CN115687948A - Power special transformer user unsupervised classification method based on load curve - Google Patents
Power special transformer user unsupervised classification method based on load curve Download PDFInfo
- Publication number
- CN115687948A CN115687948A CN202211310800.4A CN202211310800A CN115687948A CN 115687948 A CN115687948 A CN 115687948A CN 202211310800 A CN202211310800 A CN 202211310800A CN 115687948 A CN115687948 A CN 115687948A
- Authority
- CN
- China
- Prior art keywords
- user
- type
- load curve
- users
- load
- 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.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a load curve-based unsupervised classification method for power specific transformer users. The method comprises the following steps: acquiring a user load curve; pre-classifying users according to the user load curve to determine user types, wherein the user types comprise a horizontal type, a zero-value regular type, a year cycle type, a day cycle type and a random type; and respectively carrying out secondary classification on the users in the annual period type and the users in the daily period type to obtain a target classification result. According to the method, the classification of the users of the special power transformer is realized based on the characteristics of the user load curve of the users of the special power transformer, the power utilization behavior habits of the users of the special power transformer are loaded, the practicability is high, the classification effect is good, and accurate classification can be realized.
Description
Technical Field
The application relates to the technical field of power consumer classification, in particular to a load curve-based unsupervised classification method for power specific transformer users.
Background
In the classification of the power specific transformer users or the subdivision task of the power specific transformer users, a supervised classification method is generally adopted, but in practical application, label data of the power users are often lacked, and the classification of the power specific transformer users cannot be realized by directly adopting the supervised classification method. In addition, in the power-dedicated transformer user segmentation task, clustering is generally performed according to information such as the power utilization rule, the power utilization characteristics, the industry characteristics and the like of users, but the users after actual segmentation have large differences, for example, the power utilization rules may be inconsistent, the power utilization characteristics have large differences and the like. In an actual service scene, the electricity utilization rules and the electricity utilization characteristics of power special transformer users in the same subdivision industry may have great difference, and the real purpose of classification or user subdivision cannot be achieved. If the users are marked manually, because the power-specific users may have different differences in different characteristics, it is also difficult to mark the types of the users accurately, and therefore, the supervision method is difficult to implement.
The existing unsupervised classification method for the power special transformer users is mainly based on a clustering algorithm or carries out user classification according to services, but various abnormal and special power special transformer users exist in practical application, and the problems of poor clustering effect, inexplicability of results and the like can be directly caused. When the method is used for the downstream tasks, the effect of the downstream tasks can be influenced, such as the load prediction task, and if the clustering effect is poor, the load pre-precision can be directly influenced. These techniques or methods ignore special power-specific or abnormal power-specific users, such as users with a constant electrical load, users with a long-term zero electrical load, users with a random electrical load, etc. These special and abnormal users will directly affect the final classification effect, and if all the power-dedicated users are clustered, the classification effect will be poor, and there is no interpretability in the service.
Therefore, the classification method for the power specific transformer users adopted in the prior art has the problems of low practicability, low fault tolerance rate and poor classification effect.
Disclosure of Invention
The embodiment of the application aims to provide a load curve-based unsupervised classification method for power specific transformer users, which is used for solving the problems of low practicability, low fault tolerance rate and poor classification effect of the classification method for the power specific transformer users adopted in the prior art.
In order to achieve the above object, a first aspect of the present application provides a load curve-based unsupervised classification method for power specific transformer users, the method including:
acquiring a user load curve;
pre-classifying users according to user load curves to determine user types, wherein the user types comprise a horizontal type, a zero-value regular type, a year period type, a day period type and a random type;
and respectively carrying out secondary classification on the annual cycle type users and the daily cycle type users to obtain a target classification result.
In the embodiment of the present application, obtaining the user load curve includes:
and acquiring the power load data of the user according to a preset period to form a user load curve.
In the embodiment of the present application, pre-classifying users according to user load curves to determine user types includes:
judging whether the user type is a horizontal type or not according to the user load curve;
judging whether the user type is a zero-value regular type or not under the condition that the user type is not a horizontal type;
judging whether the user type is a yearly periodic type or not under the condition that the user type is not a zero-value regular type;
judging whether the user type is a day period type or not under the condition that the user type is not a year period type;
and under the condition that the user type is not the day cycle type, judging that the user type is a random type.
In the embodiment of the present application, the determining whether the user type is a horizontal type according to the user load curve includes at least one of the following:
determining the mode of a user load curve, and judging the user type to be a horizontal type under the condition that the occupation ratio of the mode is not lower than a first threshold value; or
And determining the variation coefficient of the user load curve, and judging the user type to be a horizontal type under the condition that the variation coefficient is not higher than a second threshold value.
In the embodiment of the present application, determining whether the user type is a zero-value regular type includes:
determining an initial two-dimensional matrix according to a user load curve;
based on the Hamming distance, obtaining a similarity matrix according to the initial two-dimensional matrix;
determining the mean value of the similarity matrix;
and in the case that the average value of the similarity matrix is larger than a third threshold value, judging that the user type is a zero-value regular type.
In the embodiment of the present application, determining whether the user type is a year period type includes:
determining an initial two-dimensional matrix according to a user load curve;
filtering the initial two-dimensional matrix to obtain a first correlation coefficient matrix;
determining a mean value of the first correlation coefficient matrix;
judging whether the user load curve has periodicity according to the mean value of the first correlation coefficient matrix;
and judging the user type to be a yearly periodic type under the condition that the user load curve has the periodic type.
