CN107146004A - A kind of slag milling system health status identifying system and method based on data mining - Google Patents
A kind of slag milling system health status identifying system and method based on data mining Download PDFInfo
- Publication number
- CN107146004A CN107146004A CN201710261505.7A CN201710261505A CN107146004A CN 107146004 A CN107146004 A CN 107146004A CN 201710261505 A CN201710261505 A CN 201710261505A CN 107146004 A CN107146004 A CN 107146004A
- Authority
- CN
- China
- Prior art keywords
- vertical mill
- data
- feature
- operating mode
- index
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Educational Administration (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Feedback Control In General (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of slag milling system health status identifying system and method based on data mining, mining analysis is carried out to floor data using a kind of Feature Selection method of synthesis, the key parameter for influenceing Vertical Mill stable is obtained, the index of Vertical Mill health state evaluation is used as;The index of Vertical Mill health state evaluation based on determination, the characteristics of carrying out cluster result analysis, each obtained operating mode cluster to work condition state obtains the state distribution situation in history operating mode, defines the running status classification in history operating mode;Then using ARIMA algorithms to the characteristic value training pattern determined in Vertical Mill health status feature acquisition module, the variation tendency to parameter is predicted, with the identification of predicted value secondary status.The present invention has higher accuracy of identification and generalization ability, and performance is good, it is adaptable to the health status identification and diagnosis of slag milling system.
Description
Technical field
The present invention relates to the identification of the health status of slag milling system and diagnose, it is more particularly to a kind of based on data mining
Slag milling system health status identifying system and method.
Background technology
Vertical Mill is a kind of equipment for the materials such as the slag of bulky grain to be ground to fine particle, mainly to building materials, change
The waste residue that the industries such as work, steel are produced carries out grinding, realizes the recycling of waste residue, ground obtained micro mist is usually as cement
The raw material of production.But slag milling system process is complicated, working environment is severe, and long-term heavy-duty service, system often goes out
Existing various failures, control system chain reaction can cause whole production line to shut down, and then cause the poorly efficient of production line pause
Situation.Therefore, in the urgent need to slag milling system health status is identified and assessed, the healthy shape of vertical mill system is predicted
State.
Health status Predicting Technique is to carry out overall merit in the health status to device systems, and health is characterized obtaining
On the basis of state performance parameter, its variation tendency is extended out and obtains following a period of time by the time series of analytical performance parameter
The technology of the health status changing rule of interior equipment.Recent domestic scholar to the health status Predicting Technique of complex equipment,
This problem is studied from the angle of random theory and fuzzy theory, main method has fusion forecasting method, arma modeling prediction
Method, Hidden Markov predicted method, fuzzy neural network predicted method, Kalman filter prediction method.Fusion forecasting method is based on same
Blend curve of the time series data under different weights of class equipment is as the method for prediction curve, and method is simple, directly perceived, no
Dependent on system physical model, but need more sample data.Integration technology is applied to the prediction of Gas Outburst by ox Xiao Ling etc.
Problem, obtained and preferably predicted the outcome.Arma modeling Predicting Technique is in autoregression model (AR models) and moving average model
A kind of Time Series Forecasting Methods set up on the basis of (MA models), method does not need system model when in use, but
It is only applicable to short-term forecast.Pham etc. predicts the degenerate state of change system with linear arma modeling and Nonlinear GARCH.
Hidden Markov predicted method is according to maximum likelihood theory to ask hiding according to the time series of observable performance parameter value
The markoff process of health status sequence, method is the life-cycle data extracting parameter equipped from sample, with available for long-term
Prediction, but substantial amounts of sample data is difficult acquisition.Peng Ying is by analyzing the Monitoring Data of hydraulic pump, with considering the hidden of aging factor
Semi-Markov Process (HSMM) Forecasting Methodology can describe pump performance degenerative process well.Fuzzy neural network predicted method
It is the forecast model that neural network weight is constantly trained based on fuzzy reasoning, method handles nonlinear problem using fuzzy theory,
Strong adaptability, but expertise is needed, transplantability is poor.With the automation and the raising of the level of informatization of micro mist industry, DCS
Control system has obtained commonly used in the factory, and a large amount of creation datas are have accumulated in database.
The content of the invention
In order to preferably realize the identification of Vertical Mill health status and diagnose, the present invention provides a kind of slag based on data mining
Grinding system health status identifying system and method.Concrete technical scheme is as follows:
A kind of slag milling system health status identifying system based on data mining, including:It is data preprocessing module, vertical
Grind health state evaluation index and excavate module, Vertical Mill health status Cluster Analysis module, the acquisition of Vertical Mill state estimation index feature
Module, Vertical Mill real-time characteristic parameter prediction module, wherein:
The data that Vertical Mill is gathered are carried out outlier processing, processing empty value, sliding-model control and returned by data preprocessing module
One change is handled, and is that the mining analysis of data is got ready;
Vertical Mill health state evaluation index excavates module, and floor data is carried out using a kind of Feature Selection method of synthesis
Mining analysis, obtains the key parameter for influenceing Vertical Mill stable, is used as the index of Vertical Mill health state evaluation;
Vertical Mill health status Cluster Analysis module, the index of the Vertical Mill health state evaluation based on determination, to work condition state
Cluster result analysis is carried out, stable mode operating mode storehouse is obtained;
The spy of the real time data of collection under Vertical Mill state estimation index feature acquisition module, analysis Vertical Mill running status
Point, it is determined that carrying out the characteristic value of real-time status judgement;
Vertical Mill real-time characteristic parameter prediction module, using ARIMA algorithms to true in Vertical Mill health status feature acquisition module
Fixed characteristic value carries out model training, and the variation tendency of Prediction Parameters is recognized with predicted value secondary status.
Further, in described data preprocessing module, data outliers processing, processing empty value pass through data screening
Realized with data cleansing.Sliding-model control and normalized, are realized by brief converted with data of feature.
