CN109741091B - User load classification method based on basic load reduction strategy - Google Patents

Abstract

The invention discloses a user load classification method based on a basic load reduction strategy, comprising: step 1, extracting user load data, and preprocessing the load data; step 2, selecting a load reduction reference value, taking the aggregation degree as an evaluation parameter, and using the reduction reference value The method of load numerical value distinguishes the load type; step 3, according to the load type distinction result, the load data is clustered by the fuzzy C-means algorithm. Aiming at the problem of wrong identification of user types due to user scale in the current load clustering method, the invention proposes an improved load clustering algorithm processing skill, and uses a fuzzy algorithm to verify the accuracy of the algorithm proposed by the invention. The method adopted in the present invention can effectively eliminate the interference of the user scale on the user type.

Description

User load classification method based on basic load reduction strategy

technical field

The invention belongs to the field of power system load classification, in particular to a user load classification method based on a basic load reduction strategy.

Background technique

Load clustering is to integrate a large number of users into different aggregates through certain mathematical means. According to the real-time operation of the power grid, rationally guiding different types of aggregates to use electricity in an orderly manner can generate huge economic benefits. The existing clustering methods generally cluster the trend and value of the load. However, due to the inconsistent scale and number of users in different regions, users of the same type or electricity consumption pattern cannot be identified, resulting in insufficient classification results. In view of the above problems, the present invention proposes a clustering method based on load reduction, and verifies the clustering result of the present invention through a fuzzy algorithm.

In the prior art, there are also other methods for implementing load classification. For example, the Chinese Patent Application No. 201810382946.7 discloses a load classification method based on decision tree fuzzy C clustering, which first determines the optimal number of classifications through agglomerative hierarchical clustering algorithm, and then adopts the fuzzy C-means clustering algorithm to classify the load. The data is clustered to achieve load classification.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a user load classification method based on a basic load reduction strategy to solve the problem of wrong identification of user types due to user scale in the current load clustering method.

For achieving the above object, the technical scheme adopted in the present invention is:

The user load classification method based on the basic load reduction strategy includes the following steps:

Step 1, extracting user load data, and preprocessing the load data;

Step 2, select the reference load reduction value, take the degree of aggregation as the evaluation parameter, and use the method of reducing the reference load value to distinguish the load types;

Step 3: Clustering the load data by using the fuzzy C-means algorithm according to the load type discrimination result.

Preferably, the preprocessing step in step 1 includes: defining a deviation rate η,

表示负荷数据在一时间周期内的平均值。Among them, _yi represents the load data of a certain point,

Indicates the average value of load data over a period of time.

Preferably, when η>500%, it is necessary to check whether the data is a bad data point. If most of the load points in the current load curve exceed the standard, it will be regarded as normal measurement data; if only a few points exceed the standard, it will be judged as bad data. point.

Preferably, the specific process of step 2 is:

2.1, select the reference value for reducing the load;

2.2, data standardization;

2.3, calculate the data gradient;

2.4. Determine whether the degree of aggregation meets the requirements. If so, output the differentiated load types. If not, reselect the load reduction reference value and go to 2.2.

Preferably, step 3 further comprises:

3.1, determine the number of classifications, power exponent, and membership matrix;

3.2, calculate the cluster centrality;

3.3, modify the membership function and objective function;

3.4, when the membership function satisfies the termination limit or satisfies the maximum step size, stop the iteration, otherwise go to 3.2.

时或满足最大步长时，停止迭代。Preferably, for the membership function, the termination limit ε _J > 0 is given or the maximum step size l is defined, when the

The iteration stops when the maximum step size is met.

Preferably, the iterative formulas for membership and cluster centrality are as follows:

I _j ={1≤i≤c,d _ij =0}

时，when

hour,

时，

时，when

hour,

hour,

Preferably, step 3 further includes: assuming that the sample is taken from the p-element sample population, and classifying the sample X ₁ , X ₂ , LX _n population into class c, where the value of c is not less than 2, and vi represents the _i -th cluster center , set the variable u _ij to represent the membership degree of the i-th sample to the j-th sample, then set the objective function as follows:

Restrictions:

u _ij ∈[0,1]1≤j≤n,1≤i≤c

Calculate the optimal solution according to the Lagrange multiplier method:

And according to the given constraints, it is optimized in MATLAB.