In the embodiment of the present application, determining whether the user type is a daily periodic type includes:
screening a load sequence of the latest first preset number of days in a user load curve;
determining a second phase relation number matrix according to the load sequence;
determining the mean value of the second correlation number matrix;
judging whether the user load curve has periodicity according to the mean value of the second correlation number matrix;
and under the condition that the user load curve has periodicity, judging that the user type is a day periodicity type.
In the embodiment of the present application, the performing secondary classification on the annual cycle type user and the daily cycle type user respectively to obtain the target classification result includes:
identifying an abnormal user load curve through a density clustering algorithm;
removing users corresponding to the abnormal user load curve to obtain target users;
and carrying out time series clustering processing on the user load curve of the target user to obtain a target classification result.
This application second aspect provides a special transformer user unsupervised sorter of electric power based on load curve, its characterized in that includes:
a memory configured to store instructions; and
a processor configured to call instructions from the memory and upon execution of the instructions, to enable the load curve based power specific user unsupervised classification method described above.
A third aspect of the present application provides a machine-readable storage medium, wherein the machine-readable storage medium has stored thereon instructions for causing a machine to execute the above-mentioned load curve-based power-specific user unsupervised classification method.
According to the technical scheme, a user load curve is obtained, users are pre-classified according to the user load curve to determine user types, the user types comprise a horizontal type, a zero-value regular type, a year-period type, a day-period type and a random type, and the year-period type users and the day-period type users are secondarily classified respectively to obtain target classification results. This application realizes categorisedly based on the characteristics of user load curve self of user, more accords with power consumption load user's power consumption behavior habit, can realize accurate classification, and categorised effectual and the practicality is higher.
Additional features and advantages of embodiments of the present application will be described in detail in the detailed description which follows.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the embodiments of the disclosure, but are not intended to limit the embodiments of the disclosure. In the drawings:
fig. 1 is a schematic flowchart of an unsupervised classification method for power specific transformer users based on a load curve according to an embodiment of the present application;
fig. 2 is a schematic flow chart of power specific transformer user pre-classification according to an embodiment of the present disclosure;
fig. 3 is a structural block diagram of an unsupervised classification device for a power specific transformer user based on a load curve according to an embodiment of the present application.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it should be understood that the specific embodiments described herein are only used for illustrating and explaining the embodiments of the present application and are not used for limiting the embodiments of the present application. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making creative efforts shall fall within the protection scope of the present application.
It should be noted that if directional indications (such as up, down, left, right, front, back, 8230; \8230;) are referred to in the embodiments of the present application, the directional indications are only used to explain the relative positional relationship between the components, the motion situation, etc. in a specific posture (as shown in the attached drawings), and if the specific posture is changed, the directional indications are correspondingly changed.
In addition, if there is a description of "first", "second", etc. in the embodiments of the present application, the description of "first", "second", etc. is for descriptive purposes only and is not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In addition, technical solutions between the embodiments may be combined with each other, but must be based on the realization of the technical solutions by a person skilled in the art, and when the technical solutions are contradictory to each other or cannot be realized, such a combination should not be considered to exist, and is not within the protection scope claimed in the present application.
In the existing unsupervised classification task of the special transformer users, classification is realized only by clustering, but the technologies or the methods ignore special power special transformer users or abnormal power special transformer users, such as users with unchanged power load, users with zero power load for a long time, random power load and the like. These special and abnormal users will directly affect the final classification effect, and if all the users of the power specific transformation are clustered, the classification effect will be poor, and there is no interpretability in the service. Therefore, identification and classification of specific power-specific customers is particularly important.
The classification of the power specific transformer users can take the upstream tasks of certain tasks and the classification results of the power specific transformer users as the input of the downstream tasks or the basis of model strategies, such as a load prediction task. General load prediction tasks are based on independent modeling of power specific transformer users, but when large batches of power specific transformer users are met, model training time cost, hardware cost, reasoning time cost and the like are greatly increased due to independent modeling of the users. Therefore, the number of models needs to be reduced while the prediction accuracy needs to be ensured, the classification of the power specific variable users can achieve the purpose, similar users with load curves are classified into the same class, and therefore downstream tasks are constructed for the class, and the number of models is reduced. Therefore, the embodiment of the application provides an unsupervised classification method for power specific transformer users based on load curves.
Fig. 1 is a schematic flowchart of an unsupervised classification method for power specific transformer users based on a load curve according to an embodiment of the present application. As shown in fig. 1, an embodiment of the present application provides a load curve-based unsupervised classification method for power specific transformer users, and the method may include the following steps.
In the embodiment of the application, a user load curve is a load curve of user electric quantity, the load curve can reflect the rule that the load changes along with time in a period of time, is the basis for scheduling electric power of an electric power system and planning the electric power system, and is divided into an active power load curve and a reactive power load curve according to the types of the loads; according to the time, the load curve is divided into a daily load curve and an annual load curve. In the embodiment of the application, unsupervised classification can be performed on the power special transformer users based on the user load curve, the user load curve is obtained firstly, and then the users are classified according to the user load curve. In general, the data of the user load curve is similar to a time sequence, and in one example, the data acquisition of the user load curve may adopt a manner of acquiring power load data according to a preset period, so as to form the user load curve. The load users are classified through the user load curves, the power utilization habits of the power load users are met, and accurate classification is facilitated.
And 102, pre-classifying the users according to the user load curves to determine user types, wherein the user types comprise a horizontal type, a zero-value regular type, a year period type, a day period type and a random type.
