CN105117609A - Dynamic weighing method based on generalized K-Means classification decision - Google Patents

Abstract

The invention discloses a dynamic weighing method based on generalized K-Means classification decision. The method includes the steps that 1, algorithm data are initialized through the steps of establishing a data queue of a specific length, acquiring dynamic data in windows of a specific width, calculating the mean value of the dynamic data and adding the mean value into the queue as an initialization center point; 2, the windows are moved backwards in sequence, a new mean value is calculated and added into the queue, and existing values are covered based on the first-in first-out rule if the queue is full; 3, the mean value of data in the queue serves as a new class center point, and the similarity of the data in the queue is calculated; 4, if the similarity of the data is smaller than a threshold value, a data center value can be locked, and otherwise, the steps 2 and 3 are executed repeatedly. The correct weighing value of a dynamic weighing system with amplitude fluctuation can be effectively and accurately calculated in real time through the method, and the method is high in accuracy and robustness and good in real-time performance.

Description

A kind of method of the dynamic weighing based on popularized type K-Means categorised decision

Technical field

The present invention relates to a kind of method of dynamic weighing, especially based on the method for the dynamic weighing of popularized type K-Means categorised decision.

Background technology

Dynamic weighing, refers to when measurement data is unstable, estimates by certain algorithm weight of weighing really.This type systematic is usually used in the fields such as automobile is weighed, livestock is weighed.

K-Means decision making algorithm is the objective function clustering method based on prototype, and algorithm utilizes measuring similarity that one group of data is obtained K cluster centre by continuous iteration, and data is divided into K class group.

Dynamic data centering value-based algorithm conventional is at present mainly running mean algorithm.Running mean algorithm is divided into two kinds according to the large I of moving window, stationary window length and variable window degree, and it is specific as follows:

$\stackrel{\‾}{d}=h(d_1,d_2,...,d_K)=\frac{\underset{i=1}{\overset{K}{\Σ}}{d}_i}{K}$

Wherein, d ₁, d ₂..., d _kfor raw value, K is data amount check.

For stationary window Length algorithm, K is fixed value.This algorithm is applicable to the less dynamic data of fluctuating range, and real-time is good, but accuracy is low.

For variable window Length algorithm, K is variable.Often measure data, the value of K increases by 1.Along with the increase of K numerical value, data, close to stable, try to achieve dynamic data central value.This algorithm can be used for the larger dynamic data of fluctuating range, and accuracy is high, but time delay is longer, poor real.

Summary of the invention

Problem to be solved by this invention is to provide a kind of method of dynamic weighing, can in real time effectively and also high-accuracy estimate real center value.

For solving the problem, the present invention, from the angle of K-means decision making algorithm thought, proposes a kind of method of the dynamic weighing based on popularized type K-Means decision making algorithm, comprising:

Step 1, algorithm data initialization, comprising:

1) data queue of length-specific is set up;

2) obtain dynamic data in specific width window and ask for average, as initial center point, and add queue;

Step 2, window expand successively, add new raw data, ask for new mean value and add queue, if queue is full, then the principle following first in first out covers existing value;

In step 3, queue, data mean value is as new class central point, and asks data similarity in queue;

If step 4 data similarity is less than threshold value, lockable data center is worth, otherwise repeats step 2,3.

Preferably, in step 2, the specific algorithm asking for new mean value is:

${\stackrel{\‾}{u}}_{N+1}=f(N,{\stackrel{\‾}{u}}_N,d_{N+1})=\frac{{\stackrel{\‾}{u}}_N*N+d_{N+1}}{N+1}$

Wherein, d _n+1for the raw data newly added, for the average of N number of data that previous step has been tried to achieve.

Preferably, in step 3, the specifying information of the balancing method of internal data similarity is:

$dis\tan ce=g({\stackrel{\‾}{u}}_1,{\stackrel{\‾}{u}}_2,...,{\stackrel{\‾}{u}}_K)=\sqrt{\frac{\underset{i=1}{\overset{K}{\Σ}}{({\stackrel{\‾}{u}}_i-\stackrel{\‾}{\stackrel{\‾}{u}})}^2}{K-1}}$

$\stackrel{\‾}{\stackrel{\‾}{u}}=h({\stackrel{\‾}{u}}_1,{\stackrel{\‾}{u}}_2,...,{\stackrel{\‾}{u}}_K)=\frac{\underset{i=1}{\overset{K}{\Σ}}{\stackrel{\‾}{u}}_i}{K}$

Wherein, K is queue length, for data in queue.

The present invention is based on the method for the dynamic weighing of popularized type K-Means decision making algorithm, be compared to traditional running mean estimating algorithm, the filter thought discarded tradition, from K-Means categorised decision algorithm idea, all data are divided into a class, and the target of algorithm is stretched to by classification and solves class center.The present invention has taken into account time efficiency and accuracy two aspects, can go out central value data accurately with shorter data estimation, can be used for the central value estimation with the dynamic data fluctuated widely.