Compared with the prior art, the beneficial effects of the present invention are: the present invention proposes an improved load clustering algorithm processing skill for the problem of wrongly identifying user types due to the user scale in the current load clustering method, and uses a fuzzy algorithm to verify the results. The accuracy of the algorithm proposed by the present invention; the algorithm proposed by the present invention has better clustering effect, and can perform more effective clustering of loads according to the actual needs of the project or the company.

Description of drawings

1 is a schematic flowchart of a user load classification method based on a basic load reduction strategy of the present invention;

Figure 2(a)-(e) are schematic diagrams of traditional fuzzy clustering results;

3(a)-(c) are schematic diagrams of clustering results obtained by the method of the present invention;

4 is a schematic diagram of the relationship between the cluster center and the contour value when c=3;

5 is a schematic diagram of the relationship between the cluster center and the contour value when c=5;

In the figure: Q-load; t-time; T-cluster center; S(i)-contour value.

Detailed ways

The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

As shown in FIG. 1, the present invention provides a user load classification method based on a basic load reduction strategy, including the following steps:

Step 1, extracting user load data, and preprocessing the load data;

Step 2: Select the reference load reduction value, take the degree of aggregation as the evaluation parameter, and use the method of reducing the reference load value to distinguish the load types. In order to exclude the same type of load, if the clustering results are different due to the difference in the regional scale and the number of users, the present invention adopts the method of reducing the reference load value, amplifies the difference of the load curve, and makes the gradient of the load curve larger, so as to distinguish the load type.

The specific process of step 2 is:

2.1, select the reference value for reducing the load;

2.2, data standardization;

2.3, calculate the data gradient;

2.4. Determine whether the degree of aggregation meets the requirements. If so, output the differentiated load types. If not, reselect the load reduction reference value and go to 2.2.

Step 3: Clustering the load data by using the fuzzy C-means algorithm according to the load type discrimination result. With the help of the fuzzy set theory proposed by L.A. Zadeh (1960s) and the fuzzy clustering algorithm proposed by Ruspini and Bezdek, the present invention performs direct clustering on the load curve in combination with the user's electricity usage habits and electricity usage characteristics.

Step 3 further includes:

3.1, determine the number of classifications, power exponent, and membership matrix;

3.2, calculate the cluster centrality;

3.3, modify the membership function and objective function;

3.4, when the membership function satisfies the termination limit or satisfies the maximum step size, stop the iteration, otherwise go to 3.2.

After the above steps, the elements and cluster centers in the final membership function matrix can be determined, so that the value of the objective function can be minimized.

At the beginning of the algorithm, the present invention preprocesses the load data and excludes the influence of the regional scale and the number of users on the clustering result, which has important practical significance. Then clustering is performed according to the characteristics of different types of users, and the effect is better.

Example

The data of the present invention adopts the data of the PJM power market in the United States, selects the load data of 24 hours, and analyzes and demonstrates the calculation example. In this embodiment, the load data of different regions of the PJM electricity market in the United States is selected. In order to exclude the interference of other factors, 113 groups of load data of different regions on a unified working day are selected for cluster analysis.

(1) The present invention first preprocesses the data, and eliminates the data with too severe deviation, so as to prevent the extreme value from interfering with the classification. Define the deviation rate η:

表示负荷数据在24h内的平均值。where y _i represents the load data at a certain point,

Indicates the average value of load data within 24 hours.

When η>500%, it is necessary to check whether the data is a bad data point. If most of the load points in the load curve exceed the standard, it is regarded as normal measurement data; if only a few points exceed the standard, it is judged as a bad data point.

(2) The base load reduction is worth selecting.

In order to exclude the same type of load, if the clustering results are different due to the difference in the regional scale and the number of users, the present invention adopts the method of reducing the reference load value, amplifies the difference of the load curve, and makes the gradient of the load curve larger, so as to distinguish the load type.