In the embodiment of the application, according to multiple experimental statistical analysis, the users can be divided into five types according to the user load curve, namely a horizontal type, a zero-value regular type, a year period type, a day period type and a random type. Wherein, the load of the horizontal type, that is, the load in the user load curve at any time point is basically the same, and the user type is basically kept unchanged. The zero-value rule is that the load exists in the load curve of the user only in a period of one year, and the load value of the user with the electrical load is only available in a part with the load close to the user, such as an agricultural user, in certain solar terms. The annual cycle type is an electrical load user whose load curve is in a cycle of year and the load trend on the same date in each year is very close. The daily cycle type is an electricity load user whose load curve has a certain periodicity by day or by week, and is, for example, a general industrial user, a commercial electricity user, or the like. The random type is a user of the electrical load which is not any one of a horizontal type, a zero-value regular type, a year cycle type and a day cycle type.
In one example, the power load users may be pre-classified according to the user load curves, and whether the user load curves of the power load users satisfy the determination conditions of a horizontal type, a zero-value regular type, a year period type, a day period type or a random type may be sequentially determined. Further, according to the pre-classification of the first stage, all the electric load users can be divided into five types respectively. Through multiple experimental statistical analysis, the classification mode is very practical and has strong practicability.
And 103, performing secondary classification on the annual cycle type users and the daily cycle type users respectively to obtain target classification results.
In the embodiment of the present application, after all the user load curves are classified and divided into five categories, because the users of the annual cycle type and the daily cycle type still have the problem of inconsistent cycle trend or cycle fluctuation, the user load curves of the users of the same annual cycle type or daily cycle type category may have a large difference, and thus the load prediction cannot be performed through the same model. Therefore, the second stage of classification is required for the electricity load users of the annual cycle type and the daily cycle type to realize the secondary classification and determine the final classification result, i.e., the target classification result.
In one example, by classifying users of the annual cycle type and users of the daily cycle type twice, the user load curves of the trends and periods that are not the same can be clustered, and the user load curves of the same shape can be classified into the same cluster, thereby realizing the classification of the load curves of the same period and fluctuation into the same category. In another example, after secondary classification is completed on the annual cycle type and daily cycle type users, a final classification result is obtained, and then a load prediction model can be constructed on the same class of power load users according to the final classification result, so that the power load users can be classified in the later period, and the classification efficiency is improved.
In the embodiment of the application, firstly, classification is realized based on the characteristics of the user load curves of the power-dedicated users, and compared with the existing method, the method is more in line with the power utilization behavior habits of the power-dedicated users, and the same power utilization behavior and the load curves with the same fluctuation and trend can be accurately divided together. Secondly, the embodiment of the application classifies the users of the electric loads based on an unsupervised mode, solves the problem that label data is lacked in a classification task, and is easier to apply in an actual scene. Meanwhile, in order to solve the problem of load prediction of a large number of users, after unsupervised classification, independent modeling is not needed for each user no matter different classified users are subjected to independent modeling or the classification result of the method is used as input to design the model, so that the number of the models can be greatly reduced, the training and reasoning cost is reduced, and the efficiency is improved.
According to the technical scheme, a user load curve is obtained, users are pre-classified according to the user load curve to determine user types, the user types comprise a horizontal type, a zero-value regular type, a year-period type, a day-period type and a random type, and the year-period type users and the day-period type users are secondarily classified respectively to obtain target classification results. This application realizes categorisedly based on the characteristics of user load curve self of user, more accords with power consumption load user's power consumption behavior habit, can realize accurate classification, and categorised effectual and the practicality is higher.
In this embodiment of the present application, obtaining the user load curve may include:
and acquiring the power load data of the user according to a preset period to form a user load curve.
Specifically, the user load curve is a load curve of user electric quantity, the load curve can reflect the change rule of the load along with time in a period of time, is the basis for dispatching the electric power of the electric power system and planning the electric power system, and is divided into an active power load curve and a reactive power load according to the load typesA load curve; according to the time, the load is divided into a daily load curve and an annual load curve. In one example, the data of the user load curve of the general user is similar to a time series, and the data collection is performed according to a fixed time interval, for example, once in 15 minutes, once in 1 hour, etc., and the specific collection mode is actually controlled. In another example, where the amount of data collected per day is denoted by T, the user load curve for the original user may be expressed as: x T ＝{x 1 ,x 2 ,...,x T }。
Fig. 2 is a schematic flow chart of power specific transformer user pre-classification according to an embodiment of the present disclosure. As shown in fig. 2, in the embodiment of the present application, the step 102 of pre-classifying the users according to the user load curve to determine the user type may include:
and step 205, under the condition that the user type is not the day cycle type, judging that the user type is a random type.
In the embodiment of the application, the electricity load users are pre-classified in the first stage according to the user load curves, so that all the user load curves are divided into five categories, namely a horizontal type, a zero-value regular type, an annual cycle type, a daily cycle type and a random type.
In one example, the user type of the electrical load user may be determined from the user load curve in a set order through a number of experimental statistical analyses. Specifically, the first step is to judge whether the user load curve meets the judgment condition of the horizontal type, if yes, the user type of the user corresponding to the user load curve is judged to be the horizontal type, and if not, the second step is to judge. And secondly, judging whether the user load curve meets the judgment condition of the zero-value regular type, if so, judging that the user type of the user corresponding to the user load curve is the zero-value regular type, and if not, entering the third step of judgment. And thirdly, judging whether the user load curve meets the judgment condition of the annual cycle type, if so, judging that the user type of the user corresponding to the user load curve is the annual cycle type, and if not, entering the fourth step of judgment. And fourthly, judging whether the user load curve meets the judgment condition of the daily cycle type, if so, judging that the user type of the user corresponding to the user load curve is the daily cycle type, and if not, judging that the user type of the user corresponding to the user load curve is the random type. And performing the four-step processing on the user load curve of each user until the user load curves of all the users are divided into five categories. Through multiple experimental statistical analysis, the classification mode is very practical and strong in practicability.
