Power consumer electricity consumption behavior image method considering load characteristics
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a method for drawing electricity utilization behaviors of power consumers by considering load characteristics.
Background
To date, a great deal of power data has been accumulated in power systems, which are derived from various links in the power systems, and which contain a great potential commercial value. With the improvement of the construction of the power internet of things, the development of big data and artificial intelligence becomes mature, and a new idea is provided for the research of the power industry on users. Because the big electric power data contains information such as basic information, living habits and behavior preferences of users, the value of the big electric power data is deeply mined, the upgrading innovation of electric power enterprises is favorably driven, and the power grid is favorably developed towards the intellectualization direction.
Because the number of power consumers is too large, it is difficult to completely know a certain consumer, so that the whole can be described by defining tags, and a three-dimensional and virtual consumer image, i.e. a consumer portrait, can be finally constructed by a plurality of tags. The user portrait is a data analysis tool for making an accurate marketing strategy for a target user, can play a good auxiliary role in the work of a decision maker, and a fine user portrait can help an electric power operator to understand user demands more clearly.
In recent years, in the field of power systems, research on user portrayal is increasing, and the user portrayal in the power systems aims to dig out the demand characteristics of power users, so that differentiated marketing strategies are implemented, and the service level of the power systems is improved. Therefore, it is highly desirable to provide a method for representing power consumption behaviors of power consumers in consideration of load characteristics, which can drill down the power consumption behaviors and power consumption habits of the power consumers based on the depth of the measurement data provided by the smart meters at the user terminals, finely analyze the user requirements, and study and represent the power consumption behaviors of the users.
Disclosure of Invention
The invention aims to provide a method for representing power consumption behaviors of power consumers in consideration of load characteristics, which is used for mining the power consumption behaviors and power consumption habits of the power consumers based on the aspect depth of measurement data provided by a smart electric meter at a user side, finely analyzing the user demands, researching the power consumption behaviors of the users and carrying out representation.
In order to achieve the above object, the present invention provides a power consumer electricity usage behavior imaging method considering load characteristics, comprising the steps of:
step S01, load decomposition, which is to divide the load decomposition into two types: seasonal typical loads or loads affected by other factors;
step S02, carrying out correlation analysis on the load influenced by other factors and other external influencing factors, and extracting correlation coefficients of the load and the external factors;
step S03, performing dimensionality reduction visual analysis on the seasonal base load and the load correlation coefficient influenced by other factors, so that multidimensional user electricity utilization data can be observed;
and step S04, performing cluster analysis on the seasonality basic load after the dimensionality reduction visualization analysis and the correlation coefficient of the load influenced by other factors respectively, and defining a label library.
Preferably, the step S01 specifically includes:
step S11, firstly, typical load curves of four seasons are respectively extracted;
s12, solving by adopting an optimization algorithm according to constraint conditions by using an objective function with the minimum distance between all load curves and the typical load curve, and obtaining the typical load curve;
and step S13, obtaining a load set influenced by other factors by correspondingly subtracting the typical load from the daily load set.
Preferably, the step S11 specifically includes:
extracting the daily typical load curve Y of each user in each quarterijNamely:
y in the formula (1)ijRepresenting the load set of the ith user in the jth season;
this season is m +1 days in total. Wherein
The load value of the user i in the h hour of the day d in the j season is set, and the typical daily load of the ith user in the j seasonal day is set as X
ij:
Xij=[x0,x1,x2......x23] (2)
X in the formula (2)hIs the typical load value of user i at h hour in j season.
Preferably, the step S12 specifically includes:
the objective function is the function that minimizes the distance of all load curves from the typical load curve:
in formula (3): sijRepresenting the value of the objective function; m is the number of days in the j season; n represents the nth day of the jth season, and t represents the tth hour of the nth day;
the constraint conditions are as follows:
min{Yij[n][t]}≤Xij[t]≤max{Yij[n][t]} (4)
formula (4) wherein t is 0, 1,2 … 23; n is 0, 1,2 … m;
the typical load curve can be obtained by adopting a hill climbing optimization algorithm to solve, and the obtained daily typical load is subtracted from the daily load, so that the daily load influenced by other factors can be obtained, namely:
in formula (5): y isijRepresenting a set of daily loads, XijRepresenting typical load, Zij.dRepresenting a daily load set influenced by other factors, namely a load value of the user i in the h hour on the day of the jth season influenced by other factors;
and respectively drawing a load curve before decomposition, a typical load curve after decomposition and a daily load curve influenced by other factors.
