CN104346442A - Process object data-oriented rule extracting method - Google Patents

Abstract

The invention relates to a process object data-oriented rule extracting method. The process object data-oriented rule extracting method comprises the following steps: S1: determining the optimal clustering number of process object data; S2: clustering the process object data by using a K-meAns algorithm, and verifying the reasonability of the optimal clustering number in the step S1, if the optimal clustering number in the step S1 is reasonable, going to S3, otherwise, going to S1; S3: excavating association rules between clusters at different measuring points by using an Apriori inter-dimension association rule algorithm; S4: determining the strongest association chain of the process object data; S5: according to status values of all measuring points on the strongest association chain, obtaining a status chain recording the status values of the measuring points, and guiding related industries according to the status chain. By the process object data-oriented rule extracting method, the data rule extracting efficiency and the capability of extracting knowledge from the process object data are improved.

Description

A kind of Rules extraction method of Process-Oriented object data

Technical field

The invention belongs to Data Analysis Services technical field, relate to a kind of Rules extraction method of data, especially a kind of Rules extraction method of Process-Oriented object data; Improve the efficiency that data rule extracts, and from the ability of flow object extracting data knowledge.

Background technology

Along with the development of large data technique, utilize large data to carry out Knowledge Discovery to pay close attention to by people, its major reason is because knowledge is more and more abundanter along with the accumulation of data, utilizes these data to be optimized Industry Control flow process, assists the demand of control decision constantly to increase; In Industry Control, single process control optimization means make us lack according to going more effectively to improve procedure parameter aborning, be difficult to further refinement control strategy, optimal control parameter, therefore how leaching process parameter and controling parameters to carry out arranging, extracting flow object running status Sum fanction and be used from a large amount of historical datas be the key of dealing with problems from flow object; And the process data accumulated in process industry control object, there is stronger procedure parameter complicacy, High relevancy, the non-linear and sequential of data variation and the inconsistency of sampling, find to bring difficulty to data mining Sum fanction, the operation efficiency of existing rule discovery algorithm in process too many levels, Complicated Flow object is on the low side, and the result of calculation obtained is difficult to embody the auxiliary optimization function to process industry.Therefore existing large data rule discover method is improved, make it extract flow object running status Sum fanction in a large amount of historical data of from flow object leaching process parameter and controling parameters, utilization and be used, become present stage large data processing field problem in the urgent need to address.

Summary of the invention

The object of the invention is to, for the defect existed in above-mentioned prior art, provide the Rules extraction method designing a kind of Process-Oriented object data, to solve the problems of the technologies described above.

For achieving the above object, the present invention provides following technical scheme:

A Rules extraction method for Process-Oriented object data, comprises the steps:

Step S1: determine that the best of flow object data clusters quantity;

Step S2: adopt K-means algorithm to carry out cluster to flow process object data, the best in simultaneous verification step S1 clusters the rationality of quantity, and the reasonable quantity if the best in step S1 clusters, forwards step S3 to, otherwise forward step S1 to;

Step S3: adopt association rule algorithm between Apriori dimension to excavate the correlation rule between the cluster of different measuring points;

Step S4: that determines flow object data associates the most by force chain;

Step S5: according to the state value associating the most by force all measuring points on chain, obtain the state chain recording each measuring point state value, according to state chain, relevant industries are instructed.

Preferably, described step S1 comprises the steps:

Step S101: get time period T _m, any point X in flow object data after the sequence that clocks adjustment _jvariable quantity be D _i, wherein, $D_i={\mathrm{\Σ}}_{j=t_{u_M}}^{t_{u_M+b-1}}\left|{\mathrm{\Δx}}_i\left(t_j\right)\right|;$

Step S102: obtain T _mthe measuring point that interior variable quantity is maximum, note D _m=max{D _i, i=1,2 ..., n}, obtains T _mthe measuring point X that interior variable quantity is maximum _m;

Step S103: to the measuring point X obtained in step S102 _mcarry out optimum k value calculating;

Step S104: by the measuring point X obtained in step S102 _mcarry out cluster, and result evaluation function is evaluated, determine optimum k value.

Preferably, described step S2 comprises the steps:

Step S201: adopt K-means algorithm to carry out cluster to flow process object data, clustering formula is: wherein, p is C _iin object, m _ifor a bunch C _iaverage;

Step S202: the rationality adopting the quantity that to cluster based on the best in the silhouette coefficient verification step S1 of condensation degree and degree of separation;

Step S203: the rationality of the best quantity that clusters in determining step S202, if rationally, forward step S3 to, otherwise forward step S1 to.