(3) Fuzzy algorithm

In this embodiment, the load curve is directly clustered by means of the fuzzy set theory proposed by LAZadeh (1960s) and the fuzzy clustering algorithm proposed by Ruspini and Bezdek, combined with the user's electricity usage habits and electricity usage characteristics. Suppose the sample is taken from the p-element sample population. The samples X ₁ , X ₂ , LX _n are generally classified as class c, where the value of c is not less than 2. v _i represents the i-th cluster center. Let the variable u _ij represent the membership degree of the ith sample to the jth sample.

The objective function is as follows:

Restrictions:

u _ij ∈[0,1]1≤j≤n,1≤i≤c

Calculate the optimal solution according to the Lagrange multiplier method:

And according to the constraints given by the system, it is optimized in MATLAB.

Further, according to the Lagrange multiplier method, the optimal solution of J(u,c) can be obtained. The iterative formulas for membership and cluster centrality are as follows:

I _j ={1≤i≤c,d _ij =0}

时，when

hour,

时，

时，when

hour,

hour,

The fuzzy algorithm steps are as follows:

同时处理方式是去除[0,1] 上的随机数作为矩阵初始数据。l＝1作为第1步迭代。1) Determine the number of classifications c and the power exponent m>1 and the membership matrix

At the same time, the processing method is to remove the random numbers on [0,1] as the initial data of the matrix. l=1 as the first iteration.

2) Calculate the cluster centrality according to the above formula.

3) Modify the membership function U ^(l) and the objective function J ^(l) .

in,

时或满足最大步长时，停止迭代。否则跳转步骤2)。4) For a given membership function termination limit ε _J > 0 or define the maximum step size l. when satisfied

The iteration stops when the maximum step size is met. Otherwise, skip to step 2).

After the above steps, the elements in the final membership function matrix U and the cluster center V can be determined, so that the value of J(U, V) of the objective function is minimized.

Aiming at the problem of wrong identification of user types due to user scale in the current load clustering method, the invention proposes an improved load clustering algorithm processing skill, and uses a fuzzy algorithm to verify the accuracy of the algorithm proposed by the invention.

The traditional load clustering algorithm is compared with the improved load clustering algorithm of the present invention, as shown in the accompanying drawings. Figure 2 (a)-(e) are schematic diagrams of traditional fuzzy clustering results. 3(a)-(c) are schematic diagrams of clustering results obtained by the method of the present invention. Comparing the clustering effect of the traditional method, it can be verified that the method adopted in the present invention can effectively eliminate the interference of the user scale on the user type.

For the traditional method, according to the results of the fuzzy algorithm, the load can be roughly divided into five categories, and the categories are distinguished as follows:

(1) The time points at which the load peak and valley values appear are different. Comparing the second and fifth types of loads with the fourth type of loads, the peaks appear at 9:00 and 8:00, respectively, and there is a significant time difference.

(2) The direction of the load curve is inconsistent. The second, fourth, and fifth types of loads have double peaks, while the first and third types have been in a high load state from 9:00 to 21:00.

(3) There are differences in the number of peaks and valleys. The peak value of the second type of load and the fourth type of load is 4×10 ⁴ MW, and the valley value is slightly less than 3×10 ⁴ MW. Compared with the first, third and fifth types of loads, the peak-to-valley value is reduced by about 5×10 ³ MW.

In traditional methods, due to the problem of unclear boundaries in the algorithm, the phenomenon of crossover between categories is serious. For the power system, the distinguishing point (3) is not conducive to load clustering. Because some regions have inconsistent peak-to-valley values due to regional scale, they should not be separately distinguished.

On the other hand, the present invention adopts the method of load reduction, which increases the load change gradient, and can solve the misidentification caused by different user scales. The results are shown in Figures 3(a)-(c).

Comparing Fig. 2 and Fig. 3, it can be seen that when the classification result becomes three types, the regularity of the curve is obviously strengthened, and the weight according to the change trend in the clustering basis is increased, which is conducive to identifying the electricity consumption law of users. According to Fig. 3, the load can be clearly divided into the steady load, the bimodal load and the unimodal load. It provides a good experimental basis for better formulating demand-side response strategies for users who do not use regions and electricity habits.