Further, described Vertical Mill health state evaluation index is excavated in module, a kind of Feature Selection method of synthesis
This five kinds of methods synthesis are eliminated by random lasso, ridge regression, random forest, stable Sexual behavior mode and recursive feature to constitute.Screening is calculated
Method is, by solving the relation between input variable and output variable, the importance of each feature to be given using five kinds of methods respectively
To give a mark, five kinds of scoring events are handled, the importance of feature is estimated according to the scores after processing, it is determined that
Key feature in feature set to be selected.
Further, comprising the following steps that for Vertical Mill operation key feature screening is carried out:
1) to vibrate as output y, it is characterized as inputting x with other, feature set to be selected is carried out using five kinds of methods respectively
Screening, calculates the score of each feature;
2) mechanism of different method characteristics screening is different, to eliminate the fraction difference that the difference of Filtering system is caused,
The scores of every kind of algorithm are all handled using the normalization method of maximin, score be limited in [0,1] it
Between, the average of each parameter attribute is then sought, using average value as the foundation of feature importance ranking, characteristic value choosing is carried out
Select.
3) comprehensive score to parameter is analyzed, and the controllability and physical meaning of incorporating parametric determine the pass of influence vibration
Bond parameter.In terms of scoring event, feeding capacity, micro mist are ground than table, grinding machine inlet pressure, main exhaust fan rotating speed, circulation valve area
The average value of machine inlet temperature excludes the relatively low characteristic parameter of these scores than relatively low.Several parameters of highest scoring, according to from
High to Low order is followed successively by:Thickness of feed layer, grinding machine pressure difference, grinding machine outlet temperature, circulation valve area.
4) according to step 2) and step 3) in analysis, assess characteristic parameter the selection result.The higher ginseng of four scores
In number, grinding machine pressure difference, thickness of feed layer, three parameters of Vertical Mill outlet temperature belong to outcome variable, and the value of parameter is at other
The result obtained under the combined influence of controlled variable.And it is that regulation and control variable is not suitable as sentencing for work condition state to circulate valve area
Severed finger mark.
Further, described Vertical Mill health status Cluster Analysis module, the Vertical Mill health state evaluation based on determination
Index, with reference to the data distribution in practical production experience and operating mode storehouse, it is determined that four stable judge index can cause operation different
Pretreated data are further screened by normal critical value in the range of the restriction of multiple critical values, ask satisfaction all
The data of restrictive condition, obtained the selection result as cluster input data.Clustering uses K- averages (k-
Means) K operating mode cluster in data set is found.Here K is that user specifies, and the purpose of algorithm is found in data set
K cluster barycenter, the point in data set is distributed to the barycenter nearest apart from the point, and the point is distributed into barycenter correspondence
Classification.According to the definition to data mode in cluster point group, complete to mark the classification that existing operating condition is recorded, it is trustworthy
Determine operating mode class label and be set to 0, unsteady-stage conditions label is set to 1, and therefrom extract steady working condition, set up stable mode work
Condition storehouse.
Described Vertical Mill state estimation index feature acquisition module, is exported with vibration, thickness of feed layer, grinding machine pressure difference, grinding machine
Based on the real time data of this 4 state estimation indexs of temperature, average, variance of each parameter in access window time are calculated
With exceptional value occurrence number, the characteristic variable that obtained result is judged as steady working condition.
Further, described Vertical Mill real-time characteristic parameter prediction module, is entered using time series algorithm to running status
Row prediction, and judged with obtained predicted value secondary status.Needing the parameter of prediction includes vibration, thickness of feed layer, grinding machine outlet
Time series models are respectively trained in this five parameters by temperature, grinding machine pressure difference, exceptional value number of times.Obtained model can be detected
Whether one section of sequence is stationary sequence, provides the numerical prediction of parameter, is recognized with predicted value secondary status.
According to the characteristics of Vertical Mill operating mode, because the external factor such as environment and other specification are to the combined effect of vibration, cause
Operating mode sequence belongs to non-stationary series, and the modeling of time series is carried out using ARIMA models.
Stationary sequence:Pair with sequence { X (t) }, if numerical value is fluctuated in a certain limited range, sequence has constant
Average and constant variance, and it is equal to postpone the auto-covariance and auto-correlation coefficient of the sequence variables of k phases, then and the sequence is
Stationary sequence.
Calculus of differences:It is assumed that the time interval of two sequences is T, calculus of differences is exactly pair the sequence for being mutually divided into k T
It should be worth and do subtraction, during k=1, referred to as first-order difference computing.
The essence of ARIMA models is that calculus of differences is added before ARMA computings, is then modeled using ARMA, is calculated
Formula is as follows:
xt=φ0+φ1xt-1+φ2xt-2+...+φpxt-p+εt-θ1εt-1-θ2εt-2-...-θqεt-q
The model thinks the multiple linear letter of the interference ε of the x values of p phases and preceding q phases before the variable x of t value is
Number.Error term is current random disturbances εt, it is zero-mean white noise sequence.Arma modeling think in the past the p phases sequential value and
The error term joint effect x of phase in past qtValue.
A kind of slag milling system health status recognition methods based on data mining, step is as follows:
1) floor data is analyzed using comprehensive Feature Selection method, it is determined that the stable key parameter of influence, makees
For the judge index of stable state.The span of key parameter in analysis of history data, according to its distributed area, it is determined that triggering
The critical value of stable regulation and control;
2) with step 1) in determine stable judge index be characterized, to work condition state progress clustering, using based on
The clustering algorithm of K- averages is excavated to data, the characteristics of analyzing each operating mode cluster that cluster result is obtained, and obtains history work
State distribution situation in condition;
3) according to the Result of clustering, the running status classification in history operating mode is defined, to the shape belonging to operating mode
State carries out classification mark and screening, obtains stable mode operating mode storehouse;
4) the characteristics of and then to the real time data of the collection under Vertical Mill running status, is analyzed, it is determined that carrying out real-time status
The characteristic value of judgement;
5) using ARIMA algorithms to step 4) in the characteristic value that determines carry out model training, the variation tendency to parameter enters
Row prediction, is judged with predicted value secondary status.