Preferably, the concrete steps of described step S3 are as follows:

If any two measuring point X _iand X _jcorrelation rule between any two clusters between (i, j ∈ { 1,2...n}, i ≠ j) is X _icluster k _ia→ X _jcluster k _jb, by X _icluster k _ia→ X _jcluster k _jbbe designated as ia → jb, the support of this correlation rule is S (ia → jb), $S(\mathrm{ia}\→\mathrm{jb})=P(\mathrm{ia},\mathrm{jb})=\frac{|\mathrm{ia},\mathrm{jb}|}{\left|T\right|},$

Wherein, | ia, jb| are data centralization X _icluster k _iaand X _jcluster k _jbsimultaneous affairs number, | T| represents all affairs sums, and the interest-degree of this rule is I (ia → jb),

Wherein, the degree of confidence that C (ia → jb) is this rule, S (jb) is regular consequent X _jcluster k _jbsupport, as I (ia → jb)=1, represent X _icluster k _iaand X _jcluster k _jbseparate; As I (ia → jb) > 1, represent X _icluster k _iaand X _jcluster k _jbbe positively related, I (ia → jb) is larger, and X is described _icluster k _iato X _jcluster k _jbfacilitation is larger; As I (ia → jb) < 1, represent X _icluster k _iaand X _jcluster k _jbbe negative correlation, I (ia → jb) is less, and X is described _icluster k _iato X _jcluster k _jbinhibiting effect is larger.

Preferably, described step S4 comprises the steps:

Step S401: choose any point as first node;

Step S402: be find the maximum and consequent rule not repeating chain has existed measuring point of the degree of association in all measuring point rules of this first node with preceding paragraph, and using consequent for this rule as next node;

Step S403: whether the next node determined in determining step S402 is last node of this chain, if it is forwards step S404 to, if not, then using this next node as new first node, forward step S402 to;

Step S404: stop circulation, obtain associating chain the most by force.

Preferably, described step S4 also comprises the steps:

Node headed by all nodes, what obtain headed by all nodes node associates chain the most by force.

Preferably, described step S4 also comprises the steps:

Step S45: judge whether all chains that associates the most by force comprise all measuring points, if so, then forward step S5 to, if not, then forward step S46 to;

Step S46: structure relevance tree, each associates the most by force the branch of chain as described relevance tree.

Preferably, described step S202 specifically comprises the steps:

Step S2021: get i-th data point, calculates the mean distance a of every other data point in this data point to its place bunch _i;

Step S2022: the mean distance calculating all data points in i-th data point to other bunches, and remember that value minimum in all mean distances is b _i;

Step S2023: the silhouette coefficient calculating i-th data point, is designated as

Step S2024: calculate the silhouette coefficient that clusters number is the cluster set of K, be designated as k _m=min _k{ max _Ωs _k, wherein, s _kspan be [-1,1], work as s _ijduring > 0, represent that cluster result is better, and s _ijvalue more better close to 1 cluster result; Work as s _ijwhen≤0, represent that cluster result is poor, and s _ijvalue more poorer close to-1 cluster result.

Beneficial effect of the present invention is, is divided by node data by the state of measuring point, can realize the excavation of the correlation rule between measuring point, obtains the incidence relation between any point different conditions; Meanwhile, reduce data complexity, improve data operation speed; In addition, design concept of the present invention is reliable, has application prospect widely.

As can be seen here, the present invention compared with prior art, has outstanding substantive distinguishing features and progress significantly, and its beneficial effect implemented also is apparent.

Embodiment

Below by specific embodiment, the present invention will be described in detail, and following examples are explanation of the invention, and the present invention is not limited to following embodiment.

The Rules extraction method of a kind of Process-Oriented object data provided by the invention, comprises the steps:

Step S1: determine that the best of flow object data clusters quantity;

Step S2: adopt K-means algorithm to carry out cluster to flow process object data, the best in simultaneous verification step S1 clusters the rationality of quantity, and the reasonable quantity if the best in step S1 clusters, forwards step S3 to, otherwise forward step S1 to;

Step S3: adopt association rule algorithm between Apriori dimension to excavate the correlation rule between the cluster of different measuring points;

Step S4: that determines flow object data associates the most by force chain;

Step S5: according to the state value associating the most by force all measuring points on chain, obtain the state chain recording each measuring point state value, according to state chain, relevant industries are instructed.

