CN103927431A - Power station boiler operation state monitoring method based on pyramid time frame - Google Patents

Abstract

The invention discloses a power station boiler operation state monitoring method based on a pyramid time frame. Fuzzy association rules are introduced into power industry data association mining, and the defect that traditional quantization association rules are excessively stiff in boundary division is overcome. An Apriori algorithm is expanded to a fuzzy attribute business to determine division number of each fuzzy attribute, the possible values of all attributes are mapped to fuzzy concentration, and a large item set with the support degree larger than a minimum support degree threshold is found out so as to obtain interested rules. A large amount of historical data is stored during operation of a power station, the data updates constantly in real time and comprises a plurality of rules, the pyramid time frame is built in order to effectively and simply store the (historical) rules, and associated rules are stored selectively according to the time frame so as to perform effectively monitoring, comparing and evaluating boiler efficiency. The method has the advantages that boiler efficiency can be categorized and associated to evaluation of related parameters, effective monitoring and diagnosis are achieved, and basis is provided for operators to correspondingly adjust and control boiler operation.

Description

Station boiler method for monitoring operation states based on pyramid time frame

Technical field

The present invention relates to a kind of boiler operating state monitoring method, relate in particular to the station boiler method for monitoring operation states based on pyramid time frame, belong to machine learning modeling field.

Background technology

Data mining is one of database research, the most active branch of development and application, is the result of database technology natural evolvement.Database Systems are also strong issued transaction instruments.In transaction database, Mining Association Rules is the very important research topic in Data Mining.First the people such as R.Agrawal have proposed to excavate the correlation rule problem between transaction data base middle term collection in 1993.Researchist has carried out a large amount of research to the Mining Problems of correlation rule, their work comprises original algorithm is optimized, as introduce stochastic sampling, parallel thought etc., to improve the efficiency of algorithm mining rule, the application of correlation rule is promoted.

Correlation rule can be divided into two kinds: boolean association rule and Quantitative Association Rule.Boolean association rule, namely item value is 0 or 1.Quantitatively association rule mining be a kind of from the data that comprise connection attribute the data mining technology of Mining Association Rules, can change into boolean association rule by Discretization for Continuous Attribute and excavate.For power station operational system, determine the mining process of the data characteristic of quantified property, be applicable to application quantification Mining Theories of Association Rules and solve.Quantitative Association Rule Mining Problems more complicated, a kind of naturally idea is that it is changed into Boolean Association Rules Mining Problems, when the value of whole attributes is when being all limited, need be only a boolean properties value by each best property of attribute mapping, in the time that the span of attribute is very wide, need to be divided into several sections, then each section is mapped as to a boolean properties.In recent years, also oneself has studied how to excavate power station quantified property from different angles to many scholars, and has obtained certain achievement.Wherein fuzzy concept is one of effective ways.

Summary of the invention

Goal of the invention: the invention provides a kind of station boiler method for monitoring operation states based on pyramid time frame, fuzzy association rules is introduced in the excavation of power industry data correlation, correlation rule is embedded to pyramid time frame, to realize, boiler efficiency is effectively monitored to comparative evaluation.

Technical scheme: for solving the problems of the technologies described above, the station boiler method for monitoring operation states based on pyramid time frame provided by the invention, comprises the following steps:

Step 1, by the each Transaction Information T in transaction database D _i(i=1,2 ..., each project n) utilize given membership function to map out fuzzy set

$f_i^j=\frac{{\mathrm{\μ}}_i\left(R_{j1}\right)}{R_{j1}}+\frac{{\mathrm{\μ}}_i\left(R_{j2}\right)}{R_{j2}}+...+\frac{{\mathrm{\μ}}_i\left(R_{\mathrm{jk}}\right)}{R_{\mathrm{jk}}}$

Wherein, R _jlfor project l fuzzy subregion, μ _i(R _jl) be subregion R _jlon degree of membership value;

Step 2, calculates n Transaction Information T _i(i=1,2 ..., n) in each project at corresponding fuzzy set R _js(s=1,2 ..., the k) weights of middle degree of membership

${\mathrm{weight}}_{\mathrm{js}}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\μ}}_j\left(R_{\mathrm{js}}\right);$

Step 3, to each subregion R _js(1≤j≤m, 1≤s≤k), check the weight w eight that each fuzzy set is corresponding _jswhether exceed minimum support minsupport given in advance; If subregion R _jsmeet above condition, put it into a frequent collection L ₁in, as shown in the formula

L ₁＝{R _js|weight _js≥minsupport，1≤j≤m，1≤s≤k}；

Step 4, definition r is the current number of entry in frequent item set that is retained in;

Step 5, with the Apriori class of algorithms seemingly, from frequent item set L _rmiddle generation candidate C _r+1, wherein L _rconcentrate and have and only have r-1 project identical at two items, and two subregions that belong to same project can not appear at C simultaneously _r+1same in;

Step 6, candidate C _r+1in each newly-generated r+1 item collection t=(t ₁, t ₁..., t _r+1), it is handled as follows:

