CN105975744A - D-S evidence theory-based textile process data fusion system - Google Patents

Abstract

The invention discloses a D-S evidence theory-based textile process data fusion system. The system comprises sensor modules used for acquiring workshop data, local decision-making modules and a D-S synthesis module, wherein the sensor modules and the local decision-making modules are connected in a one-to-one correspondence manner; the local decision-making modules are connected with the D-S synthesis module; each local decision-making module adopts an adaptive weighted data fusion algorithm; and the D-S synthesis module adopts a D-S evidence theory. According to the system, two stages of sensors are adopted for information fusion; the first stage of the sensors are used for local fusion, a classic adaptive weighted fusion estimation algorithm is adopted, and the uncertainty and limitation of a single sensor are overcome, so that consistent explaination and description of a tested object are obtained; the second stage of the sensors are used for global fusion, and a D-S evidence theory is adopted; and the D-S evidence theory allows people to perform modeling on uncertain problems and perform reasoning, so that the uncertainty of things can be reflected more objectively.

Description

A kind of fabrication processes data fusion system based on D-S evidence theory

Technical field

The present invention relates to data monitoring technical field of weaving, be specifically related to a kind of fabrication processes data based on D-S evidence theory Emerging system.

Background technology

Along with textile machine equipment automatization, networking and intelligentized development, whole weaving manufacture process is with unprecedented Speed produce the process equipment of magnanimity, production process and operational management data, the most also include controlling feeder number According to, the raw material of text type, sensing data, the unstructured data such as yarn defect inspection image data, and have between data There are higher-dimension, non-linear, strong correlation, and many noise behaviors, therefore fabrication processes data have that data volume is big, type is many, Real-time and be worth big feature, substantially possess the feature of big data " 4V ", be one typical " big data ".

Under this " big data " environment, weaving manufacturing execution system is again a kind of non-linear, multi-variable system of time-varying, Making to be often accompanied by immesurable uncertain factor in production process data, this disturbance is easily caused the multiplication of data volume, And along with the raising of application precision is then incremented by geometry level so that existing model of Information Integration and method are difficult to reply should " big data ", the correctness ultimately resulting in fabrication processes data fusion result is difficult to ensure that.Therefore, how to these seas Amount weaving data carry out Formal Representation and data fusion reason, thus are advantageously implemented effective integration and the pipe of mass data Reason, the weaving manufacturing execution system being more beneficial for building under big data environment is realistic problem urgently to be resolved hurrily.

Summary of the invention

For solving the problems referred to above, the invention provides a kind of fabrication processes data fusion system based on D-S evidence theory, choosing Select two or more sensor group to detect the situation of anomalous event, then use two-stage sensor data fusion, One-level is that local (i.e. pixel level) is merged, and it uses the adaptive weighted Fusion Estimation Algorithm of classics, overcomes single biography The uncertainty of sensor and limitation, thus the concordance obtaining measurand is explained and describes.Two grades is in the overall situation (i.e. Decision-making level) merge, use Dempster-Shafer (D-S) evidence theory.Owing to D-S evidence theory allows people Uncertain problem is modeled, and make inferences such that it is able to more objectively reflect the uncertainty of things.

For achieving the above object, the technical scheme that the present invention takes is:

A kind of fabrication processes data fusion system based on D-S evidence theory, including the sensor for gathering each workshop data Module, local decision module and D-S synthesis module, sensor assembly arranges with local decision-making module one_to_one corresponding and is connected, Described local decision module is connected with D-S synthesis module, and described local decision module uses adaptive weighting data fusion to calculate Method, not only can optimize the data of each workshop appliance sensor, additionally it is possible to effectively reject environmental disturbances signal, its center Thought is according to each sensing data by mistake extent, distributes different flexible strategy, and the high data of precision are little due to error, Distributional weight is relatively big, otherwise less；Described D-S synthesis module uses Dempster-Shafer (D-S) evidence theory.

