confidence level of the current power load data being processed: highly reliable, moderately reliable, low reliable, or significantly abnormal. and denoted as mean and standard deviation, respectively, of a random variable that follows a higher-order Gaussian distribution.

2. A power grid load anomaly detection device, applied to the power grid load anomaly detection method as described in claim 1, characterized in that, The module retrieves the anomaly detection dataset. The extraction module extracts a point-in-time data point of power supply load from the anomaly detection dataset; The processing module sends the power supply load data at specific times to the power grid load database, and retrieves the power grid load data S from the power grid load database using an element generation function. all ; The fitting module will use the grid power supply load data S all The fit is a lower-level Gaussian distribution; The training module takes the lower-level Gaussian distribution as input to train a multi-level Gaussian distribution mixture optimization model to obtain the optimal weights. The Aggregate Gaussian module performs an aggregation operation on the lower-level Gaussian distribution with the optimal weights to obtain the upper-level Gaussian distribution. The mean and standard deviation of the random variable that follows the upper-level Gaussian distribution are obtained through the upper-level Gaussian distribution. The anomaly assessment module generates a reliability rating for the current power load time point data by using the mean and standard deviation of random variables that follow a higher-level Gaussian distribution. Based on the reliability rating, it determines whether the power load time point data is abnormal.

Images