Preferably, in step S02, the pearson product-moment correlation coefficient ρ is usedXYTo reflect the degree of linear correlation of two variables:
x, Y in the formula (6) represent two variables, respectively; mu.sX、μYMean values of variables X and Y, respectively; sigmaXAnd σYThe standard deviations of the variables X and Y are indicated, respectively.
Preferably, the step S03 specifically includes:
after obtaining the correlation coefficients of the seasonal typical load and the load influenced by other factors, respectively carrying out dimensionality reduction visualization on the two types of loads with higher dimensionalities through a t-SNE algorithm;
for seasonal typical loads, the load data is first subjected to a Z-Score normalization process:
x in the formula (7)ijRepresenting the load of the ith user in the jth season; μ represents the mean of the load data; σ represents the standard deviation; and finally, mapping the data from a high-dimensional space to a two-dimensional space by adopting a t-SNE algorithm to obtain a dimension reduction result graph.
Compared with the prior art, the invention has the advantages that:
1. considering both the typical load under the influence of seasonal factors and the load influenced by other factors, then respectively performing dimension reduction visualization and clustering analysis on the seasonal typical loads of all users and the correlation coefficients of the loads influenced by other factors to respectively obtain corresponding tag libraries;
2. the method can help the power company to better know the electricity utilization behaviors of different types of users, and can make personalized marketing strategies according to personalized differences so as to improve the experience of the users and the service level of the power operators. In addition, the method can also be used for rapidly and accurately analyzing the power utilization behavior pattern of the user, assisting power operators to observe and analyze the power utilization behavior of the user, mining the power utilization behavior rules of the user and searching potential customers.
Drawings
Fig. 1 is a flowchart of a power consumer electricity consumption behavior portrayal method considering load characteristics according to the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, those skilled in the art will now describe the present invention in further detail with reference to the accompanying drawings.
Referring to fig. 1, the present invention provides a method for mapping power consumption behavior of a power consumer in consideration of load characteristics, including the following steps:
step S01, load decomposition, which is to divide the load decomposition into two types: seasonal typical loads or loads affected by other factors.
Wherein, step S01 specifically includes:
step S11, firstly, typical load curves of four seasons are respectively extracted;
s12, solving by adopting an optimization algorithm according to constraint conditions by using an objective function with the minimum distance between all load curves and the typical load curve, and obtaining the typical load curve;
and step S13, obtaining a load set influenced by other factors by correspondingly subtracting the typical load from the daily load set.
Step S11 specifically includes:
extracting the daily typical load curve Y of each user in each quarterijNamely:
y in the formula (1)ijRepresenting the load set of the ith user in the jth season;
this season is m +1 days in total. Wherein
The load value of the user i in the h hour of the day d in the j season is set, and the typical daily load of the ith user in the j seasonal day is set as X
ij:
Xij=[x0,x1,x2......x23] (2)
X in the formula (2)hIs the typical load value of user i at h hour in j season.
Step S12 specifically includes:
the objective function is the function that minimizes the distance of all load curves from the typical load curve:
in formula (3): sijRepresenting the value of the objective function; m is the number of days in the j season; n represents the nth day of the jth season, and t represents the tth hour of the nth day;
the constraint conditions are as follows:
min{Yij[n][t]}≤Xij[t]≤max{Yij[n][t]} (4)
formula (4) wherein t is 0, 1,2 … 23; n is 0, 1,2 … m;
the typical load curve can be obtained by adopting a hill climbing optimization algorithm to solve, and the obtained daily typical load is subtracted from the daily load, so that the daily load influenced by other factors can be obtained, namely:
in formula (5): y isijRepresenting a set of daily loads, XijRepresenting typical load, Zij.dRepresenting a daily load set influenced by other factors, namely a load value of the user i in the h hour on the day of the jth season influenced by other factors;
and respectively drawing a load curve before decomposition, a typical load curve after decomposition and a daily load curve influenced by other factors.
And step S02, performing correlation analysis on the load influenced by other factors and other external influencing factors, and extracting correlation coefficients of the load and the external factors.
For the loads influenced by other factors, the invention firstly carries out correlation analysis on the external factors (such as air temperature, electricity price, working day and the like) which can cause the load change, and analyzes the correlation between the load influenced by the external factors and the external factors such as air temperature, electricity price, working day and the like for each season of each user.
In step S02, the pearson product moment correlation coefficient ρ is usedXYTo reflect the degree of linear correlation of two variables:
x, Y in the formula (6) represent two variables, respectively; mu.sX、μYMean values of variables X and Y, respectively; sigmaXAnd σYThe standard deviations of the variables X and Y are indicated, respectively.
Step S03, performing dimensionality reduction visual analysis on the seasonal base load and the load correlation coefficient influenced by other factors, so that multidimensional user electricity utilization data can be observed;
step S03 specifically includes:
after obtaining the correlation coefficients of the seasonal typical load and the load influenced by other factors, respectively carrying out dimensionality reduction visualization on the two types of loads with higher dimensionalities through a t-SNE algorithm;
for seasonal typical loads, the load data is first subjected to a Z-Score normalization process:
x in the formula (7)ijRepresenting the load of the ith user in the jth season; μ represents the mean of the load data; σ represents the standard deviation; and finally, mapping the data from a high-dimensional space to a two-dimensional space by adopting a t-SNE algorithm to obtain a dimension reduction result graph.
And step S04, performing cluster analysis on the seasonality basic load after the dimensionality reduction visualization analysis and the correlation coefficient of the load influenced by other factors respectively, and defining a label library.
1) Seasonal typical load
After dimension reduction visualization, the class number GMM clustering algorithm capable of obtaining load classification respectively considers K classes of loads to be in accordance with certain Gaussian distribution, and probability density of the K classes of Gaussian distribution and weight alpha of each class can be respectively estimated through Gaussian model training datakSecondly, calculating the probability of each load data appearing in the K-class Gaussian distribution, and finally, classifying the data into a class with the maximum probability value by comparing the probability of the data appearing in different classes. Finally, obtaining the classification result of the typical load, and respectively averaging each type of load to obtain three typical curves Xk(k ═ 1,2,3), typical curve X, since each user has differences in both the number of appliances and the energy consumption, and the power usage behavior is more focused on the trend than the magnitude of the order of magnitudekThe following treatments were carried out:
after clustering, each typical load is labeled with a category label 1.
2) Loads influenced by other factors
Similarly, the load correlation coefficient affected by other factors is subjected to dimensionality reduction visualization to obtain a classification number K, and the correlation coefficient is clustered through a GMM algorithm, and the principle is the same as the above. After clustering, attaching a class label 2 to each load according to the sensitivity of the load to different influence factors.
And finally, the result of the two types of labels is used for representing the electricity utilization behaviors of the power consumers.
According to the method, the typical load under the influence of seasonal factors and the load influenced by other factors are considered, dimension reduction visualization and clustering analysis are respectively carried out on the seasonal typical loads of all users and the correlation coefficients of the loads influenced by other factors, and corresponding label libraries are respectively obtained.
Aiming at the two types of loads of different types, user labels under the two types of loads are respectively obtained based on a t-SNE dimension reduction algorithm and a GMM clustering algorithm, so that a more accurate user portrait can be obtained. The method comprises the following steps that a t-SNE dimension reduction algorithm is adopted in load analysis, multidimensional data are reduced to a two-dimensional space to be displayed, results are clearer and more visual, and meanwhile local and global structures of the data are well reserved; the GMM clustering algorithm is adopted to carry out clustering analysis on the load in the clustering analysis of the load, and the data set can be in an arbitrary elliptical shape due to the adoption of mean value and standard deviation parameters, so that the clustering effect is good.
The construction, features and functions of the present invention are described in detail in the embodiments illustrated in the drawings, which are only preferred embodiments of the present invention, but the present invention is not limited by the drawings, and all equivalent embodiments modified or changed according to the idea of the present invention should fall within the protection scope of the present invention without departing from the spirit of the present invention covered by the description and the drawings.