In this embodiment of the application, determining whether the user type is a horizontal type according to the user load curve may include at least one of:
determining the mode of a user load curve, and judging the user type to be a horizontal type under the condition that the occupation ratio of the mode is not lower than a first threshold value; or
And determining the variation coefficient of the user load curve, and judging the user type to be a horizontal type under the condition that the variation coefficient is not higher than a second threshold value.
In the embodiment of the present application, the load of the horizontal type, i.e., the load in the user load curve at any time point, is substantially the same, and the user type is substantially kept unchanged. The first threshold is a minimum value of a proportion of a mode in data of a user load curve when a user type of a user corresponding to the user load curve is a horizontal type. The second threshold is a maximum value of the variation coefficient of the user load curve when the user type of the user corresponding to the user load curve is a horizontal type.
In a specific embodiment of the present application, the first user load curve may be represented as X T ＝{x 1 ,x 2 ,...,x T Therein ofAnd T is the data amount in the user load curve.
Based on the user load curve X T The user type division of the first stage is performed. Firstly, judging whether the user is a horizontal user, wherein the load of the horizontal user at any time point is basically the same and basically remains unchanged, and two judging methods of the horizontal user can be adopted.
The specific judgment steps of the first method are as follows:
calculating a user load curve X T If phi is larger than or equal to 90%, the user is a horizontal user, wherein 90% is a first threshold, the numerical value of the first threshold can be determined according to actual conditions, and the first threshold is set to be 90% through experimental statistics.
The second method comprises the following specific judging steps:
calculating the user load curve X T The coefficient of variation cv satisfies the formula (1):
wherein cv is the coefficient of variation, T is the data volume in the user load curve, and x t For the load data of the user load curve, μ is the sequence X T σ is the sequence X T Standard deviation of (2).
If cv is less than or equal to 0.01, the user load curve is a horizontal user for the electricity load user, wherein 0.01 is a second threshold, the numerical value of the first threshold can be determined according to the actual situation, and it can be understood that the closer the second threshold is to 0, that is, the closer the coefficient of variation is to 0, the more accurate the determination result is.
It can be understood that, when the user load curve satisfies any determination condition of the first method or the second method, the user type corresponding to the user load curve may be determined to be a horizontal type. By the method, whether the user load curve is horizontal or not is judged, the judgment process is simple and clear, and the accuracy of the judgment result is improved. By the method, the horizontal users are judged, the power special transformer users meeting the judgment conditions of the horizontal users are classified into one class, a basis is provided for the class input of a downstream load prediction task or a prediction model construction strategy, and the quantity of models in a large number of power special transformer user load prediction tasks is reduced.
In this embodiment of the present application, determining whether the user type is a zero-valued rule type may include:
determining an initial two-dimensional matrix according to a user load curve;
based on the Hamming distance, obtaining a similarity matrix according to the initial two-dimensional matrix;
determining the mean value of the similarity matrix;
and in the case that the average value of the similarity matrix is larger than a third threshold value, judging that the user type is a zero-value regular type.
In the embodiment of the application, the zero-value rule is that the load exists in the load curve of the user only in a period of one year, and the load value of the part of the electric load users with the loads very close to each other, such as agricultural type users, only exists in certain solar terms. And when the user load curve is judged not to meet the condition of the horizontal type, judging whether the user load curve is of a zero-value regular type.
In particular, the first user load curve may be represented as X T ＝{x 1 ，x 2 ，...，x T T is the data amount in the user load curve. The specific process of judging whether the user type is a zero-value regular type is as follows:
curve X of user load T And converting the year cycle data format. Load sequence X for user electricity T Convert the annual period into X y，d，T And calculating the average value of daily load curve (the average value of annual daily calculated load values) to obtain an initial two-dimensional matrix X y，d Where y is the y-th year on which the curve is located and d is the date of the year.
If the load value on the same date in the year is 0, the date (row) on which the date is located is deleted, and X is obtained y，m Wherein m is less than or equal to d, and m is the number of the dates left after deletion.
Based on Hamming distanceCalculating X y，m Similarity of the matrix on the y axis to obtain a similarity matrix, wherein the similarity satisfies a formula (2):
wherein u and v are each X y，m Sequences in year y.
Acquiring a lower triangular matrix of the similarity matrix and calculating the mean value of the lower triangular matrix
If matrix
Greater than threshold similarity threshold by a third threshold s 0 . Then a zero-scale pattern.
By the method, the zero-value regular users are judged, the power special transformer users meeting the judgment conditions of the zero-value regular users are classified into one class, a basis is provided for category input of a downstream load prediction task or a prediction model construction strategy, and the number of models in a large number of power special transformer user load prediction tasks is reduced.
In this embodiment, determining whether the user type is a year period type may include:
determining an initial two-dimensional matrix according to a user load curve;
filtering the initial two-dimensional matrix to obtain a first correlation coefficient matrix;
determining a mean value of the first correlation coefficient matrix;
judging whether the user load curve has periodicity according to the mean value of the first correlation coefficient matrix;
and under the condition that the user load curve has a periodic type, judging that the user type is a year periodic type.