Accompanying drawing explanation

Fig. 1 is the raw data distribution plan of first group of data in the preferred embodiment of the present invention;

Fig. 2 is the dynamic weighing data profile obtained after first group of data acquisition running mean algorithm in the preferred embodiment of the present invention;

Fig. 3 is the dynamic weighing data profile obtained after first group of data acquisition algorithm of the present invention in the preferred embodiment of the present invention;

Fig. 4 is the raw data distribution plan of second group of data in the preferred embodiment of the present invention;

Fig. 5 is the dynamic weighing data profile obtained after second group of data acquisition running mean algorithm in the preferred embodiment of the present invention;

Fig. 6 is the dynamic weighing data profile obtained after second group of data acquisition algorithm of the present invention in the preferred embodiment of the present invention.

Embodiment

Illustrate that the present invention is further detailed explanation with simulation process below in conjunction with accompanying drawing:

The present invention is based on the method for the dynamic weighing of popularized type K-Means decision making algorithm, this algorithm make use of the Clustering of K-Means, utilizes the iteration of cluster and decision process to ask for class center, in this, as the central value of dynamic data.The concrete steps of this algorithm are as follows:

1. algorithm data initialization, comprising:

1) data queue that length is K is set up;

2) specific width window N is obtained ₀interior dynamic data asks for average as initial center point, and add queue;

${\stackrel{\‾}{u}}_1=\frac{d_1+d_2+...+d_{N_0}}{N_0}$

2. window expands successively, adds new raw data d _i, ask for new mean value and add queue if queue is full, then the principle following first in first out covers existing value;

Data mean value in step 3, queue as new class central point, and ask data similarity distance in queue;

$\stackrel{\‾}{\stackrel{\‾}{u}}=h({\stackrel{\‾}{u}}_1,{\stackrel{\‾}{u}}_2,...,{\stackrel{\‾}{u}}_K)=\frac{\underset{i=1}{\overset{K}{\Σ}}{\stackrel{\‾}{u}}_i}{K}$

$dis\tan ce=g({\stackrel{\‾}{u}}_1,{\stackrel{\‾}{u}}_2,...,{\stackrel{\‾}{u}}_K)=\sqrt{\frac{\underset{i=1}{\overset{K}{\Σ}}{({\stackrel{\‾}{u}}_i-\stackrel{\‾}{\stackrel{\‾}{u}})}^2}{K-1}}$

If step 4 data similarity is less than threshold value, lockable data center is worth, otherwise repeats step 2,3.

As the simulation result that Fig. 1-6 is two groups of data, as described in illustrate, according to weighing system, general 1s obtains the actual time delay demand of 10 data, gets moving window initial size N here ₀=20, queue length K=10.It is comparatively large that this emulates data fluctuations used, and when namely virtual center value is D, fluctuation data can fluctuation in the scope of [0,2D].As figure, contrast visible, running mean algorithm can only obtain the central value data of certain error, and when dynamic data fluctuation is larger, the central value that running mean algorithm is estimated also has certain fluctuation, and algorithm of the present invention can obtain comparatively stable central value data.

As seen from the figure, when fluctuating larger, the present invention may occur the state (diagram is shown as and recovers null value) that data cannot lock but algorithm of the present invention can relock central value data in 2s time range, meets the demand that practical application is immediately stable.

The above, be only present pre-ferred embodiments, therefore can not limit scope of the invention process according to this, the equivalence change namely done according to the scope of the claims of the present invention and description with modify, all should still belong in scope that the present invention contains.

Claims (3)

1., based on a method for the dynamic weighing of popularized type K-Means categorised decision, it is characterized in that comprising:

Step 1, algorithm data initialization, comprising:

1) data queue of length-specific is set up;

2) obtain dynamic data in specific width window and ask for average, as initial center point, and add queue;

Step 2, window expand successively, add new raw data, ask for new mean value and add queue, if queue is full, then the principle following first in first out covers existing value;

In step 3, queue, data mean value is as new class central point, and asks data similarity in queue;

If step 4 data similarity is less than threshold value, lockable data center is worth, otherwise repeats step 2,3.

2., as claimed in claim 1 based on the method for the dynamic weighing of popularized type K-Means categorised decision, it is characterized in that: in step 2, the specific algorithm asking for new mean value is:

${\stackrel{\‾}{u}}_{N+1}=f(N,{\stackrel{\‾}{u}}_N,d_{N+1})=\frac{{\stackrel{\‾}{u}}_N*N+d_{N+1}}{N+1}$

Wherein, d _n+1for the raw data newly added, for the average of N number of data that previous step has been tried to achieve.

3., as claimed in claim 1 based on the method for the dynamic weighing of popularized type K-Means categorised decision, it is characterized in that: in step 3, the specifying information of the balancing method of internal data similarity is:

$dis\tan ce=g({\stackrel{\‾}{u}}_1,{\stackrel{\‾}{u}}_2,...,{\stackrel{\‾}{u}}_K)=\sqrt{\frac{\underset{i=1}{\overset{K}{\Σ}}{({\stackrel{\‾}{u}}_i-\stackrel{\‾}{\stackrel{\‾}{u}})}^2}{K-1}}$

$\stackrel{\‾}{\stackrel{\‾}{u}}=h({\stackrel{\‾}{u}}_1,{\stackrel{\‾}{u}}_2,...,{\stackrel{\‾}{u}}_K)=\frac{\underset{i=1}{\overset{K}{\Σ}}{\stackrel{\‾}{u}}_i}{K}$

Wherein, K is queue length, for data in queue.