Beneficial effects of the present invention, which are mainly manifested in, to monitor slag milling system in real time based on an accurate model
The healthy running status of system, overall merit is carried out in the health status to device systems, obtains sign health status performance ginseng
On the basis of number, its variation tendency is extended out and obtains the strong of interior Vertical Mill of following a period of time by the time series of analytical performance parameter
Health state change rule, increases the security reliability of slag milling system, is conducive to preventing accident.The present invention has higher
Accuracy of identification and generalization ability, predicated error is relatively low, and prediction effect is good.
Brief description of the drawings
Fig. 1 is the slag milling system health status identifying system structural representation based on data mining.
Fig. 2 is the preprocessing process figure of Vertical Mill data.
Fig. 3 is Vertical Mill health state evaluation index mining process figure.
Fig. 4 is Vertical Mill health status K-means clustering flow charts.
When Fig. 5 is k=3, the parameter distribution probability density figure of clustering point group, (a) is classification 0, and (b) is classification 1,
(c) it is classification 2.
Fig. 6 is that Vertical Mill state estimation index feature obtains flow chart.
Fig. 7 is the time series modeling procedure chart of Vertical Mill real-time characteristic parameter prediction.
Fig. 8 is the original series figure in vibration a period of time.
Fig. 9 is the partial autocorrelation figure after the original series first-order difference in vibration a period of time.
Figure 10 is the predicted value of system and the graph of a relation of actual value.
Embodiment
Refer to the attached drawing, which can be more fully described on the present invention, figure, shows certain embodiments of the present invention, but not institute
Some embodiments.In fact, the present invention can be embodied as in many different forms, it should not be regarded as and be only limitted to institute here
The embodiment of elaboration, and embodiments of the invention should be regarded as in order that present disclosure meets applicable conjunction
What method was required and provided.Present invention is elaborated explanation with reference to Figure of description and specific implementation.
Fig. 1 lists the function of each module of slag milling system health status identifying system based on data mining and each
Logical relation between module.
The data that Vertical Mill is gathered are carried out outlier processing, processing empty value, sliding-model control and returned by data preprocessing module
One change is handled, and is that the mining analysis of data is got ready;
Vertical Mill health state evaluation index excavates module, and floor data is carried out using a kind of Feature Selection method of synthesis
Mining analysis, obtains the key parameter for influenceing Vertical Mill stable, is used as the index of Vertical Mill health state evaluation;
Vertical Mill health status Cluster Analysis module, the index of the Vertical Mill health state evaluation based on determination, to work condition state
Cluster result analysis is carried out, stable mode operating mode storehouse is obtained;
The spy of the real time data of collection under Vertical Mill state estimation index feature acquisition module, analysis Vertical Mill running status
Point, it is determined that carrying out the characteristic value of real-time status judgement;
Vertical Mill real-time characteristic parameter prediction module, using ARIMA algorithms to true in Vertical Mill health status feature acquisition module
Fixed characteristic value carries out model training, and the variation tendency of Prediction Parameters is recognized with predicted value secondary status.
It is illustrated in figure 2 the preprocessing process figure of Vertical Mill data.The quality of data has very big to the analysis result of data mining
Influence.A large amount of attributes are contained in the Vertical Mill initial data of acquisition, there is improper value and exceptional value, it is necessary to be carried out to data preliminary
Screening, removes improper value and exceptional value, it is ensured that the accuracy of data, and removes the attribute unrelated with excavation, and to ensure sample
The diversity of notebook data and the completeness of characteristic information.In addition it is also necessary to be handled according to algorithm requirements data, make data
Meet the input requirements of algorithm.
In described data preprocessing module, data outliers processing, processing empty value pass through data screening and data cleansing
Realize:Vertical Mill feed, grinding, ventilation apparatus, dust separation equipment, hydraulic station, hot-blast stove, warehouse etc. are contained in data with existing
The parameter attribute that 65 partial measuring points are obtained.Obtain including 30 main works of Vertical Mill from 65 attributes after attribute selection
The attribute set of skill and performance parameter, including the vibration of Vertical Mill, feeding capacity, electric current, grinding pressure, thickness of feed layer, air feed system
The aperture of cold and hot air-valve, the aperture for circulating air-valve, powder concentrator rotating speed, each main electrical current etc..In Vertical Mill startup, shut down and failure
Before and after occurring, because operating mode is highly unstable, parameter can big ups and downs.And there is record missing in Vertical Mill data, it is abnormal and
The situation of misregistration.Some records lack some parameter values, have plenty of the factors such as manual entry mistake or sensor fault and lead
The data deviation of cause, missing or abnormal.In order to exclude interference of these factors to data, it is necessary to these missing records and mistake
Value is handled, it is ensured that data it is correct, credible, so just can guarantee that the reliable and validity of Result.
In described data preprocessing module, sliding-model control and normalized are converted in fact by feature is brief with data
It is existing:Consider the artificial setting of the feature distribution, enterprise of Vertical Mill data to parameter, and in actual motion parameter controllability
Situations such as, it is brief to data progress, to reduce the dimension of data, save data processing time.By the brief residue of feature
In 14 characteristic parameters, grinding machine main frame electric current, powder concentrator electric current, three main electrical current parameters of main exhaust fan electric current are contained.Due to
Reduction can take more concerned be overall energy consumption reduction, rather than single part energy consumption change, therefore construct one
New attribute is used for characterizing the size of power consumption, is named as total current.The value of total current is equal to grinding machine main frame electric current, powder concentrator electricity
Stream, the algebraical sum of main exhaust fan electric current.Feature set so to be selected is simplified to 12 features.