In the embodiment of the present application, the annual cycle type is an electricity load user whose load curve is in a cycle of year and the load trend on the same date in each year is very close. And when the user load curve is judged not to meet the condition of the zero-value regular pattern, judging whether the user load curve is a year period type.
Specifically, if the user load curve is a non-zero-value regular pattern, the user load curve X is determined T Whether the load trend is a year-periodic user, the year-periodic user is a user with a load curve in a year period, and the load trends of the same date in each year are very close to each other, and the specific judgment method comprises the following steps:
curve X of user load T And (4) converting the annual cycle data format. Load sequence X for users T Convert the annual period into X y，d，T And calculating the average value of daily load curve (the average value of annual daily load value) to obtain an initial two-dimensional matrix X d，y Where y is the y-th year on which the curve is located and d is the date of the year.
For the initial two-dimensional matrix X d，y The HP filtering calculation is respectively carried out according to the columns, the HP filtering can decompose the sequence into a trend component and a period component, and in the embodiment of the application, only the trend component g is needed to be used d，y . The HP filtering calculation satisfies formula (3):
X d，y ＝f(X d，y ；λ)＝g d，y +c d，y ； (3)
wherein, g d，y As a trend component, c d，y For the period component, λ is a smoothing parameter, and the specific value of the smoothing parameter can be adjusted according to the actual requirement, and after many experiments, for the user load curve of T =96, preferably, the value of the smoothing parameter can be set to 7 × T.
Based on the smoothed trend part g d，y Calculating Pearson correlation coefficient to obtain R d×d First correlation coefficient matrix R y×y The Pearson correlation coefficient calculation satisfies formula (4):
wherein R is a correlation coefficient, u, v are g respectively d，y Vector of middle row (daily average load sequence of a year)Columns);
are the mean values of the vectors u, v, respectively; d is g d，y D days (1).
Determining a first correlation coefficient matrix R y×y Based on the first correlation coefficient matrix R y×y Is that the user load curve has a period, if so, the user load curve X has a period T The corresponding user is a yearly periodic user.
In one example, the presence or absence of periodicity determination method is as follows:
using the first correlation coefficient matrix R y×y As input, a first correlation coefficient matrix R is applied y×y Converting into a lower triangular matrix, and removing a rows with minimum row mean and a columns with minimum column mean (the value of a can be defined but a < m) to obtain a lower triangular matrix R (m-a)×(m-a) Then calculate the lower triangular matrix R (m-a)×(m-a) And (4) average value of non-empty part, if the average value is more than 0.6, the periodicity is considered to exist, otherwise, the periodicity does not exist.
By the method, the annual cycle type users are judged, the power special transformer users meeting the judgment conditions of the annual cycle type users with the zero-value rule are classified into one class, a basis is provided for a downstream load prediction task by class input or a prediction model construction strategy, and the number of models in a large number of power special transformer user load prediction tasks is reduced.
In this embodiment of the present application, determining whether the user type is a day cycle type may include:
screening a load sequence of the latest first preset number of days in a user load curve;
determining a second phase relation number matrix according to the load sequence;
determining the mean value of the second correlation number matrix;
judging whether the user load curve has periodicity according to the mean value of the second correlation number matrix;
and under the condition that the user load curve has periodicity, judging the user type to be a day periodic type.
In the embodiment of the present application, the daily periodicity type is an electricity load user whose load curve has a certain periodicity by day or by week, such as a general industrial user, a commercial electricity user, and the like. And when the user load curve is judged not to meet the condition of the annual cycle type, judging whether the user load curve is the daily cycle type.
Specifically, if the user load curve corresponds to a non-annual periodic user, the user load curve X is determined T Whether the user is a daily periodic user or not, the load curve of the daily periodic user has certain periodicity by day or week, such as general industrial users, commercial power users and the like. The specific determination method for the day-periodic user is as follows:
selecting a user load curve X T The load sequence of the last first preset number of days, for example, the load sequence of the last 90 days is selected. Converting the load curve to X d×T Each row is k load values for one day.
According to the load matrix X d×T Calculating a Pearson correlation coefficient matrix according to days to obtain a second correlation coefficient matrix R of the correlation matrix between the days d×d 。
According to a second correlation number matrix R d×d And judging whether the user load curve has periodicity, if so, determining that the user is a daily periodic user, and otherwise, determining that the user is a random user.
In one example, the presence or absence of periodicity determination method is as follows:
using the first correlation coefficient matrix R y×y As input, a first correlation coefficient matrix R is applied y×y Converting into a lower triangular matrix, and removing a rows with minimum row mean and a columns with minimum column mean (the value of a can be defined but a < m) to obtain a lower triangular matrix R (m-a)×(m-a) Then calculate the lower triangular matrix R (m-a)×(m-a) And (4) average value of non-empty part, if the average value is more than 0.6, the periodicity is considered to exist, otherwise, the periodicity does not exist.
By the method, the daily periodic users are judged, the power special transformer users meeting the judgment conditions of the daily periodic users are classified into one class, a basis is provided for category input or a prediction model construction strategy for a downstream load prediction task, and the quantity of models in a large number of power special transformer user load prediction tasks is reduced.
In this embodiment of the present application, the performing secondary classification on the annual cycle type user and the daily cycle type user respectively to obtain the target classification result may include:
identifying an abnormal user load curve through a density clustering algorithm;
removing users corresponding to the abnormal user load curve to obtain target users;
and carrying out time series clustering processing on the user load curve of the target user to obtain a target classification result.