The present invention uses the contour value as an index of the algorithm.

Specifically, the contour value S(i) is used for comparative analysis,

where a represents the normalized distance between the i-th point and points of the same category; b represents the normalized distance from points of different categories. The value range of the contour value S(i) is [-1, 1], and the larger S(i) is, the more reasonable the classification is. When S(i)<0, it means that the classification of this point is unreasonable, and there is a more reasonable classification method.

As shown in the accompanying drawings, FIG. 4 is a schematic diagram of the relationship between the cluster center and the contour value when c=3. FIG. 5 is a schematic diagram of the relationship between the cluster center and the contour value when c=5.

According to the processing technique proposed by the present invention, after the load value is reduced by a certain reference value, the change gradient between the loads is enlarged, and the load is re-divided into clustering criteria. According to Figure 4, when the clustering standard is divided into 5 categories, the membership degree of the samples in the first and second categories has a negative value, indicating that the sample at this point has a low degree of membership to the cluster center, and there is a better clustering Way. According to Figure 5, when the samples are divided into three categories, the contour values of the samples are all positive, and for most of the samples, the degree of membership is high, which is a more reasonable classification method. And this classification method avoids the phenomenon of misclassification caused by the scale of power load data.

Comparing the clustering effect of the traditional method, it can be seen that when the number of classifications is reduced from 5 to 3, it is found from the comparison of Fig. 4 and Fig. 5 that the classification result of the algorithm of the present invention is more accurate.

Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, and substitutions can be made in these embodiments without departing from the principle and spirit of the invention and modifications, the scope of the present invention is defined by the appended claims and their equivalents.

Claims (4)

1. The user load classification method based on the basic load reduction strategy is characterized by comprising the following steps:

step 1, extracting user load data and preprocessing the load data;

the pretreatment step comprises the following steps: defining deviation ratio

，

Wherein,

the load data of a certain point is represented,

represents an average value of the load data over a period of time;

when in use

In the process, whether the data are bad data points or not needs to be checked, if most load points of the current load curve exceed the standard, the data are determined to be normal measurement data; if only the individual point exceeds the standard, determining the data point as a bad data point;

step 2, selecting a load reduction reference value, taking the polymerization degree as an evaluation parameter, and distinguishing the load types by adopting a method of reducing a reference load value; the specific process is as follows:

2.1, selecting a load reduction reference value;

2.2, data standardization;

2.3, calculating a data gradient;

2.4, judging whether the polymerization degree meets the requirement, if so, outputting the distinguished load type, and if not, reselecting a load reduction reference value and turning to 2.2;

step 3, clustering the load data by adopting a fuzzy C-means algorithm according to the load type distinguishing result; further comprising:

3.1, determining the classification number, the power index and the membership matrix;

3.2, calculating the clustering centrality;

3.3, modifying the membership function and the target function;

and 3.4, stopping iteration when the membership function meets the termination limit or the maximum step length, and otherwise, turning to 3.2.

2. The method of claim 1, wherein a termination limit is given for the membership function

Or define a maximum step size

When it is satisfied

And stopping iteration when the maximum step length is met.

3. The method for classifying user loads based on the basic load shedding strategy according to claim 1, wherein the iterative formulas of the membership degree and the clustering center degree are as follows:

when in use

When the temperature of the water is higher than the set temperature,

when in use

When the temperature of the water is higher than the set temperature,

when the temperature of the water is higher than the set temperature,

。

4. the method of claim 1 wherein step 3 further comprises the step of classifying the user load based on the basic load shedding strategyThe method comprises the following steps: provided that the sample is taken from

Meta-sample population, sample

Is divided into

Class I, wherein

The value of (a) is not less than 2,

is shown as

Individual cluster center, set variable

Is shown as

And the membership degree of each sample to the jth sample, setting an objective function as follows:

constraint conditions are as follows:

calculating an optimal solution according to a Lagrange multiplier method:

and optimizing in MATLAB according to given constraint conditions.

Images