Preferably, described adaptive weighting data fusion algorithm specifically includes following steps:

S1, being provided with n sensor to detect the anomalous event that certain uncertain factor causes, their variance is respectively The measured value of each sensor is respectively x₁, x₂..., x_n, and it is separate；Assuming that each sensor Weighter factor be respectively w₁, w₂..., w_n, then after weighter factor introduces, the Data Fusion of Sensor value of system is:In formulaTotal mean square deviation is:

${\mathrm{\δ}}^2=E\[{(x-\hat{x})}^2\]=E\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-\hat{x})}^2+2E\underset{i=1,j=1,i=j}{\overset{n}{\mathrm{\Σ}}}{w}_i{w}_j(x-{\hat{x}}_i)(x-{\hat{x}}_j);$

Due to x₁, x₂..., x_nIt is mutually independent, and is the unbiased esti-mator of x, therefore:

And i ≠ j；I, j=1,2 ..., n；

Then have:

In formula, δ is each weighter factor w_iMultiple quadratic function, asking for of its minima is exactly at weighter factor w₁, w₂..., w_nMeet asking for of function of many variables extreme value under normalization constraints.

S2, basis seek extreme value theory, when weighter factor is:

Then have:

It is above the estimation carried out according to each sensor data acquisition set value at a time, when estimating that true value x is constant Time, then can estimate according to the average of each sensor historic data；

S3, set:Q=1,2 ..., n；

Estimated value now isOverall mean square error is:

${\stackrel{\‾}{\mathrm{\δ}}}^2=E\[{(x-\widehat{\stackrel{\‾}{x}})}^2\]=E\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-{\stackrel{\‾}{x}}_{iw})}^2+2E\underset{i=1,j=1,i=j}{\overset{n}{\mathrm{\Σ}}}{w}_i{w}_j(x-{\stackrel{\‾}{x}}_i)(x-x_{jw});$

Because x₁, x₂..., x_nFor unbiased esti-mator, therefore:

${\stackrel{\‾}{\mathrm{\δ}}}^2=E\[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-{\stackrel{\‾}{x}}_{jw})}^2\]=\frac{1}{k}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{{\mathrm{\σ}}_i}^2=\frac{{{\mathrm{\σ}}^2}_{\min }}{k};$

Obviously,And along with the increase of k, δ is gradually reduced.

Preferably, the construction process of the Basic probability assignment function in Dempster-Shafer (D-S) evidence theory is as follows:

S1, distance and the structure of relativity measurement relation

If the characteristic vector of evidence body (sensor) is X_i, target (monitoring type) A_jMaster sample characteristic vector be Y_j. then Both Manhattan distances are: d_ij(X_i, Y_j)=∑ | X_ik-Y_ik|；

From above formula, distance d_ij(X_i, Y_j) the biggest, then evidence body i and target A_jDegree of correlation get over the end；Otherwise, distance d_ij(X_i, Y_j) the least, then evidence body i and target A_jDegree of correlation the highest；Therefore, definition: C_i(A_j)=1/d_ij(X_i, Y_j), Then evidence body with the maximum correlation of target is:

α_i=max{C_i(A_j)=1/min{d_ij(X_i, Y_j)}；

Evidence body i with the breadth coefficient of each target correlation coefficient is:

In formula, N is target type number to be measured；

The safety factor of evidence body (sensor) i is:

S2, Basic probability assignment function construct

Formula in comprehensive S1 obtains evidence body i and gives target A_jBasic probability assignment and give identification framework Θ the most general Rate is distributed, i.e. the computational methods of the uncertain probit of sensor are as follows:

$m_i\left(A_j\right)=\frac{w_i{C}_i\left(A_j\right)}{\underset{i}{\mathrm{\Σ}}{w}_i{C}_i\left(A_j\right)+N_s(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}---\left(1\right)$

$m_i\left(\mathrm{\θ}\right)=\frac{N_s(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}{\underset{j}{\mathrm{\Σ}}{w}_i{C}_i\left(A_j\right)+N_c(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}---\left(2\right)$

In formula, N_sFor number of sensors, w_iFor weight coefficient, and according to correlation coefficient C_i(A_j) size value, and 0 ＜ w_i＜ 1；

Analysis mode (1) and (2), first, it meets the definition condition of mass function, and is calculating evidence body (sensor) During to the basic probability assignment of target (including Θ), set up the spacing of evidence body (sensor) and target (monitoring type) With corresponding relation and the distribution of dependency of dependency, and sensor safety factor；Then, according to the phase of evidence body with target Guan Xing, introduces weight coefficient, carries out the output of the result of decision, effectively enhances the correctness of the result of decision.