In the embodiment of the application, after the electricity load users are pre-classified in the first stage according to the user load curve, the electricity load users are divided into five categories. However, the problem of inconsistent period trend or period fluctuation still exists for users in the annual period and daily period types, and the load curves of users in the same type may have larger differences, so that load prediction cannot be performed through a unified model. And performing secondary classification on the two types of users in the second stage to obtain a final classification result, namely a target classification result.
In one example, the load sequence of the second preset number of days in the power load curve of the user can be screened out to obtain an initial load data set; carrying out normalization processing on the initial load data set to obtain a sample data set; carrying out zero equalization processing on the sample data set to obtain a zero equalization data set; smoothing the zero-mean data set to obtain a smoothed data set; and then carrying out abnormity identification on the smoothed data set through a density clustering algorithm DBSCAN, and then realizing user load curve clustering with different trends and periods through a time series clustering algorithm K-Shape.
Specifically, the specific process of performing secondary classification on the annual periodic users and the daily periodic users according to the user load curve is as follows:
selecting each user load sequenceX T Constructing a new load data set X from the load data of the latest second preset number of days m×l For example, 7-day load data is selected to construct a new load data set X m×l If l is the length of the sequence 7 × T, and m is the number of users in the year cycle or day cycle, also referred to as the number of samples.
To X m×l And (3) preprocessing data, firstly, normalizing the data, and unifying the load values of all samples into a scale of 0-1. Thus, X is m×l Normalizing according to the sample to obtain X' m×l The normalized calculation satisfies formula (5):
where t = l, is the length of the user payload data sequence.
In order to eliminate the differences caused by different load levels of different samples, the load sequence of the samples may be subjected to zero-averaging processing to obtain a zero-averaged data set, where the zero-averaging processing satisfies formula (6):
smoothing the zero-mean data set by adopting HP filtering, and selecting a trend part g m×l As the smoothed data, formula (7) is satisfied:
X″ m×l ＝f(X″ m×l ；λ)＝g m×l +c m×l ； (7)
where λ is a smoothing coefficient, preferably, the smoothing coefficient may be a multiple of 8, and in the embodiment of the present application, λ =8.
Using the smoothed data set g m×l And performing DBSCAN clustering, wherein partial abnormal sequences can be obtained through density clustering based on DBSCAN, and the partial sequences are load users of independent types and are divided into independent categories.
Using DTW as samples in DBSCAN clustering processThe distance measurement method of the sequence adjusts the proportion of independent or abnormal samples according to the density control parameter of the field, obtains the clustering label of each sample after clustering by using DBSCAN, wherein the samples with the label of-1 are divided into independent classes C 0 Sample division of other labels into C 1 To C 1 The samples of the category are clustered twice.
From the data set g m×l Reject class C 0 To obtain a sample of
And (3) carrying out the K-Shape clustering processing of the second stage on the data set, clustering the K-Shape clusters according to the similarity of the load curve shapes of different users, dividing the load curves of the users with the same Shape into the same cluster, namely dividing the load curves with the same period and fluctuation into the same category, and constructing the load prediction model of the users with the same category after division.
In K-Shape clustering, DTW is still used as a distance measurement mode, the clustering number N _ Class can be determined by a mode based on contour coefficient searching, and can also be manually specified, but the clustering number is not too small when the number of samples is more, the clustering number is recommended to be between 5 and 12, and the clustering number is set to be 6.
If the annual periodic user represents the category of the first-stage division by using a numerical value 3, the final division result can be represented as:
wherein
Is of independent class C 0 (ii) a Similarly, a daily periodic user is represented by a value of 4, and the final classification result can be expressed as:
wherein
As independent class C 0 。
In a specific embodiment of the present application, the power load users are classified by the above method, and the labels of the users of the first-stage pre-classified horizontal type, zero-degree regular type, year period type, day period type and random type are respectively represented by numerals 1, 2, 3, 4 and 5, so that the final classification result of the power load users is shown in table 1.
TABLE 1
In the pre-classification process of the first stage, the users are preliminarily divided into a plurality of types by the method, each type represents a large class of load curve users, and the method is equivalent to identifying and screening special or abnormal load curve users, eliminating abnormal samples of the clustering algorithm of the second stage and improving the classification effect of the second stage. In the embodiment of the application, the HP filtering is used for extracting the trend item of the load curve, the effect of data smoothing is achieved, and meanwhile, a periodic type judgment method based on Pearson correlation coefficients is adopted.
And in the second stage, users with periodic types are divided more finely, firstly, abnormal user load curves are identified through DBSCAN clustering, and then, user load curve clustering with different trends and periods is realized through a time series clustering algorithm K-Shape. The DTW sequence distance measurement method is used in the clustering algorithms of DBSCAN and K-Shape, so that the similarity between user load curves can be measured more accurately, and the clustering effect is better.
According to the technical scheme, a user load curve is obtained, users are pre-classified according to the user load curve to determine user types, the user types comprise a horizontal type, a zero-value regular type, a year-period type, a day-period type and a random type, and the year-period type users and the day-period type users are secondarily classified respectively to obtain target classification results. The user load curve self characteristics based on the user realize classification, more accord with the power consumption behavior habit of the power consumption load user, can realize accurate classification, and classification effect is good and the practicality is higher.