In the present embodiment, described step S1 comprises the steps:

Step S101: get time period T _m, any point X in flow object data after the sequence that clocks adjustment _jvariable quantity be D _i, wherein, $D_i={\mathrm{\&Sigma;}}_{j=t_{u_M}}^{t_{u_M+b-1}}\left|{\mathrm{\&Delta;x}}_i\left(t_j\right)\right|;$

Step S102: obtain T _mthe measuring point that interior variable quantity is maximum, note D _m=max{D _i, i=1,2 ..., n}, obtains T _mthe measuring point X that interior variable quantity is maximum _m;

Step S103: to the measuring point X obtained in step S102 _mcarry out optimum k value calculating;

Step S104: by the measuring point X obtained in step S102 _mcarry out cluster, and result evaluation function is evaluated, determine optimum k value.

In the present embodiment, described step S2 comprises the steps:

Step S201: adopt K-means algorithm to carry out cluster to flow process object data, clustering formula is: wherein, p is C _iin object, m _ifor a bunch C _iaverage;

Step S202: the rationality adopting the quantity that to cluster based on the best in the silhouette coefficient verification step S1 of condensation degree and degree of separation;

Step S203: the rationality of the best quantity that clusters in determining step S202, if rationally, forward step S3 to, otherwise forward step S1 to.

In the present embodiment, the concrete steps of described step S3 are as follows:

If any two measuring point X _iand X _jcorrelation rule between any two clusters between (i, j ∈ { 1,2...n}, i ≠ j) is X _icluster k _ia→ X _jcluster k _jb, by X _icluster k _ia→ X _jcluster k _jbbe designated as ia → jb, the support of this correlation rule is S (ia → jb), $S(\mathrm{ia}\&RightArrow;\mathrm{jb})=P(\mathrm{ia},\mathrm{jb})=\frac{|\mathrm{ia},\mathrm{jb}|}{\left|T\right|},$

Wherein, | ia, jb| are data centralization X _icluster k _iaand X _jcluster k _jbsimultaneous affairs number, | T| represents all affairs sums, and the interest-degree of this rule is I (ia → jb),

Wherein, the degree of confidence that C (ia → jb) is this rule, S (jb) is regular consequent X _jcluster k _jbsupport, as I (ia → jb)=1, represent X _icluster k _iaand X _jcluster k _jbseparate; As I (ia → jb) > 1, represent X _icluster k _iaand X _jcluster k _jbbe positively related, I (ia → jb) is larger, and X is described _icluster k _iato X _jcluster k _jbfacilitation is larger; As I (ia → jb) < 1, represent X _icluster k _iaand X _jcluster k _ibbe negative correlation, I (ia → jb) is less, and X is described _icluster k _iato X _jcluster k _jbinhibiting effect is larger.

In the present embodiment, described step S4 comprises the steps:

Step S401: choose any point as first node;

Step S402: be find the maximum and consequent rule not repeating chain has existed measuring point of the degree of association in all measuring point rules of this first node with preceding paragraph, and using consequent for this rule as next node;

Step S403: whether the next node determined in determining step S402 is last node of this chain, if it is forwards step S404 to, if not, then using this next node as new first node, forward step S402 to;

Step S404: stop circulation, obtain associating chain the most by force.

In the present embodiment, described step S4 also comprises the steps:

Node headed by all nodes, what obtain headed by all nodes node associates chain the most by force.

In the present embodiment, described step S4 also comprises the steps:

Step S45: judge whether all chains that associates the most by force comprise all measuring points, if so, then forward step S5 to, if not, then forward step S46 to;

Step S46: structure relevance tree, each associates the most by force the branch of chain as described relevance tree.

In the present embodiment, described step S202 specifically comprises the steps:

Step S2021: get i-th data point, calculates the mean distance a of every other data point in this data point to its place bunch _i;

Step S2022: the mean distance calculating all data points in i-th data point to other bunches, and remember that value minimum in all mean distances is b _i;

Step S2023: the silhouette coefficient calculating i-th data point, is designated as

Step S2024: calculate the silhouette coefficient that clusters number is the cluster set of K, be designated as k _m=min _k{ max _Ωs _k, wherein, s _kspan be [-1,1], work as s _ijduring > 0, represent that cluster result is better, and s _ijvalue more better close to 1 cluster result; Work as s _ijwhen≤0, represent that cluster result is poor, and s _ijvalue more poorer close to-1 cluster result.

The preferred embodiment of the present invention is only above; but the present invention is not limited thereto; any those skilled in the art can think there is no creationary change, and some improvements and modifications done without departing from the principles of the present invention, all should drop in protection scope of the present invention.