Vertical Mill health state evaluation index mining process figure is illustrated in figure 3, specific excavation step is as follows:
1) to vibrate as output y, it is characterized as inputting x with other, feature set to be selected is carried out using five kinds of methods respectively
Screening, calculates the score of each feature;
2) mechanism of different method characteristics screening is different, to eliminate the fraction difference that the difference of Filtering system is caused,
The scores of every kind of algorithm are all handled using the normalization method of maximin, score has been limited in [0,
1] between, the average of each parameter attribute is then sought, using average value as the foundation of feature importance ranking, feature is carried out
Value selection.Applied in Vertical Mill data, result is obtained after algorithm process as shown in table 1 below.
The scoring event of the different feature selection approach of table 1 feature to be selected
Method characteristic | Random LASSO | Ridge regression | Random forest | Stable Sexual behavior mode | Recursive feature is eliminated | Average |
Feeding capacity | 0.1 | 0 | 0.03 | 0.08 | 0.18 | 0.08 |
Micro mist compares table | 0.31 | 0.39 | 0.07 | 0.0 | 0.09 | 0.17 |
Thickness of feed layer | 0.6 | 1.0 | 1.0 | 0.8 | 0.71 | 0.82 |
Grinding machine outlet temperature | 0.21 | 0.45 | 0.32 | 0.66 | 0.42 | 0.41 |
Grinding machine inlet temperature | 0.0 | 0.0 | 0.23 | 0.0 | 0.14 | 0.07 |
Grinding machine inlet pressure | 0.11 | 0.0 | 0.43 | 0.24 | 0.13 | 0.18 |
Powder concentrator rotating speed | 0.06 | 0.0 | 0.27 | 0.0 | 0.59 | 0.18 |
Grinding machine pressure difference | 0.5 | 0.79 | 0.67 | 0.95 | 0.95 | 0.77 |
Cold wind valve opening | 0.29 | 0.0 | 0.0 | 0.0 | 0.09 | 0.08 |
Hot blast valve opening | 0.21 | 0.0 | 0.01 | 0.12 | 0.0 | 0.07 |
Circulate valve area | 0.6 | 0.21 | 0.14 | 0.24 | 0.33 | 0.3 |
Main exhaust fan rotating speed | 0.01 | 0.1 | 0.01 | 0.0 | 0.0 | 0.02 |
3) comprehensive score to parameter is analyzed, and the controllability and physical meaning of incorporating parametric are determined to influence vibration
Key parameter.In terms of scoring event, feeding capacity, micro mist are than table, grinding machine inlet pressure, main exhaust fan rotating speed, circulation valve area
The average value of grinding machine inlet temperature excludes the relatively low characteristic parameter of these scores than relatively low.Several parameters of highest scoring, according to
Order from high to low is followed successively by:Thickness of feed layer, grinding machine pressure difference, grinding machine outlet temperature, circulation valve area.
4) according to step 2) and step 3) in analysis, assess characteristic parameter the selection result.The higher ginseng of four scores
In number, grinding machine pressure difference, thickness of feed layer, three parameters of Vertical Mill outlet temperature belong to outcome variable, and the value of parameter is at other
The result obtained under the combined influence of controlled variable.And it is that regulation and control variable is not suitable as sentencing for work condition state to circulate valve area
Severed finger mark.
In summary analyze, it is final to determine that vibration, thickness of feed layer, grinding machine pressure difference, 4 parameters one of grinding machine outlet temperature are acted as
The index judged for stable state.
It is illustrated in figure 4 Vertical Mill health status K-means clustering flow charts.With reference to practical production experience and operating mode storehouse
In data distribution, it is determined that four stable judge index can cause the critical value of operation exception, in the restriction of multiple critical values
In the range of pretreated data are further screened, seek the data for meeting all restrictive conditions, obtained the selection result
It is used as the input data of cluster.Clustering uses K- averages (k-means) to find K operating mode cluster in data set.
Here K is that user specifies, and the purpose of algorithm is to find the barycenter of K cluster in data set, and the point in data set is distributed
The corresponding classification of the barycenter is distributed to the barycenter nearest apart from the point, and by the point.
When choosing K=3, cluster result is as follows, and the data point number in cluster centre and each cluster as shown in table 2, divides group
Parameter distribution probability density figure it is as shown in Figure 5.
Table 2k=3, point group's cluster centre table
Classification | Thickness of feed layer | Grinding machine outlet temperature | Grinding machine pressure difference | Grinding machine shell vibrates | Class number |
0 | -0.464651 | 0.564229 | -0.110864 | 0.520456 | 2276 |
1 | -0.182965 | -0.963877 | -0.437062 | -0.448334 | 2178 |
2 | 1.551188 | 0.872879 | 1.284423 | -0.219220 | 937 |
As can be seen from Figure 5:
The feature of classification 0:The thick span of the bed of material between 125~135mm, grinding machine outlet temperature at 100~108 DEG C,
Grinding machine pressure difference is in 2800~3200Pa, and vibration values concentrate on 7, near 8,9 three values.
The feature of classification 1:The thick span of the bed of material is between 125~144mm, and grinding machine outlet temperature is at 95~103 DEG C, mill
Machine pressure difference is in 2800~3200Pa, and vibration values concentrate on 6, near 7,8 three values.
The feature of classification 2:The thick span of the bed of material between 140~150mm, grinding machine outlet temperature at 102~108 DEG C,
Grinding machine pressure difference is concentrated between 6~8 in 3200~3500Pa, vibration values.
When choosing K=3, the plyability of vibration is larger, and the distance interval of other three parameters is more reasonable, comes with reference to data
The design production suggestion of source Vertical Mill, take three cluster center when obtained classification 0 be defined as in unsteady state, classification 1 and 2
Record is defined as stable state.
It is illustrated in figure 6 Vertical Mill state estimation index feature and obtains flow chart, specific acquisition process is as follows:
1) the real-time working condition data at T moment are gathered, null value and rejecting outliers are carried out to the data collected, if read
During there is null value, give up data or fill up null value with history average.It is disposed, according to the stability index number of setting
According to sampling interval △ t, continue to read the data at next collection moment, carry out Data Detection, repeat this process until obtaining n bars
Record;
2) during collection n bars record, if exceptional value occurs, time that the exceptional value of each parameter occurs is added up
Number.The basis for estimation of exceptional value is with reference to the parameters span obtained from steady working condition pattern base, when the ginseng collected
Number exceeds normal range (NR), then it is assumed that the data at the moment are exceptional value.