Fig. 3 is a structural block diagram of an unsupervised classification device for a power specific transformer user based on a load curve according to an embodiment of the present application. As shown in fig. 3, an embodiment of the present application provides a load curve-based unsupervised classification device for power specific transformer users, which may include:
a memory 310 configured to store instructions; and
a processor 320 configured to call the instructions from the memory 310 and when executing the instructions, to implement the method for controlling the boom described above.
Specifically, in the embodiment of the present application, the processor 320 may be configured to:
acquiring a user load curve;
pre-classifying users according to user load curves to determine user types, wherein the user types comprise a horizontal type, a zero-value regular type, a year period type, a day period type and a random type;
and respectively carrying out secondary classification on the annual cycle type users and the daily cycle type users to obtain a target classification result.
Further, the processor 320 may also be configured to:
the step of obtaining the user load curve comprises the following steps:
and acquiring the power load data of the user according to a preset period to form a user load curve.
Further, the processor 320 may also be configured to:
the pre-classifying the users according to the user load curve to determine the user types comprises the following steps:
judging whether the user type is a horizontal type or not according to the user load curve;
judging whether the user type is a zero-value regular type or not under the condition that the user type is not a horizontal type;
judging whether the user type is a yearly periodic type or not under the condition that the user type is not a zero-value regular type;
judging whether the user type is a day period type or not under the condition that the user type is not a year period type;
and under the condition that the user type is not a day cycle type, judging that the user type is a random type.
Further, the processor 320 may also be configured to:
judging whether the user type is a horizontal type according to the user load curve comprises at least one of the following steps:
determining the mode of a user load curve, and judging the user type to be a horizontal type under the condition that the occupation ratio of the mode is not lower than a first threshold value; or alternatively
And determining the variation coefficient of the user load curve, and judging the user type to be a horizontal type under the condition that the variation coefficient is not higher than a second threshold value.
Further, the processor 320 may also be configured to:
judging whether the user type is a zero-value regular type comprises the following steps:
determining an initial two-dimensional matrix according to a user load curve;
based on the Hamming distance, obtaining a similarity matrix according to the initial two-dimensional matrix;
determining the mean value of the similarity matrix;
and in the case that the average value of the similarity matrix is larger than a third threshold value, judging that the user type is a zero-value regular type.
Further, the processor 320 may also be configured to:
judging whether the user type is a year cycle type comprises the following steps:
determining an initial two-dimensional matrix according to a user load curve;
filtering the initial two-dimensional matrix to obtain a first correlation coefficient matrix;
determining a mean value of the first correlation coefficient matrix;
judging whether the user load curve has periodicity according to the mean value of the first correlation coefficient matrix;
and judging the user type to be a yearly periodic type under the condition that the user load curve has the periodic type.
Further, the processor 320 may also be configured to:
judging whether the user type is a daily cycle type comprises the following steps:
screening a load sequence of the latest first preset number of days in a user load curve;
determining a second phase relation number matrix according to the load sequence;
determining the mean value of the second correlation number matrix;
judging whether the user load curve has periodicity according to the mean value of the second correlation number matrix;
and under the condition that the user load curve has periodicity, judging that the user type is a day periodicity type.
Further, the processor 320 may also be configured to:
the secondary classification of the annual cycle type users and the daily cycle type users is respectively carried out to obtain target classification results, and the target classification results comprise:
identifying an abnormal user load curve through a density clustering algorithm;
removing users corresponding to the abnormal user load curve to obtain target users;
and carrying out time series clustering processing on the user load curve of the target user to obtain a target classification result.
By the technical scheme, the user load curve is obtained, the users are pre-classified according to the user load curve to determine the user types, the user types comprise a horizontal type, a zero-value regular type, a year period type, a day period type and a random type, and the year period type users and the day period type users are secondarily classified respectively to obtain a target classification result. The user load curve self characteristics based on the user realize classification, more accord with the power consumption behavior habit of the power consumption load user, can realize accurate classification, and classification effect is good and the practicality is higher.
The embodiment of the application also provides a machine-readable storage medium, and the machine-readable storage medium stores instructions for enabling a machine to execute the above-mentioned load curve-based power specific variable user unsupervised classification method.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.
Computer-readable media, including both permanent and non-permanent, removable and non-removable media, may implement the information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
The above are merely examples of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. An unsupervised classification method for power specific transformer users based on load curves is characterized by comprising the following steps:
acquiring a user load curve;
pre-classifying users according to the user load curve to determine user types, wherein the user types comprise a horizontal type, a zero-value regular type, a year cycle type, a day cycle type and a random type;
and respectively carrying out secondary classification on the annual cycle type users and the daily cycle type users to obtain a target classification result.
2. The power specific transformer user unsupervised classification method according to claim 1, wherein the obtaining of the user load curve comprises:
and acquiring the power load data of the user according to a preset period to form the user load curve.
3. The power specific transformer user unsupervised classification method according to claim 2, wherein the pre-classifying users according to the user load curve to determine the user type comprises:
judging whether the user type is the horizontal type or not according to the user load curve;
if the user type is not the horizontal type, judging whether the user type is the zero-value regular type;
if the user type is not the zero-value regular type, judging whether the user type is the annual cycle type;
judging whether the user type is the day cycle type or not under the condition that the user type is not the year cycle type;
and under the condition that the user type is not the day cycle type, judging that the user type is the random type.