Claims (8)

1. a Rules extraction method for Process-Oriented object data, comprises the steps:

Step S1: determine that the best of flow object data clusters quantity;

Step S2: adopt K-means algorithm to carry out cluster to flow process object data, the best in simultaneous verification step S1 clusters the rationality of quantity, and the reasonable quantity if the best in step S1 clusters, forwards step S3 to, otherwise forward step S1 to;

Step S3: adopt association rule algorithm between Apriori dimension to excavate the correlation rule between the cluster of different measuring points;

Step S4: that determines flow object data associates the most by force chain;

Step S5: according to the state value associating the most by force all measuring points on chain, obtain the state chain recording each measuring point state value, according to state chain, relevant industries are instructed.

2. the Rules extraction method of Process-Oriented object data according to claim 1, is characterized in that: described step S1 comprises the steps:

Step S101: get time period T _m, any point X in flow object data after the sequence that clocks adjustment _jvariable quantity be D _i, wherein,

Step S102: obtain T _mthe measuring point that interior variable quantity is maximum, note D _m=max{D _i,i=1,2 ..., n}, obtains T _mthe measuring point X that interior variable quantity is maximum _m;

Step S103: to the measuring point X obtained in step S102 _mcarry out optimum k value calculating;

Step S104: by the measuring point X obtained in step S102 _mcarry out cluster, and result evaluation function is evaluated, determine optimum k value.

3. the Rules extraction method of Process-Oriented object data according to claim 1 and 2, is characterized in that: described step S2 comprises the steps:

Step S201: adopt K-means algorithm to carry out cluster to flow process object data, clustering formula is: wherein, p is C _iin object, m _ifor a bunch C _iaverage;

Step S202: the rationality adopting the quantity that to cluster based on the best in the silhouette coefficient verification step S1 of condensation degree and degree of separation;

Step S203: the rationality of the best quantity that clusters in determining step S202, if rationally, forward step S3 to, otherwise forward step S1 to.

4. the Rules extraction method of Process-Oriented object data according to claim 3, is characterized in that: the concrete steps of described step S3 are as follows:

If any two measuring point X _iand X _jcorrelation rule between any two clusters between (i, j ∈ { 1,2...n}, i ≠ j) is X _icluster k _ia→ X _jcluster k _jb, by X _icluster k _ia→ X _jcluster k _jbbe designated as ia → jb, the support of this correlation rule is S (ia → jb),

Wherein, | ia, jb| are data centralization X _icluster k _iaand X _jcluster k _jbsimultaneous affairs number, | T| represents all affairs sums, and the interest-degree of this rule is I (ia → jb),

Wherein, the degree of confidence that C (ia → jb) is this rule, S (b) is regular consequent X _jcluster k _jbsupport.

5. the Rules extraction method of Process-Oriented object data according to claim 4, is characterized in that: described step S4 comprises the steps:

Step S401: choose any point as first node;

Step S402: be find the maximum and consequent rule not repeating chain has existed measuring point of the degree of association in all measuring point rules of this first node with preceding paragraph, and using consequent for this rule as next node;

Step S403: whether the next node determined in determining step S402 is last node of this chain, if it is forwards step S404 to, if not, then using this next node as new first node, forward step S402 to;

Step S404: stop circulation, obtain associating chain the most by force.

6. the Rules extraction method of Process-Oriented object data according to claim 5, is characterized in that: described step S4 also comprises the steps:

Node headed by all nodes, what obtain headed by all nodes node associates chain the most by force.

7. the Rules extraction method of Process-Oriented object data according to claim 6, is characterized in that: described step S4 also comprises the steps:

Step S45: judge whether all chains that associates the most by force comprise all measuring points, if so, then forward step S5 to, if not, then forward step S46 to;

Step S46: structure relevance tree, each associates the most by force the branch of chain as described relevance tree.

8. the Rules extraction method of Process-Oriented object data according to claim 7, is characterized in that: described step S202 specifically comprises the steps:

Step S2021: get i-th data point, calculates the mean distance a of every other data point in this data point to its place bunch _i;

Step S2022: the mean distance calculating all data points in i-th data point to other bunches, and remember that value minimum in all mean distances is b _i;

Step S2023: the silhouette coefficient calculating i-th data point, is designated as

Step S2024: calculate the silhouette coefficient that clusters number is the cluster set of K, be designated as k _m=min _k{ max _ns _k, wherein, s _kspan be [-1,1].