3) average and standard deviation of parameters in n bars record are calculated.Each parameter is finally obtained to obtain within the access cycle
To average, three dimension characteristic values that totally 12 numerical value judges as operating mode of variance and exceptional value number of times, to stable state
Judgement.
It is illustrated in figure 7 the time series modeling procedure chart of Vertical Mill real-time characteristic parameter prediction.According to the spy of Vertical Mill operating mode
Point, because the external factor such as environment and other specification are to the combined effect of vibration, cause operating mode sequence to belong to non-stationary series, adopts
The modeling of time series is carried out with ARIMA models.
Stationary sequence:Pair with sequence { X (t) }, if numerical value is fluctuated in a certain limited range, sequence has constant
Average and constant variance, and it is equal to postpone the auto-covariance and auto-correlation coefficient of the sequence variables of k phases, then and the sequence is
Stationary sequence.
Calculus of differences:It is assumed that the time interval of two sequences is T, calculus of differences is exactly pair the sequence for being mutually divided into k T
It should be worth and do subtraction, during k=1, referred to as first-order difference computing.
The essence of ARIMA models is that calculus of differences is added before ARMA computings, is then modeled using ARMA, is calculated
Formula is as follows:
xt=φ0+φ1xt-1+φ2xt-2+...+φpxt-p+εt-θ1εt-1-θ2εt-2-...-θqεt-q
The model thinks the multiple linear letter of the interference ε of the x values of p phases and preceding q phases before the variable x of t value is
Number.Error term is current random disturbances εt, it is zero-mean white noise sequence.Arma modeling think in the past the p phases sequential value and
The error term joint effect x of phase in past qtValue.
The process that explanation is modeled using time series by taking vibration values as an example.First to being collected in one section of continuous time
Vibration values carry out stationarity detection, fetch at intervals of 5 seconds, it is continuous 35 vibration value data be illustrated in fig. 8 shown below, can see
Go out the sequence on the rise, belong to non-stationary series.Auto-correlation coefficient is asked for sequence, the absolute value of coefficient correlation is big for a long time
In zero, show that the sequence has long-term correlation.The partial autocorrelation figure such as Fig. 9 for this sequence obtain after first-order difference
It is shown.It can be seen that the timing diagram of sequence is fluctuated near average after first-order difference, and fluctuation range is less, so first-order difference
Sequence afterwards is stationary sequence.
Then white noise sound detection is carried out to the sequence after first-order difference, obtained P values are less than 0.05, so after first-order difference
Sequence belong to steady non-white noise sequence, can be fitted with arma modeling.Next arma modeling is carried out determining rank,
It is exactly the parameter in modulus type, the size of the BIC information content obtained according to p, q all combinations determines that selection makes BIC information
Amount reaches p, q combination of minimum.It can be just predicted after model order using the ARIMA models set up.
Forecast model can provide the predicted value, standard error and confidential interval of continuous 5 minutes, predicted value and actual value
Relation is as shown in Figure 10.As can be seen from the figure predicated error is relatively low, and predicted value can reflect the variation tendency of numerical value, mould substantially
The prediction effect of type is good.
Claims (10)
1. a kind of slag milling system health status identifying system based on data mining, it is characterised in that the system includes:Number
Data preprocess module, Vertical Mill health state evaluation index are excavated module, Vertical Mill health status Cluster Analysis module, Vertical Mill state and commented
Estimate index feature acquisition module, Vertical Mill real-time characteristic parameter prediction module, wherein:
The data that Vertical Mill is gathered are carried out outlier processing, processing empty value, sliding-model control and normalization by data preprocessing module
Processing;
Vertical Mill health state evaluation index excavates module, and floor data is excavated using a kind of Feature Selection method of synthesis
Analysis, obtains the key parameter for influenceing Vertical Mill stable, is used as the index of Vertical Mill health state evaluation;
Vertical Mill health status Cluster Analysis module, the index of the Vertical Mill health state evaluation based on determination is carried out to work condition state
Cluster result is analyzed, and obtains stable mode operating mode storehouse;
The characteristics of real time data of collection under Vertical Mill state estimation index feature acquisition module, analysis Vertical Mill running status, really
Surely the characteristic value of real-time status judgement is carried out;
Vertical Mill real-time characteristic parameter prediction module, using ARIMA algorithms to determining in Vertical Mill health status feature acquisition module
Characteristic value carries out model training, and the variation tendency of Prediction Parameters is recognized with predicted value secondary status.
2. the system as claimed in claim 1, it is characterised in that in described data preprocessing module, data outliers processing,
Processing empty value, is realized by data screening and data cleansing.Sliding-model control and normalized, are become by feature is brief with data
Change realization.
3. the system as claimed in claim 1, it is characterised in that described Vertical Mill health state evaluation index is excavated in module,
A kind of Feature Selection method of synthesis eliminates this by random lasso, ridge regression, random forest, stable Sexual behavior mode and recursive feature
Five kinds of methods integrate composition, and mining analysis is carried out to floor data, obtain the key parameter for influenceing Vertical Mill stable, it is determined that vibration,
The index of thickness of feed layer, grinding machine pressure difference, 4 parameters of grinding machine outlet temperature collectively as Vertical Mill health state evaluation.
4. system as claimed in claim 3, it is characterised in that described comprehensive characteristics screening technique, is become by solving input
Measure the relation between output variable, to the scores of every kind of algorithm using the normalization method of maximin at
Reason, score has been limited between [0,1], has then sought the average of each parameter attribute, important using average value as feature
Property sequence foundation, the importance of feature is estimated according to the scores after processing, carry out characteristic value selection.