4. The power specific transformer user unsupervised classification method according to claim 3, wherein the determining whether the user type is the horizontal type according to the user load curve comprises at least one of:
determining a mode of the user load curve, and judging that the user type is the horizontal type under the condition that the occupation ratio of the mode is not lower than a first threshold; or alternatively
Determining a variation coefficient of the user load curve, and judging the user type to be the horizontal type under the condition that the variation coefficient is not higher than a second threshold value.
5. The power-specific transformer user unsupervised classification method according to claim 3, wherein the determining whether the user type is the zero-value regular type comprises:
determining an initial two-dimensional matrix according to the user load curve;
based on the Hamming distance, obtaining a similarity matrix according to the initial two-dimensional matrix;
determining a mean value of the similarity matrix;
if the mean of the similarity matrix is greater than a third threshold, determining that the user type is the zero-valued regular type.
6. The power-specific transformer user unsupervised classification method according to claim 3, wherein the determining whether the user type is the year-round type comprises:
determining an initial two-dimensional matrix according to the user load curve;
filtering the initial two-dimensional matrix to obtain a first correlation coefficient matrix;
determining a mean value of the first correlation coefficient matrix;
judging whether the user load curve has periodicity according to the mean value of the first correlation coefficient matrix;
and under the condition that the user load curve has a periodic type, judging that the user type is the annual periodic type.
7. The power specific transformer user unsupervised classification method according to claim 3, wherein the determining whether the user type is the day cycle type comprises:
screening out a load sequence of the latest first preset number of days in the user load curve;
determining a second phase relation number matrix according to the load sequence;
determining a mean value of the second correlation number matrix;
judging whether the user load curve has periodicity according to the average value of the second correlation number matrix;
and under the condition that the user load curve has periodicity, judging that the user type is the day periodicity type.
8. The unsupervised classification method for power specific transformer users according to claim 1, wherein the secondarily classifying the users of the year period type and the users of the day period type respectively to obtain target classification results comprises:
identifying an abnormal user load curve through a density clustering algorithm;
removing users corresponding to the abnormal user load curve to obtain target users;
and carrying out time series clustering processing on the user load curve of the target user to obtain a target classification result.
9. The utility model provides a special transformer user unsupervised sorter of electric power based on load curve which characterized in that includes:
a memory configured to store instructions; and
a processor configured to invoke the instructions from the memory and when executing the instructions to enable the load curve based power specific user unsupervised classification method according to any one of claims 1 to 8.
10. A machine-readable storage medium having stored thereon instructions for causing a machine to perform the load curve based power specific customer unsupervised classification method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211310800.4A CN115687948A (en) | 2022-10-25 | 2022-10-25 | Power special transformer user unsupervised classification method based on load curve |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211310800.4A CN115687948A (en) | 2022-10-25 | 2022-10-25 | Power special transformer user unsupervised classification method based on load curve |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115687948A true CN115687948A (en) | 2023-02-03 |
Family
ID=85098922
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211310800.4A Pending CN115687948A (en) | 2022-10-25 | 2022-10-25 | Power special transformer user unsupervised classification method based on load curve |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115687948A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118130976A (en) * | 2024-05-10 | 2024-06-04 | 国网四川省电力公司广安供电公司 | Power grid fault diagnosis system and method based on multi-source heterogeneous data |
-
2022
- 2022-10-25 CN CN202211310800.4A patent/CN115687948A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118130976A (en) * | 2024-05-10 | 2024-06-04 | 国网四川省电力公司广安供电公司 | Power grid fault diagnosis system and method based on multi-source heterogeneous data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110717535B (en) | Automatic modeling method and system based on data analysis processing system | |
| CN112070125A (en) | Prediction method of unbalanced data set based on isolated forest learning | |
| CN111724278A (en) | Fine classification method and system for power multi-load users | |
| CN112734135B (en) | Power load prediction method, intelligent terminal and computer readable storage medium | |
| CN111028100A (en) | Refined short-term load prediction method, device and medium considering meteorological factors | |
| CN111143685A (en) | Recommendation system construction method and device | |
| CN113821542B (en) | Automatic significant feature recommendation system and method | |
| CN111861759A (en) | Matching method and system of product and customer group | |
| CN112732786A (en) | Financial data processing method, device, equipment and storage medium | |
| CN115687948A (en) | Power special transformer user unsupervised classification method based on load curve | |
| CN114611738A (en) | Load prediction method based on user electricity consumption behavior analysis | |
| CN111324790A (en) | Load type identification method based on support vector machine classification | |
| Ahmed et al. | Enhancing stock portfolios for enterprise management and investment in energy industry | |
| CN110458581B (en) | Method and device for identifying business turnover abnormality of commercial tenant | |
| CN117913808A (en) | Distributed photovoltaic power generation prediction method and device | |
| CN115481803A (en) | Financial time sequence prediction method, device and equipment based on industry crowding degree | |
| CN116739649A (en) | User response potential evaluation method and device | |
| CN115936184A (en) | Load prediction matching method suitable for multi-user types | |
| CN112365280A (en) | Power demand prediction method and device | |
| Cheng et al. | Chaotic load series forecasting based on MPMR | |
| Gómez-Boix et al. | Consumer segmentation through multi-instance clustering time-series energy data from smart meters | |
| CN113487080B (en) | Wind speed dynamic scene generation method, system and terminal based on wind speed classification | |
| CN115271274B (en) | Short-term daily load prediction method for power system and related equipment | |
| CN112836769B (en) | Demand response user classification method and system based on principal component analysis | |
| Chekole | APPLICATION OF MACHINE LEARNING TOOLS FOR PREDICTING DETERMINANT FACTORS. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |