CN112308117A - Homogeneous crowd identification method based on double-index particle swarm algorithm - Google Patents

Abstract

The invention provides a homogeneous crowd identification method based on a double-index particle swarm algorithm, aiming at the defect that a public health service platform cannot be comprehensively analyzed by a single-index clustering algorithm, and the method comprises the following steps: collecting the use crowd information of a public health service platform as a user information data set; obtaining two initial adaptive values of the user information data set through a clustering algorithm; and iterating the two initial adaptive values as adaptive functions to obtain a clustering result and obtain homogeneous crowd information data. According to the method, optimization guidance is performed on the particle swarm clustering on the basis of two index results through double-index adaptive value evaluation, the evaluation tendency of a single index is eliminated, the application unicity of an internal index is expanded, the labor and the time can be saved, and the complex and diverse crowd information can be comprehensively analyzed.

Description

Homogeneous crowd identification method based on dual-index particle swarm algorithm

technical field

The invention relates to the field of swarm intelligence evolution, and mainly relates to a homogeneous crowd identification method based on a dual-index particle swarm algorithm.

Background technique

At present, for the identification of homogeneous crowds, many clustering algorithms have been applied to crowd identification at home and abroad, such as the k-means algorithm, but it has the shortcomings of being sensitive to the initial center and the value of K; based on The grid method has defects in accuracy; the multiple regression method is too sensitive to the data; the density-based method is not strong against noise and is more dependent on the value of the field radius.

In addition, due to the tendency of the design of internal indicators, the expressive ability of a single indicator is limited. Therefore, the genetic algorithm combined with the clustering algorithm, the difference algorithm combined with the clustering algorithm and other methods take a single indicator as the optimal fitness value evaluation method, and the results obtained is also singular.

For workers engaged in public health services, it is extremely important to optimize the work platform according to the various users of the service platform, so that public health services can meet the needs of people to the greatest extent, and it is always necessary to carry out with the use of the population. . The population of the public health service platform is diverse, and there will be differences in different age groups and different living environments. In the past platform optimization, more people were used to understand the user population in the form of online questionnaires and offline surveys, or to analyze the user population data through a single-index clustering algorithm, but the above methods require a lot of manpower and Time, and in the face of the complexity and diversity of crowd information, the analysis is not comprehensive enough.

SUMMARY OF THE INVENTION

Aiming at the deficiency that the single-index clustering algorithm cannot comprehensively analyze the public health service platform, the present invention proposes a homogeneous population identification method based on the dual-index particle swarm algorithm, and uses the dual-index fitness value evaluation to perform the homogeneous population clustering analysis , the efficiency and comprehensiveness of platform optimization have been fully improved.

The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a method for identifying a homogeneous crowd based on a dual-index particle swarm algorithm, including:

Collect the information on the user population of the public health service platform as a user information data set;

Obtaining two initial fitness values from the user information data set through a clustering algorithm;

The two initial fitness values are used as the fitness function to iterate to obtain the clustering result, and the information data of the homogeneous population is obtained.

The “collection of user information of the public health service platform as a user information dataset” includes:

Convert all the information on the users of the public health service platform into numbers to obtain a data set;

converting the dataset into a readable file format;

Eliminate useless attribute data columns in the data set to obtain a processed data set;

Standardize the processed data set to obtain the user information data set.

The "readable file format" includes: csv format and/or bat format.

The "information on the population of users of the public health service platform" includes: nationality, place of residence, and age information.

The two "initial fitness values" are: Fitness1 (CH), Fitness2 (S_Dbw).

The steps for obtaining the Fitness1(CH) are:

Indicator formula:

表示类别中离差矩阵的迹；

represents the trace of the dispersion matrix in the category;

表示类别间离差矩阵的迹，m表示整个数据集的平均值向量；N为样本数量；K为迭代次数。

Represents the trace of the dispersion matrix between classes, m represents the mean vector of the entire dataset; N is the number of samples; K is the number of iterations.

The steps for obtaining the Fitness2 (S_Dbw) are:

Indicator formula: S_Dbw(c)=Scat(c)+Dens_bw(c)

in:

Dens_bw(c) is used to evaluate the relationship between the density of the two classes together and the density of each individual class;

density(u) is used to characterize the number of points around u, and the threshold for comparison is stdev in 1;

stdev represents the average deviation of each cluster in a dataset;

Scater(c) represents the average dispersion between classes.

The process of "iterating the two initial fitness values as adaptation functions to obtain clustering results and obtaining homogeneous crowd information data" is as follows:

a. Initialize the parameters of the particle swarm algorithm:

b. Calculate the fitness value Fitness1 and fitness value of each particle according to the CH and S_Dbw index formulas respectively;

c. For each particle, compare the fitness value Fitness1 of its current position with the fitness value Fitness2 of the two fitness values corresponding to its historical best position, if the two fitness values of the current position are higher, then use The current location updates the best location in history, otherwise it will not be updated;

d. For each particle, compare the fitness value Fitness1 of its current position with the fitness value Fitness2 of the two fitness values corresponding to its global best position, if the two fitness values of the current position are higher, then use The current position updates the global best position, otherwise it will not be updated;

e. Update the position and velocity of the particle:

The particle velocity update formula:

The particle position update formula:

Among them, Vidk represents the d-dimensional component of the velocity vector of particle i in the k-th iteration; xidk represents the d-dimensional component of the position vector of particle i in the k-th iteration; c1, c2 represent acceleration constants; r1, r2 represent two random parameters, Value range [0,1]; w represents inertia weight;

f. If the end condition is not met, return to step b. If the end condition is met, the algorithm ends, and the global optimal position is the global optimal solution;

Finally, the clustering results are output to obtain accurate homogeneous population information data.

The beneficial effects of the present invention are as follows: the present invention conducts optimization guidance for particle swarm clustering on the basis of the results of the two indicators through the evaluation of the fitness value of the two indicators, eliminates the evaluation tendency of a single indicator and expands the application singleness of the internal indicators; The single-index and dual-index fitness value evaluation can more completely retain the crowd data information in the clustering process, and to the greatest extent ensure that the clustering results based on the particle swarm algorithm have the optimal results of inter-class dispersion and intra-class compactness; the present invention In the case of a large number of crowd data attributes, the method described above can more inclusively obtain the most similar classification of all samples, save manpower and time, and comprehensively analyze complex and diverse crowd information.

Description of drawings

FIG. 1 is a flowchart of user crowd information extraction in the present invention.

Fig. 2 is a coding diagram used by the algorithm of the present invention.

FIG. 3 is a flowchart of the dual-index particle swarm algorithm according to the present invention.

Detailed ways

The present invention will be further described below with reference to the accompanying drawings.

As shown in Figures 1 to 3, the method for identifying homogeneous crowds based on the dual-index particle swarm algorithm according to the present invention includes:

Collect the information on the user population of the public health service platform as a user information data set;

Obtaining two initial fitness values from the user information data set through a clustering algorithm;

The two initial fitness values are used as the fitness function to iterate to obtain the clustering result, and the information data of the homogeneous population is obtained.

The specific steps are:

First, all the information on the user population of the public health service platform is converted into numbers, such as nationality and place of residence, age information, body mass index (BMI), activity of daily living score (ADL); then the data set can be converted into this patented procedure. Read the file format, such as csv, bat, read the data set, read all the attribute values of the crowd information, attribute filtering, eliminate the useless attribute data column, contain N samples, each sample has D attributes, and then analyze all the data. Carry out standardization processing to obtain a user information data set. In the actual situation obj[i].dataX[j] represents the jth attribute of the ith sample.

After completing all the above steps, the clustering algorithm program enters the population individual coding, initializes the population, and uses the nearest neighbor clustering rule NMP (nearest multiple prototypes) to obtain the initial clustering clusters, and uses the Euclidean distance formula to calculate each class. The initial class center of the cluster, and then obtain the initial fitness value Fitness1(CH), Fitness2(S_Dbw)

①CH (Calinski-Harabasz):

表示类别中离差矩阵的迹，是各个数据点到所在类的质心点的距离之和。

Represents the trace of the dispersion matrix in the class, which is the sum of the distances from each data point to the centroid of the class.

表示类别间离差矩阵(由向量距离值组成的矩阵)的迹，m表示整个数据集的平均值向量。in,

represents the trace of the between-class dispersion matrix (a matrix of vector distance values), and m represents the mean vector for the entire dataset.

②S_Dbw:

S_Dbw(c)=Scat(c)+Dens_bw(c)

Dens_bw(c) is used to evaluate the relationship between the density of two classes together, and the density of each individual class. The density of the two classes together, which is significantly less than the density of each individual class, indicates better clustering.

density(u) is used to characterize the number of points around u, and the threshold for comparison is stdev in 1.

stdev represents the average deviation of each class in a dataset.

Scater(c) represents the average dispersion between classes.

Calculate the above two internal indicators as the fitness value function, and then enter the following iterative process:

while (the number of loops is less than the set value) do

a. Initialize the parameters of the particle swarm algorithm:

Position of particle i: x _i =(xi1,xi2,...,xiD);

The speed of particle i: vi=(vi1,vi1,...,viD);

The best position experienced by particle i: pbesti=(pi1,pi2,...,piD), its corresponding bFitness1 and bFitness2;

The best position experienced by all particles: gbesti=(gi1,gi2,...,giD), its corresponding gFitness1 and gFitness2;

The position of each dimension should be limited to the interval: [Xmin,d,Xmax,d];

The speed of each dimension should be limited to the interval: [-Vmax,d,Vmax,d];

b. Calculate the fitness values Fitness1 and Fitness2 of each particle according to the CH and S_Dbw index formulas;

c. For each particle, compare the two fitness values of its current position with the two fitness values corresponding to its historical best position (pbest). If the two fitness values of the current position are higher, update the history with the current position The best location, otherwise it will not be updated;

d. For each particle, compare the two fitness values of its current position with the two fitness values corresponding to its global best position (gbest). If the two fitness values of the current position are higher, update the global position with the current position. The best location, otherwise it will not be updated;

e. Update the particle's position and velocity according to the formula:

The particle velocity update formula:

The particle position update formula:

Among them, Vidk represents the d-dimensional component of the velocity vector of particle i in the k-th iteration; xidk represents the d-dimensional component of the position vector of particle i in the k-th iteration; c1, c2 represent the acceleration constant, and adjust the maximum learning step size; r1, r2 Represents two random parameters, the value range is [0, 1], to increase the randomness of the search w represents the inertia weight, a non-negative number, used to adjust the search range of the solution space;

f. If the end condition is not met, return to step b. If the end condition is met, the algorithm ends, and the global best position (gbest) is the global optimal solution.

end while

Finally, the clustering results are output, and accurate homogeneous population information data can be obtained.

It should be clear that in addition to CH and S_Dbw used in this paper, other internal indicators can also be selected to replace the fitness value according to the characteristics of the indicators and the purpose of problem analysis. At the same time, clustering based on genetic algorithm and clustering combined with differential algorithm can be used to replace the clustering scheme based on particle swarm algorithm for clustering analysis.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical scope disclosed by the present invention can easily change or replace them, all belonging to the present invention. within the scope of protection. Therefore, the protection scope of the present invention is described in accordance with the protection scope of the claims.

Claims (8)

1. A homogeneous population identification method based on a double-index particle swarm algorithm is characterized by comprising the following steps:

collecting the use crowd information of a public health service platform as a user information data set;

obtaining two initial adaptive values of the user information data set through a clustering algorithm;

and iterating the two initial adaptive values as adaptive functions to obtain a clustering result and obtain homogeneous crowd information data.

2. The homogeneous population recognition method based on the dual-index particle swarm algorithm according to claim 1, wherein the collecting of the use population information of the public health service platform as the user information data set comprises:

the method comprises the steps that information of users of a public health service platform is converted into numbers to obtain a data set;

converting the data set into a readable file format;

removing useless attribute data columns in the data set to obtain a processed data set;

and carrying out standardization processing on the processed data set to obtain the user information data set.

3. The homogeneous population recognition method based on the dual-index particle swarm algorithm of claim 2, wherein the "readable file format" comprises: csv format or bat format.

4. The homogeneous population identification method based on the dual-index particle swarm algorithm according to claim 2, wherein the information of the people using the public health service platform comprises: nationality, place of residence, age information.

5. The homogeneous population recognition method based on the dual-index particle swarm algorithm according to claim 1, wherein the two "initial adaptive values" are: fitness1(CH), Fitness2(S _ Dbw).

6. The homogeneous population recognition method based on the dual-index particle swarm algorithm of claim 5, wherein the Fitness1(CH) obtaining step is:

traces representing dispersion matrices in the categories;

the trace of the inter-class dispersion matrix is represented, and m represents the mean vector of the whole data set; n is the number of samples; and K is the iteration number.

7. The homogeneous population recognition method based on the dual-index particle swarm algorithm according to claim 5, wherein the Fitness2(S _ Dbw) obtaining step is:

S_Dbw(c)＝Scat(c)+Dens_bw(c)

wherein:

dens _ bw (c) to evaluate the relationship of the density of two classes together and the density of each individual class;

density (u) is used to characterize the number of points around u, and the threshold for comparison is stdev in 1;

stdev represents the average deviation of classes of a data set;

scater (c) represents the mean dispersion between the classes.

8. The homogeneous population identification method based on the dual-index particle swarm algorithm according to claim 5, wherein the process of obtaining the homogeneous population information data by iterating the two initial adaptive values as adaptive functions to obtain clustering results is as follows:

a. initializing various parameters of a particle swarm algorithm:

b. respectively calculating an adaptive value Fitness1 and an adaptive value Fitness2 of each particle according to CH and S _ Dbw index formulas;

c. for each particle, comparing the adaptive value Fitness1 of the current position with the adaptive value Fitness2 and two adaptive values corresponding to the historical optimal position of the particle, if the two adaptive values of the current position are higher, updating the historical optimal position by using the current position, otherwise, not updating;

d. for each particle, comparing the adaptive value Fitness1 of the current position with the adaptive value Fitness2 and two adaptive values corresponding to the global optimal position of the particle, if the two adaptive values of the current position are higher, updating the global optimal position by using the current position, otherwise, not updating;

e. update of position and velocity of particles:

particle velocity update formula:

particle position update formula:

wherein, V_id ^kA d-dimension component representing a velocity vector of a particle i at the k-th iteration; x is the number of_i ^dkA d-dimension component representing a location vector of a particle i at the k-th iteration; c. C₁，c₂Represents an acceleration constant; r is₁，r₂Represents two random parameters with the value range of [0,1]](ii) a w represents an inertial weight;

f. if the end condition is not met, returning to the step b, if the end condition is met, ending the algorithm, and obtaining a global optimal position, namely a global optimal solution;

and finally, outputting a clustering result to obtain accurate homogeneous crowd information data.

Images