5. the system as claimed in claim 1, it is characterised in that described Vertical Mill health status Cluster Analysis module, based on true
The index of fixed Vertical Mill health state evaluation, the characteristics of cluster result analysis, each obtained operating mode cluster are carried out to work condition state,
The state distribution situation in history operating mode is obtained, the running status classification in history operating mode is defined, the state belonging to operating mode is entered
Row classification is marked and screened, and obtains stable mode operating mode storehouse.
6. system as claimed in claim 5, it is characterised in that described cluster result analysis is k-means clusterings, choosing
K=3 is taken, the record that obtained classification 0 is defined as in unsteady state, classification 1 and 2 is defined as stable state.The feature of classification 0:
The thick span of the bed of material between 125~135mm, grinding machine outlet temperature at 100~108 DEG C, grinding machine pressure difference 2800~
3200Pa, vibration values concentrate on 7, near 8,9 three values.The feature of classification 1:The thick span of the bed of material 125~144mm it
Between, grinding machine outlet temperature is at 95~103 DEG C, and grinding machine pressure difference is in 2800~3200Pa, and vibration values concentrate on 6,7,8 three values are attached
Closely.The feature of classification 2:The thick span of the bed of material is between 140~150mm, and grinding machine outlet temperature is in 102~108 DEG C, grinding machine pressure
Difference is in 3200~3500Pa, and vibration values are concentrated between 6~8.
7. the system as claimed in claim 1, it is characterised in that described Vertical Mill state estimation index feature acquisition module, with
Based on vibration, thickness of feed layer, grinding machine pressure difference, the real time data of grinding machine outlet temperature this 4 state estimation indexs, calculate each
Average, variance and exceptional value occurrence number of the parameter in access window time, judge obtained result as steady working condition
Characteristic variable.
8. the system as claimed in claim 1, it is characterised in that described Vertical Mill real-time characteristic parameter prediction module, under
The ARIMA algorithms of formula carry out model training to the characteristic value determined in Vertical Mill health status feature acquisition module:
xt=φ0+φ1xt-1+φ2xt-2+...+φpxt-p+εt-θ1εt-1-θ2εt-2-...-θqεt-q
Wherein, xt-pFor the x values of preceding p phases, εt-qFor the interference value of preceding q phases, φ0~φpFor the coefficient of different time x values, θ1
~θqFor the coefficient of different time ε values.
9. a kind of slag milling system health status recognition methods based on data mining, it is characterised in that including step:
1) floor data is analyzed using comprehensive Feature Selection method, it is determined that the stable key parameter of influence, as steady
Determine the judge index of state.The span of key parameter in analysis of history data, according to its distributed area, it is determined that triggering is stable
The critical value of regulation and control;
2) with step 1) in determine stable judge index be characterized, to work condition state progress clustering, use clustering algorithm
Data are excavated, the characteristics of analyzing each operating mode cluster that cluster result is obtained, obtain the state distribution feelings in history operating mode
Condition;
3) according to the Result of clustering, the running status classification in history operating mode is defined, the state belonging to operating mode is entered
Row classification is marked and screened, and obtains stable mode operating mode storehouse;
4) to the real time data of the collection under Vertical Mill running status the characteristics of is analyzed, it is determined that carrying out the spy of real-time status judgement
Value indicative;
5) using ARIMA algorithms to step 4) in the characteristic value that determines carry out model training, the variation tendency of parameter is carried out pre-
Survey, judged with predicted value secondary status.
10. method as claimed in claim 9, it is characterised in that the Feature Selection method of the synthesis, including random lasso,
Ridge regression, random forest, stable Sexual behavior mode and recursive feature eliminate this five kinds of methods.By solving input variable and output variable
Between relation, the importance of each feature is given a mark using five kinds of methods respectively, five kinds of scoring events are handled,
The importance of feature is estimated according to the scores after processing, the key feature in feature set to be selected is determined.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710261505.7A CN107146004B (en) | 2017-04-20 | 2017-04-20 | A kind of slag milling system health status identifying system and method based on data mining |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710261505.7A CN107146004B (en) | 2017-04-20 | 2017-04-20 | A kind of slag milling system health status identifying system and method based on data mining |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107146004A true CN107146004A (en) | 2017-09-08 |
CN107146004B CN107146004B (en) | 2018-02-16 |
Family
ID=59775267
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710261505.7A Active CN107146004B (en) | 2017-04-20 | 2017-04-20 | A kind of slag milling system health status identifying system and method based on data mining |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107146004B (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052970A (en) * | 2017-12-08 | 2018-05-18 | 深圳市智物联网络有限公司 | A kind of data processing method and processing equipment |
CN109376919A (en) * | 2018-10-12 | 2019-02-22 | 西安科技大学 | A kind of prediction technique of coal mine fully-mechanized mining working gas emission |
CN109507535A (en) * | 2018-12-10 | 2019-03-22 | 国网河南省电力公司电力科学研究院 | Grounding net of transformer substation operation phase and service life prediction technique and device |