1) to each Transaction Information T _ithe degree of membership of project t is concentrated in calculated candidate sport

${\mathrm{\μ}}_{\mathrm{it}}={\mathrm{\μ}}_i\left(R_{t_1}\right)\mathrm{\Λ}{\mathrm{\μ}}_i\left(R_{t_2}\right)\mathrm{\Λ}...\mathrm{\Λ}{\mathrm{\μ}}_i\left(R_{t_{r+1}}\right)$

Wherein, for Transaction Information T _iat subregion degree of membership value

${\mathrm{\μ}}_{\mathrm{it}}=\underset{j=1}{\overset{r+1}{\min }}{\mathrm{\μ}}_i\left(R_{t_j}\right);$

2) weights of a project are concentrated in calculated candidate sport

3) if weight _tmeet the requirement of given minimum support, by project t=(t ₁, t ₁..., t _r+1) put into L _r+1in;

Step 7, if L _r+1for sky, carry out next step; If L _r+1non-NULL, puts r=r+1, goes to step five

Step 8, iterative step finishes, and all have a project (t ₁, t ₁..., t _q) sport collection t structure correlation rule, form all possible fuzzy association rules, calculate the degree of confidence of all fuzzy association rules, output degree of confidence meets the rule of given min confidence;

Step 9, by the correlation rule access time snapshot (section) obtaining;

Step 10, is divided into [log by time snapshot _at]+1 section, be respectively [0,1 ..., [1og _at]], wherein [] represents to round downwards, and T is that initial time arrives interval at once, a, l is time coefficient;

1) initialization s=0,

2) to being not more than in the number of T, choose last a ^l+ 1 numerical value deposits s item, s=s+1 in;

3) judge s≤[log _at]+1 whether set up, enter next step if set up, otherwise return to previous step;

4) from s=[log _at]+1 beginning, reduce successively s, delete the time repeating and sign, finally obtain pyramid time frame;

5) from time frame, intercept the correlation rule in a respective items snapshot, after integration, obtain taking into account the integration correlation rule of real time data and historical data characteristic.

Beneficial effect: correlation rule is embedded pyramid time frame by the present invention, has realized boiler efficiency has effectively been monitored to comparative evaluation, has the following advantages:

1, reflect more timely and accurately the situation of change of boiler operation performance situation, take into account history run trend simultaneously, for boiler performance optimization and status monitoring provide new thinking and method;

2, for power plant's monitoring information system Premium Features module (as boiler operatiopn optimization, condition monitoring and fault diagnosis etc.) provides can reference model;

3, consider the continuity of the running status of the each subsystem in power station (containing boiler), pyramid time window is introduced to association analysis, monitoring boiler efficiency changes and is associated with its material impact parameter is innovation of the present invention, can also carry out further field practice and research work according to result.

Brief description of the drawings

Fig. 1 is the process flow diagram of the embodiment of the present invention;

Fig. 2 is the membership function figure of step 2 in the embodiment of the present invention.

Embodiment

Embodiment: taking the 300MW of Fire Monitoring power plant unit boiler efficiency and correlation parameter thereof as example, the concrete processing procedure that the fuzzy association rules of description taken in conjunction pyramid time frame is excavated.

Certain day historical data under N300-16.7/537/537-3 type 300MW boiler typical load steady running condition is by load section (70%, 80%, 90%, 100%) classification, and taking out 10 groups is that example is analyzed.To load, excess air coefficient, unburned carbon in flue dust, exhaust gas temperature and 5 parameters of boiler efficiency of being closely related with boiler combustion situation as research object, given minimum support minsupport=0.3, min confidence minconfidence=0.75.

Each operating mode is by five major parameters: load (note is Mw), excess air coefficient (note is Air), unburned carbon in flue dust (note is Cf), exhaust gas temperature (note is FlueT) and boiler efficiency (note is Q1).The each attribute kit of typical condition history data set is containing 10 data.Whole process mainly contains the pre-service of input data, obfuscation, builds the nucleus modules such as correlation rule and pyramid time frame.Detailed process is as shown in Figure 1:

1, data enter input data pre-service link, and each operating mode is by five major parameters: load (note is Mw), excess air

Coefficient (note is Air), unburned carbon in flue dust (note is Cf), exhaust gas temperature (note is FlueT) and boiler efficiency (note is Q1).

The each attribute kit of typical condition history data set is containing 10 data (taking high load capacity as example) as shown in table 1.

Table 1 high load capacity operating mode history data set

2, every attribute takes out the fuzzy concepts such as basic, normal, high according to actual conditions, and input data-mapping after pretreatment is gone out to corresponding fuzzy set, and its membership function figure as shown in Figure 2.

3, according to minimum support and min confidence restriction, build candidate, determine frequent item set.Determine under this load, meet the high section boiler efficiency of minimum support and min confidence requirement and all the other parameters high in the \ fuzzy association rules of low section.