The method have the advantages that

Select two or more sensor group to detect the situation of anomalous event, then use two-stage sensor information Merging, one-level is that local (i.e. pixel level) is merged, and it uses the adaptive weighted Fusion Estimation Algorithm of classics, overcomes The uncertainty of single sensor and limitation, thus the concordance obtaining measurand is explained and describes.Two grades be The overall situation (i.e. decision-making level) merges, and uses Dempster-Shafer (D-S) evidence theory.Owing to D-S evidence is managed Opinion allows people to be modeled uncertain problem, and makes inferences such that it is able to more objectively reflection things is not Definitiveness.

Accompanying drawing explanation

Fig. 1 is the system block diagram of a kind of fabrication processes data fusion system based on D-S evidence theory of the embodiment of the present invention.

Detailed description of the invention

In order to make objects and advantages of the present invention clearer, below in conjunction with embodiment, the present invention is carried out the most specifically Bright.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.

As it is shown in figure 1, embodiments provide a kind of fabrication processes data fusion system based on D-S evidence theory, bag Include for gathering the sensor assembly of each workshop data, local decision module and D-S synthesis module, sensor assembly drawn game Portion's decision-making module one_to_one corresponding arranges connected, and described local decision module is connected with D-S synthesis module, described local decision Module uses adaptive weighting data fusion algorithm, not only can optimize the data of each workshop appliance sensor, additionally it is possible to have Effect rejects environmental disturbances signal, and its central idea is according to each sensing data extent by mistake, distributes different power Number, the high data of precision are little due to error, and distributional weight is relatively big, otherwise less；Described D-S synthesis module uses Dempster-Shafer (D-S) evidence theory.

Described adaptive weighting data fusion algorithm specifically includes following steps:

S1, being provided with n sensor to detect the anomalous event that certain uncertain factor causes, their variance is respectively The measured value of each sensor is respectively x₁, x₂..., x_n, and it is separate；Assuming that each sensor Weighter factor be respectively w₁, w₂..., w_n, then after weighter factor introduces, the Data Fusion of Sensor value of system is:

In formulaTotal mean square deviation is:

${\mathrm{\δ}}^2=E\[{(x-\hat{x})}^2\]=E\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-\hat{x})}^2+2E\underset{i=1,j=1,i=j}{\overset{n}{\mathrm{\Σ}}}{w}_i{w}_j(x-{\hat{x}}_i)(x-{\hat{x}}_j).$

Due to x₁, x₂..., x_nIt is mutually independent, and is the unbiased esti-mator of x, therefore:

And i ≠ j；I, j=1,2 ..., n.

Then have:

In formula, δ is each weighter factor w_iMultiple quadratic function, asking for of its minima is exactly at weighter factor w₁, w₂..., w_nMeet asking for of function of many variables extreme value under normalization constraints.

S2, basis seek extreme value theory, when weighter factor is:

Then have:

It is above the estimation carried out according to each sensor data acquisition set value at a time, when estimating that true value x is constant Time, then can estimate according to the average of each sensor historic data；

S3, set:Q=1,2 ..., n；

Estimated value now isOverall mean square error is:

${\stackrel{\‾}{\mathrm{\δ}}}^2=E\[{(x-\widehat{\stackrel{\‾}{x}})}^2\]=E\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-{\stackrel{\‾}{x}}_{iw})}^2+2E\underset{i=1,j=1,i=j}{\overset{n}{\mathrm{\Σ}}}{w}_i{w}_j(x-{\stackrel{\‾}{x}}_i)(x-x_{jw}).$

Because x₁, x₂..., x_nFor unbiased esti-mator, therefore:

${\stackrel{\‾}{\mathrm{\δ}}}^2=E\[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-{\stackrel{\‾}{x}}_{jw})}^2\]=\frac{1}{k}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{{\mathrm{\σ}}_i}^2=\frac{{{\mathrm{\σ}}^2}_{\min }}{k}.$

Obviously,And along with the increase of k, δ is gradually reduced.