CN109886443A (en) * | 2017-12-26 | 2019-06-14 | 广东电网有限责任公司电力调度控制中心 | A kind of failure Prediction System based on gray system theory |
CN109902871A (en) * | 2019-02-27 | 2019-06-18 | 万洲电气股份有限公司 | A kind of intelligent optimization energy conserving system of combination enterprise production line differentiation feature |
CN110086829A (en) * | 2019-05-14 | 2019-08-02 | 四川长虹电器股份有限公司 | A method of Internet of Things unusual checking is carried out based on machine learning techniques |
CN110245842A (en) * | 2019-05-24 | 2019-09-17 | 电子科技大学 | A kind of production line Risk Scheduling method of equipment oriented burst major break down |
CN110569279A (en) * | 2019-08-23 | 2019-12-13 | 长沙学院 | Time series signal reconstruction method based on variable projection algorithm |
CN110618984A (en) * | 2019-08-27 | 2019-12-27 | 西安因联信息科技有限公司 | Shutdown vibration data cleaning method |
CN111259730A (en) * | 2019-12-31 | 2020-06-09 | 杭州安脉盛智能技术有限公司 | State monitoring method and system based on multivariate state estimation |
CN111639842A (en) * | 2020-05-20 | 2020-09-08 | 湖北博华自动化系统工程有限公司 | Equipment health evaluation method, evaluation system and equipment health prediction method |
CN112015620A (en) * | 2020-08-19 | 2020-12-01 | 浙江无极互联科技有限公司 | Method for automatically adjusting and optimizing parameters of website service end system |
CN112015619A (en) * | 2020-08-19 | 2020-12-01 | 浙江无极互联科技有限公司 | Method for optimizing and screening core key indexes of system through parameters |
CN112365458A (en) * | 2020-11-02 | 2021-02-12 | 中材邦业(杭州)智能技术有限公司 | Grate cooler snowman identification method and system based on ANN neural network |
CN112381965A (en) * | 2020-11-03 | 2021-02-19 | 浙大城市学院 | Aeroengine health state identification system and method based on data mining |
CN112699926A (en) * | 2020-12-25 | 2021-04-23 | 浙江中控技术股份有限公司 | Method for identifying saturated grinding abnormity of cement raw material vertical mill based on artificial intelligence technology |
TWI771531B (en) * | 2018-11-23 | 2022-07-21 | 中華電信股份有限公司 | Method and system for predicting system health using machine learning |
CN115964907A (en) * | 2023-03-17 | 2023-04-14 | 中国人民解放军火箭军工程大学 | Complex system health trend prediction method and system, electronic device and storage medium |
CN117436353A (en) * | 2023-12-21 | 2024-01-23 | 泰安维创游乐设备有限公司 | Intelligent recreation facility fault prediction method based on big data |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102319612A (en) * | 2011-07-05 | 2012-01-18 | 中国科学院沈阳自动化研究所 | Method for intelligently controlling pressure difference of cement raw meal vertical mill |
CN102323751A (en) * | 2011-06-28 | 2012-01-18 | 浙江大学 | Preparatory grinding system control method based on fuzzy intelligence control and optimization method |
CN103028480A (en) * | 2012-12-10 | 2013-04-10 | 上海凯盛节能工程技术有限公司 | Intelligent control system for vertical mill based on fuzzy PID (proportion integration differentiation) algorithm |
US9336493B2 (en) * | 2011-06-06 | 2016-05-10 | Sas Institute Inc. | Systems and methods for clustering time series data based on forecast distributions |
CN105894027A (en) * | 2016-03-31 | 2016-08-24 | 华北电力科学研究院有限责任公司 | Principal element degree of association sensor fault detection method and apparatus based on density clustering |
CN106250709A (en) * | 2016-08-18 | 2016-12-21 | 中国船舶重工集团公司第七�三研究所 | Gas turbine abnormality detection based on sensors association network and fault diagnosis algorithm |
CN106407589A (en) * | 2016-09-29 | 2017-02-15 | 北京岳能科技股份有限公司 | Wind turbine state evaluation and prediction method and system |
-
2017
- 2017-04-20 CN CN201710261505.7A patent/CN107146004B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9336493B2 (en) * | 2011-06-06 | 2016-05-10 | Sas Institute Inc. | Systems and methods for clustering time series data based on forecast distributions |
CN102323751A (en) * | 2011-06-28 | 2012-01-18 | 浙江大学 | Preparatory grinding system control method based on fuzzy intelligence control and optimization method |
CN102319612A (en) * | 2011-07-05 | 2012-01-18 | 中国科学院沈阳自动化研究所 | Method for intelligently controlling pressure difference of cement raw meal vertical mill |
CN103028480A (en) * | 2012-12-10 | 2013-04-10 | 上海凯盛节能工程技术有限公司 | Intelligent control system for vertical mill based on fuzzy PID (proportion integration differentiation) algorithm |
CN105894027A (en) * | 2016-03-31 | 2016-08-24 | 华北电力科学研究院有限责任公司 | Principal element degree of association sensor fault detection method and apparatus based on density clustering |
CN106250709A (en) * | 2016-08-18 | 2016-12-21 | 中国船舶重工集团公司第七�三研究所 | Gas turbine abnormality detection based on sensors association network and fault diagnosis algorithm |
CN106407589A (en) * | 2016-09-29 | 2017-02-15 | 北京岳能科技股份有限公司 | Wind turbine state evaluation and prediction method and system |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108052970A (en) * | 2017-12-08 | 2018-05-18 | 深圳市智物联网络有限公司 | A kind of data processing method and processing equipment |
CN109886443A (en) * | 2017-12-26 | 2019-06-14 | 广东电网有限责任公司电力调度控制中心 | A kind of failure Prediction System based on gray system theory |
CN109376919A (en) * | 2018-10-12 | 2019-02-22 | 西安科技大学 | A kind of prediction technique of coal mine fully-mechanized mining working gas emission |
CN109376919B (en) * | 2018-10-12 | 2022-04-26 | 西安科技大学 | Prediction method for gas emission quantity of coal mine fully-mechanized coal mining face |
TWI771531B (en) * | 2018-11-23 | 2022-07-21 | 中華電信股份有限公司 | Method and system for predicting system health using machine learning |
CN109507535B (en) * | 2018-12-10 | 2021-02-05 | 国网河南省电力公司电力科学研究院 | Method and device for predicting operation stage and operation life of transformer substation grounding grid |
CN109507535A (en) * | 2018-12-10 | 2019-03-22 | 国网河南省电力公司电力科学研究院 | Grounding net of transformer substation operation phase and service life prediction technique and device |
CN109902871A (en) * | 2019-02-27 | 2019-06-18 | 万洲电气股份有限公司 | A kind of intelligent optimization energy conserving system of combination enterprise production line differentiation feature |