Table 2 operational factor fuzzy set table

The frequent collection of table 3

The frequent binomial collection of table 4

Frequent three collection of table 5

Filter out frequent binomial collection and frequent three collection as shown in table 4 table 5 according to min confidence and minimum support.At 300MW unit operation during in high load capacity section, excess air coefficient is in the time of 1.42 left and right, and boiler efficiency is up to optimum, therefore can be considered 1.42 load section excess air coefficient optimal values for this reason; From burning efficiency theoretically, more arcola efficiency is higher for flying dust and exhaust gas temperature, and from historical data Result, history run optimal value scope is respectively in the scope of 1.338 and 134.2 DEG C of left and right.

4,, according to initial time and current time, build pyramid time frame.To after the gelatinization of correlation rule reverse, deposit in pyramid time frame by time tag.

5, extract the interval value after the reverse gelatinization of storing in framework, made average value processing, can obtain the Air under high section boiler efficiency, Cf, the isoparametric numerical value of FlueT.

Taking above-mentioned foundation desirable embodiment of the present invention as enlightenment, by above-mentioned description, relevant staff can, not departing from the scope of this invention technological thought, carry out various change and amendment completely.The raw scope of technology of this invention is not limited to the content on instructions, must determine its technical scope according to claim scope.

Claims (1)

1. the station boiler method for monitoring operation states based on pyramid time frame, is characterized in that comprising the following steps:

Step 1, by the each Transaction Information T in transaction database D _i(i=1,2 ..., each project n) utilize given membership function to map out fuzzy set

$f_i^j=\frac{{\mathrm{\μ}}_i\left(R_{j1}\right)}{R_{j1}}+\frac{{\mathrm{\μ}}_i\left(R_{j2}\right)}{R_{j2}}+...+\frac{{\mathrm{\μ}}_i\left(R_{\mathrm{jk}}\right)}{R_{\mathrm{jk}}}$

Wherein, R _jlfor project l fuzzy subregion, μ _i(R _jl) be subregion R _jlon degree of membership value;

Step 2, calculates n Transaction Information T _i(i=1,2 ..., n) in each project at corresponding fuzzy set R _js(s=1,2 ..., the k) weights of middle degree of membership

${\mathrm{weight}}_{\mathrm{js}}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\μ}}_j\left(R_{\mathrm{js}}\right);$

Step 3, to each subregion R _js(1≤j≤m, 1≤s≤k), check the weight w eight that each fuzzy set is corresponding _jswhether exceed minimum support minsupport given in advance; If subregion R _jsmeet above condition, put it into a frequent collection L ₁in, as shown in the formula

L ₁＝{R _js|weight _js≥minsupport，1≤j≤m，1≤s≤k}；

Step 4, definition r is the current number of entry in frequent item set that is retained in;

Step 5, adopts Apriori algorithm, from frequent item set L _rmiddle generation candidate C _r+1, wherein L _rconcentrate and have and only have r-1 project identical at two items, and two subregions that belong to same project can not appear at C simultaneously _r+1same in;

Step 6, to candidate C _r+1in each newly-generated r+1 item collection t=(t ₁, t ₁..., t _r+1) be handled as follows:

1) to each Transaction Information T _ithe degree of membership of project t is concentrated in calculated candidate sport

${\mathrm{\μ}}_{\mathrm{it}}={\mathrm{\μ}}_i\left(R_{t_1}\right)\mathrm{\Λ}{\mathrm{\μ}}_i\left(R_{t_2}\right)\mathrm{\Λ}...\mathrm{\Λ}{\mathrm{\μ}}_i\left(R_{t_{r+1}}\right)$

Wherein μ _i for Transaction Information T _iat subregion degree of membership value

${\mathrm{\μ}}_{\mathrm{it}}=\underset{j=1}{\overset{r+1}{\min }}{\mathrm{\μ}}_i\left(R_{t_j}\right);$

2) weights of a project are concentrated in calculated candidate sport

3) if weight _tmeet the requirement of given minimum support, by project t=(t ₁, t ₁..., t _r+1) put into L _r+1in;

Step 7, if L _r+1for sky, carry out next step; Otherwise put r=r+1, go to step five;

Step 8, iterative step finishes, and all have a project (t ₁, t ₁..., t _q) sport collection t structure correlation rule, form all possible fuzzy association rules, calculate the degree of confidence of all fuzzy association rules, output degree of confidence meets the rule of given min confidence;

Step 9, by the correlation rule access time snapshot (section) obtaining;

Step 10, is divided into [log by time snapshot _at]+1 section, be respectively [0,1 ..., [log _at]], wherein [] represents to round downwards, and T is that initial time arrives interval at once, a, l is time coefficient;

1) initialization s=0,

2) to being not more than in the number of T, choose last a ^l+ 1 numerical value deposits s item, s=s+1 in;

3) judge s≤[log _at]+1 whether set up, enter next step if set up, otherwise return to previous step;

4) from s=[log _at]+1 beginning, reduce successively s, delete the time repeating and sign, finally obtain pyramid time frame;

5) from time frame, intercept the correlation rule in a respective items snapshot, after integration, obtain taking into account the integration correlation rule of real time data and historical data characteristic.