According to being defined below of D-S evidence theory:

Definition 1: set Θ as identification framework, if set function m:2^Θ→ [0,1] (2^ΘPower set for Θ) meet ∑ m (A)=1, then m is called the basic brief inference on identification framework Θ；M (A) is referred to as the basic certainty value of A, M (A) reflects the size of the reliability to A itself；

Definition 2: if m is a basic brief inference, thenThen determined The function Bel of justice is a belief function, and Bel (A) reflects the reliability that on A, all subsets are total；

If there is A to belong to identification framework Θ, definition Dou (A)=Bel (A), Pl (A)=1-Bel (A), then Pl (A) is called The likelihood function of Bel, Pl (A) is called the likelihood degree of A, i.e. describes the likelihood of A or reliable degree；Dou is the bosom of Bel Doubting function, Dou (A) is the suspicious degree of A, and describe A does not be sure of degree；It practice, [Bel (A), Pl (A)] illustrates The bound of the indeterminacy section of A, i.e. probability；

In D-S evidence theory, owing to there is no Basic Probability As-signment (Basic Probability Assignment, BPA) function Specific descriptions, it is generally the case that mainly rely on expert to specify, or obtain Basic Probability As-signment according to certain experience, Therefore there is bigger subjective randomness.To this end, for the uncertainty of process of textile production,

The construction process of the Basic probability assignment function in Dempster-Shafer (D-S) evidence theory is as follows:

S1, distance and the structure of relativity measurement relation

If the characteristic vector of evidence body (sensor) is X_j, target (monitoring type) A_jMaster sample characteristic vector be Y_j.Then Both Manhattan distances are: d_ij(X_i, Y_j)=∑ | X_ik-Y_ik|；

From above formula, distance d_ij(X_i, Y_j) the biggest, then evidence body i and target A_jDegree of correlation get over the end；Otherwise, distance d_ij(X_j, Y_j) the least, then evidence body i and target A_jDegree of correlation the highest；Therefore, definition: C_i(A_j)=1/d_ij(X_i, Y_j), Then evidence body with the maximum correlation of target is:

α_i=max{C_i(A_j)=1/min{d_ij(X_i, Y_j)}。

Evidence body i with the breadth coefficient of each target correlation coefficient is:

In formula, N is target type number to be measured；

The safety factor of evidence body (sensor) i is:

S2, Basic probability assignment function construct

Formula in comprehensive S1 obtains evidence body i and gives target A_jBasic probability assignment and give identification framework Θ the most general Rate is distributed, i.e. the computational methods of the uncertain probit of sensor are as follows:

$m_i\left(A_j\right)=\frac{w_i{C}_i\left(A_j\right)}{\underset{i}{\mathrm{\Σ}}{w}_i{C}_i\left(A_j\right)+N_s(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}---\left(1\right)$

$m_i\left(\mathrm{\θ}\right)=\frac{N_s(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}{\underset{j}{\mathrm{\Σ}}{w}_i{C}_i\left(A_j\right)+N_c(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}---\left(2\right)$

In formula, N_sFor number of sensors, w_iFor weight coefficient, and according to correlation coefficient C_i(A_j) size value, and 0 ＜ w_i＜ 1；

Analysis mode (1) and (2), first, it meets the definition condition of mass function, and is calculating evidence body (sensor) During to the basic probability assignment of target (including Θ), set up the spacing of evidence body (sensor) and target (monitoring type) With corresponding relation and the distribution of dependency of dependency, and sensor safety factor；Then, according to the phase of evidence body with target Guan Xing, introduces weight coefficient, carries out the output of the result of decision, effectively enhances the correctness of the result of decision.

The above is only the preferred embodiment of the present invention, it is noted that for those skilled in the art, Without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications also should be regarded as the present invention Protection domain.

Claims (3)

1. a fabrication processes data fusion system based on D-S evidence theory, it is characterised in that include for gathering each The sensor assembly of workshop data, local decision module and D-S synthesis module, sensor assembly and local decision-making module one One is correspondingly arranged connected, and described local decision module is connected with D-S synthesis module, and described local decision module uses adaptive Answering weighted fusion algorithm algorithm, described D-S synthesis module uses Dempster-Shafer (D-S) evidence theory.

A kind of fabrication processes data fusion system based on D-S evidence theory the most according to claim 1, its feature exists Following steps are specifically included in, described adaptive weighting data fusion algorithm:

S1, being provided with n sensor to detect the anomalous event that certain uncertain factor causes, their variance is respectively The measured value of each sensor is respectively x₁, x₂..., x_n, and it is separate；Assuming that each sensor Weighter factor be respectively w₁, w₂..., w_n, then after weighter factor introduces, the Data Fusion of Sensor value of system is:

In formulaTotal mean square deviation is:

${\mathrm{\δ}}^2=E\[{(x-\hat{x})}^2\]=E\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-\hat{x})}^2+2E\underset{i=1,j=1,i=j}{\overset{n}{\mathrm{\Σ}}}{w}_i{w}_j(x-{\hat{x}}_i)(x-{\hat{x}}_j);$

Due to x₁, x₂..., x_nIt is mutually independent, and is the unbiased esti-mator of x, therefore:

And i ≠ j；I, j=1,2 ..., n；

Then have:

In formula, δ is each weighter factor w_iMultiple quadratic function；

S2, basis seek extreme value theory, when weighter factor is:

Then have:

S3, set:Q=1,2 ..., n；

Estimated value now isOverall mean square error is:

${\stackrel{\‾}{\mathrm{\δ}}}^2=E\[{(x-\widehat{\stackrel{\‾}{x}})}^2\]=E\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-{\stackrel{\‾}{x}}_{iw})}^2+2E\underset{i=1,j=1,i=j}{\overset{n}{\mathrm{\Σ}}}{w}_i{w}_j(x-{\stackrel{\‾}{x}}_i)(x-x_{jw});$

Because x₁, x₂..., x_nFor unbiased esti-mator, therefore:

${\stackrel{\‾}{\mathrm{\δ}}}^2=E\[\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{(x-{\stackrel{\‾}{x}}_{jw})}^2\]=\frac{1}{k}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{w_i}^2{{\mathrm{\σ}}_i}^2=\frac{{{\mathrm{\σ}}^2}_{min}}{k};$

Obviously,And along with the increase of k, δ is gradually reduced.

A kind of fabrication processes data fusion system based on D-S evidence theory the most according to claim 1, its feature Being, the construction process of the Basic probability assignment function in Dempster-Shafer (D-S) evidence theory is as follows:

S1, distance and the structure of relativity measurement relation

If the characteristic vector of evidence body (sensor) is X_i, target (monitoring type) A_jMaster sample characteristic vector be Y_j. then Both Manhattan distances are: d_ij(X_i, Y_j)=∑ | X_ik-Y_ik|；

From above formula, distance d_ij(X_i, Y_j) the biggest, then evidence body i and target A_jDegree of correlation get over the end；Otherwise, distance d_ij(X_i, Y_j) the least, then evidence body i and target A_jDegree of correlation the highest；Therefore, definition: C_i(A_j)=1/d_ij(X_i, Y_j), Then evidence body with the maximum correlation of target is:

α_i=max{C_i(A_j)=1/min{d_ij(X_i, Y_j)}；

Evidence body i with the breadth coefficient of each target correlation coefficient is:

In formula, N is target type number to be measured；

The safety factor of evidence body (sensor) i is:

S2, Basic probability assignment function construct

Formula in comprehensive S1 obtains evidence body i and gives target A_jBasic probability assignment and give identification framework Θ the most general Rate is distributed, i.e. the computational methods of the uncertain probit of sensor are as follows:

$m_i\left(A_j\right)=\frac{w_i{C}_i\left(A_j\right)}{\underset{i}{\mathrm{\Σ}}{w}_i{C}_i\left(A_j\right)+N_s(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}---\left(1\right)$

$m_i\left(\mathrm{\θ}\right)=\frac{N_s(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}{\underset{j}{\mathrm{\Σ}}{w}_i{C}_i\left(A_j\right)+N_c(1-R_i)(1-{\mathrm{\α}}_i{\mathrm{\β}}_j)}---\left(2\right)$

In formula, N_sFor number of sensors, w_iFor weight coefficient, and according to correlation coefficient C_i(A_j) size value, and 0 ＜ w_i＜ 1；

Analysis mode (1) and (2), first, it meets the definition condition of mass function, and is calculating evidence body (sensor) During to the basic probability assignment of target (including Θ), set up the spacing of evidence body (sensor) and target (monitoring type) With corresponding relation and the distribution of dependency of dependency, and sensor safety factor；Then, according to the phase of evidence body with target Guan Xing, introduces weight coefficient, carries out the output of the result of decision.