CN109902871B (en) * | 2019-02-27 | 2023-02-17 | 万洲电气股份有限公司 | Intelligent optimization energy-saving system combining differentiation characteristics of enterprise production line |
CN110086829A (en) * | 2019-05-14 | 2019-08-02 | 四川长虹电器股份有限公司 | A method of Internet of Things unusual checking is carried out based on machine learning techniques |
CN110245842A (en) * | 2019-05-24 | 2019-09-17 | 电子科技大学 | A kind of production line Risk Scheduling method of equipment oriented burst major break down |
CN110245842B (en) * | 2019-05-24 | 2022-11-25 | 电子科技大学 | Production line risk scheduling method for sudden large faults of equipment |
CN110569279A (en) * | 2019-08-23 | 2019-12-13 | 长沙学院 | Time series signal reconstruction method based on variable projection algorithm |
CN110569279B (en) * | 2019-08-23 | 2021-04-16 | 长沙学院 | Time series signal reconstruction method based on variable projection algorithm |
CN110618984B (en) * | 2019-08-27 | 2023-02-03 | 西安因联信息科技有限公司 | Shutdown vibration data cleaning method |
CN110618984A (en) * | 2019-08-27 | 2019-12-27 | 西安因联信息科技有限公司 | Shutdown vibration data cleaning method |
CN111259730B (en) * | 2019-12-31 | 2022-08-23 | 杭州安脉盛智能技术有限公司 | State monitoring method and system based on multivariate state estimation |
CN111259730A (en) * | 2019-12-31 | 2020-06-09 | 杭州安脉盛智能技术有限公司 | State monitoring method and system based on multivariate state estimation |
CN111639842A (en) * | 2020-05-20 | 2020-09-08 | 湖北博华自动化系统工程有限公司 | Equipment health evaluation method, evaluation system and equipment health prediction method |
CN112015619A (en) * | 2020-08-19 | 2020-12-01 | 浙江无极互联科技有限公司 | Method for optimizing and screening core key indexes of system through parameters |
CN112015620A (en) * | 2020-08-19 | 2020-12-01 | 浙江无极互联科技有限公司 | Method for automatically adjusting and optimizing parameters of website service end system |
CN112365458A (en) * | 2020-11-02 | 2021-02-12 | 中材邦业(杭州)智能技术有限公司 | Grate cooler snowman identification method and system based on ANN neural network |
CN112381965A (en) * | 2020-11-03 | 2021-02-19 | 浙大城市学院 | Aeroengine health state identification system and method based on data mining |
CN112699926A (en) * | 2020-12-25 | 2021-04-23 | 浙江中控技术股份有限公司 | Method for identifying saturated grinding abnormity of cement raw material vertical mill based on artificial intelligence technology |
CN112699926B (en) * | 2020-12-25 | 2023-01-20 | 浙江中控技术股份有限公司 | Method for recognizing saturated grinding abnormity of cement raw material vertical mill based on artificial intelligence technology |
CN115964907A (en) * | 2023-03-17 | 2023-04-14 | 中国人民解放军火箭军工程大学 | Complex system health trend prediction method and system, electronic device and storage medium |
CN115964907B (en) * | 2023-03-17 | 2023-12-01 | 中国人民解放军火箭军工程大学 | Complex system health trend prediction method, system, electronic equipment and storage medium |
CN117436353A (en) * | 2023-12-21 | 2024-01-23 | 泰安维创游乐设备有限公司 | Intelligent recreation facility fault prediction method based on big data |
CN117436353B (en) * | 2023-12-21 | 2024-03-22 | 泰安维创游乐设备有限公司 | Intelligent recreation facility fault prediction method based on big data |
Also Published As
Publication number | Publication date |
---|---|
CN107146004B (en) | 2018-02-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107146004B (en) | A kind of slag milling system health status identifying system and method based on data mining | |
CN106990763B (en) | A kind of Vertical Mill operation regulator control system and method based on data mining | |
CN107239066B (en) | A kind of Vertical Mill operation closed-loop control device and method based on data mining | |
CN108584592B (en) | A kind of shock of elevator car abnormity early warning method based on time series predicting model | |
Capozzoli et al. | Fault detection analysis using data mining techniques for a cluster of smart office buildings | |
JP5538597B2 (en) | Anomaly detection method and anomaly detection system | |
EP2752722B1 (en) | Facility state monitoring method and device for same | |
JP5301310B2 (en) | Anomaly detection method and anomaly detection system | |
US9483049B2 (en) | Anomaly detection and diagnosis/prognosis method, anomaly detection and diagnosis/prognosis system, and anomaly detection and diagnosis/prognosis program | |
JP5501903B2 (en) | Anomaly detection method and system | |
WO2011086805A1 (en) | Anomaly detection method and anomaly detection system | |
Roemer et al. | An overview of selected prognostic technologies with application to engine health management | |
CN109948860A (en) | A kind of mechanical system method for predicting residual useful life and system | |
US20070088550A1 (en) | Method for predictive maintenance of a machine | |
CN111178553A (en) | Industrial equipment health trend analysis method and system based on ARIMA and LSTM algorithms | |
KR101948604B1 (en) | Method and device for equipment health monitoring based on sensor clustering | |
JPH06170696A (en) | System and method for using real time expert system for diagnosing tool life and predicted tool wear | |
CN112381965A (en) | Aeroengine health state identification system and method based on data mining | |
CN116976707B (en) | User electricity consumption data anomaly analysis method and system based on electricity consumption data acquisition | |
Barbieri et al. | Sensor-based degradation prediction and prognostics for remaining useful life estimation: Validation on experimental data of electric motors | |
Ramezani et al. | Prognostics and health management in machinery: A review of methodologies for RUL prediction and roadmap | |
JP2014056598A (en) | Abnormality detection method and its system | |
Zhu et al. | A data-driven decision-making framework for online control of vertical roller mill | |
CN117076869B (en) | Time-frequency domain fusion fault diagnosis method and system for rotary machine | |
CN105302476B (en) | A kind of reliability data online acquisition for nuclear power plant equipment analyzes storage system and its storage method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |