CN105718754A - Method and device for generating dynamic alarm threshold value of parameters of refining process - Google Patents

Abstract

The invention provides a method and a device for generating a dynamic alarm threshold value of parameters of a refining process, and relates to the technical field of petroleum refining fault monitoring. The method comprises the following steps of obtaining historic data of the parameters of a refining process production device; determining initial historic training data of the parameters according to the length of a sliding window and the step length, and updating historic training data for once every other step length time from the initial historic training data; forming normalized historic training data; carrying out kernel density estimation on each parameter, and generating a probability density function of the parameters; determining a probability distribution function of each parameter; determining an alarm threshold value corresponding to each parameter according to the probability distribution function of each parameter and a preset alarm threshold confidence degree. According to the method for generating the dynamic alarm threshold value of the parameters of the refining process, provided by the invention, the dynamic alarm threshold value of the parameters can be obtained, and the problems that because a static model parameter threshold value method is adopted at present, and the alarm threshold value of the parameters is artificially regulated, false alarm or leakage alarm can be easily caused, and the sensitivity of abnormal monitoring can be easily reduced can be solved.

Description

A kind of generation method of refinery procedure parameter dynamic alert threshold value and device

Technical field

The present invention relates to oil refining malfunction monitoring technical field, particularly relate to generation method and the device of a kind of refinery procedure parameter dynamic alert threshold value.

Background technology

Currently, as typical process industrial, refinery process is complicated and process units quantity and of a great variety.In refinery process, refinery process and other devices in addition will be had influence on once break down.It could even be possible to cause some major accidents.Therefore, the production process in refinery process is monitored increasingly important.

At present, most of refinery industry adopt static models parameter threshold method that production process is carried out status monitoring, and the theoretical method based on static models parameter threshold is built upon it is assumed hereinafter that on: namely modeling data contains the data space of all normal refinery operating modes substantially, and refinery operating mode process is that stable state is constant.But, in actual refinery process, great majority all exist and slowly and normally drift about, i.e. slow time-varying feature, in addition actual refinery process has multi-state unstable state Interim, therefore static parameter alarm threshold value will be no longer applicable after production status changes, if continuing the static parameter alarm threshold value that application is original, then easily cause large-scale false alarm and fail to report police.It addition, in most refinery industry, parameter alarm threshold value is generally adopted manual adjustment, namely device produces director and suitably adjusts alarm threshold value to evade redundant warning or reduce according to operation needs or expand alarm threshold value scope according to current process condition.The setting of artificial parameter threshold value frequently can lead to alarm threshold value and updates not in time or update by mistake, thus causing new false alarm and failing to report police or reduce the susceptiveness of exception monitoring.

Visible, currently employed static models parameter threshold method, and can not have been met refinery industrial processes complicated and changeable by manual adjustment parameter alarm threshold value.

Summary of the invention

Embodiments of the invention provide generation method and the device of a kind of refinery procedure parameter dynamic alert threshold value, to solve currently employed static models parameter threshold method, and by manual adjustment parameter alarm threshold value, it is easy to cause false alarm or fail to report police, and the problem being easily reduced the sensitivity of exception monitoring.

For reaching above-mentioned purpose, the present invention adopts the following technical scheme that

A kind of generation method of refinery procedure parameter dynamic alert threshold value, including:

Obtain the historical data of the parameter of refinery process process units；

Determine the sliding window length corresponding to the historical data of described parameter and step-length；

According to described sliding window length and step-length, it is determined that the initial history training data of parameter, and update a historic training data from described initial history training data every a step-length time；Described initial history training data is the historical data of the parameter of the sliding window length before the initial time determining historic training data；

Described historic training data is arranged to normalized, formed normalization historic training data；

According to described normalization historic training data, each parameter is carried out Density Estimator, generate the probability density function of each parameter；

Probability density function according to described each parameter, it is determined that the probability-distribution function of each parameter；

Probability-distribution function according to described each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter.

Concrete, according to described sliding window length and step-length, it is determined that the initial history training data of parameter, and update a historic training data from described initial history training data every a step-length time, including:

The initial history training data matrix determining described parameter is:Wherein, l₀For historic training data length；M is the process variable number of described historic training data；

Updating once described historic training data every a step-length D time, the historic training data after kth time renewal is:l_kThe length of the historic training data after updating for kth time.

Concrete, described historic training data is arranged to normalized, form normalization historic training data, including:

According to formula:

${x_{i,j}}^{\′}=\frac{x_{i,j}-x_{min}\left(j\right)}{x_{max}\left(j\right)-x_{min}\left(j\right)}$

Determine described normalization historic training data；Wherein, x_i,j' in described normalization historic training data i-th row jth row data；X_i,jFor historic training data X_kIn i-th row jth row data；X_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Concrete, according to described normalization historic training data, each parameter is carried out Density Estimator, generate the probability density function of each parameter, including:

According to formula:

$f\left(x\right)=\frac{1}{l_k{h}^d}\underset{i=1}{\overset{l_k}{\Σ}}k\left(\frac{x-{x_{i,j}}^{\′}}{h}\right)$

Each parameter is carried out Density Estimator, generates probability density function f (x) of each parameter；

Wherein,l_kThe length of the historic training data after updating for kth time；D is the space dimensionality pre-set；For gaussian kernel function,Normalization historic training data average for jth parameter.

Concrete, according to described each parameter probability density function, it is determined that the probability-distribution function of each parameter, including:

According to formula:

$F\left(x\right)=\underset{-\mathrm{\∞}}{\overset{x}{\∫}}f\left(x\right)dx$

Determine probability-distribution function F (x) of each parameter.

Concrete, according to the probability-distribution function of described each parameter and the alarm threshold value confidence level that pre-sets, it is determined that the alarm threshold value corresponding to each parameter, including:

According to formula:

F (x)=α

Determine the first parameter value S corresponding to each parameter₁；Wherein, α is the alarm threshold value confidence level pre-set；

According to formula:

x₁=S₁×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value lower limit x corresponding to jth parameter₁；Wherein, x_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Concrete, according to the probability-distribution function of described each parameter and the alarm threshold value confidence level that pre-sets, it is determined that the alarm threshold value corresponding to each parameter, also include:

According to formula:

F (x)=1-α

Determine the second parameter value S corresponding to each parameter₂；

According to formula:

x₂=S₂×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value upper limit x corresponding to jth parameter₂。

A kind of generation device of refinery procedure parameter dynamic alert threshold value, including:

Historical data acquiring unit, for obtaining the historical data of the parameter of refinery process process units；

Sliding window length and step size determination unit, for determining the sliding window length corresponding to historical data and the step-length of described parameter；

Historic training data updating block, for according to described sliding window length and step-length, it is determined that the initial history training data of parameter, and updates a historic training data from described initial history training data every a step-length time；Described initial history training data is the historical data of the parameter of the sliding window length before the initial time determining historic training data；

Normalized unit, for described historic training data arranged to normalized, form normalization historic training data；

Density Estimator unit, for each parameter being carried out Density Estimator according to described normalization historic training data, generates the probability density function of each parameter；

Probability-distribution function determines unit, for the probability density function according to described each parameter, it is determined that the probability-distribution function of each parameter；

Alarm threshold value determines unit, is used for the probability-distribution function according to described each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter.

Additionally, described historic training data updating block, specifically for:

The initial history training data matrix determining described parameter is:Wherein, l₀For historic training data length；M is the process variable number of described historic training data；

Updating once described historic training data every a step-length D time, the historic training data after kth time renewal is:l_kThe length of the historic training data after updating for kth time.

Additionally, described normalized unit, specifically for:

According to formula:

${x_{i,j}}^{\′}=\frac{x_{i,j}-x_{min}\left(j\right)}{x_{max}\left(j\right)-x_{min}\left(j\right)}$

Determine described normalization historic training data；Wherein, x_i,j' in described normalization historic training data i-th row jth row data；X_i,jFor historic training data X_kIn i-th row jth row data；X_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Additionally, described Density Estimator unit, specifically for:

According to formula:

$f\left(x\right)=\frac{1}{l_k{h}^d}\underset{i=1}{\overset{l_k}{\Σ}}k\left(\frac{x-{x_{i,j}}^{\′}}{h}\right)$

Each parameter is carried out Density Estimator, generates probability density function f (x) of each parameter；

Wherein,l_kThe length of the historic training data after updating for kth time；D is the space dimensionality pre-set；For gaussian kernel function,Normalization historic training data average for jth parameter.

Additionally, described probability-distribution function determines unit, specifically for:

According to formula:

$F\left(x\right)=\underset{-\mathrm{\∞}}{\overset{x}{\∫}}f\left(x\right)dx$

Determine probability-distribution function F (x) of each parameter.

It addition, described alarm threshold value determines unit, specifically for:

According to formula:

F (x)=α

Determine the first parameter value S corresponding to each parameter₁；Wherein, α is the alarm threshold value confidence level pre-set；

According to formula:

x₁=S₁×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value lower limit x corresponding to jth parameter₁；Wherein, x_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Further, described alarm threshold value determines unit, is additionally operable to:

According to formula:

F (x)=1-α

Determine the second parameter value S corresponding to each parameter₂；

According to formula:

x₂=S₂×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value upper limit x corresponding to jth parameter₂。

The generation method of a kind of refinery procedure parameter dynamic alert threshold value that the embodiment of the present invention provides and device, the historic training data of the parameter by constantly updating carries out Density Estimator, generate the probability density function of each parameter, so that it is determined that the probability-distribution function of each parameter, and then determine the alarm threshold value corresponding to each parameter.Present invention transient process suitable in refinery process, multiple operating modes process and operating mode be the malfunction monitoring of normal Drift Process slowly, significantly reduces false alarm rate and improves the susceptiveness of malfunction monitoring.The present invention can solve currently employed static models parameter threshold method, and by manual adjustment parameter alarm threshold value, it is easy to cause false alarm or fail to report police, and the problem being easily reduced the sensitivity of exception monitoring.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, the accompanying drawing used required in embodiment or description of the prior art will be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.

The flow chart one of the generation method of a kind of refinery procedure parameter dynamic alert threshold value that Fig. 1 provides for the embodiment of the present invention；

The flowchart 2 of the generation method of a kind of refinery procedure parameter dynamic alert threshold value that Fig. 2 provides for the embodiment of the present invention；

Fig. 3 is the historic training data curve synoptic diagram in the embodiment of the present invention；

Fig. 4 is the historic training data curve synoptic diagram after the normalization in the embodiment of the present invention；

Fig. 5 is the probability-distribution function curve synoptic diagram of the parameter obtained after Density Estimator in the embodiment of the present invention；

Fig. 6 is the schematic diagram of the dynamic state of parameters threshold monitor curve of the propane tower raw material surge tank level monitoring process in the embodiment of the present invention and static parameter threshold monitor curve；

Fig. 7 is the schematic diagram of the dynamic state of parameters threshold monitor curve of the monitoring process of the propylene anti-tower temperature of bottom reboiler gas hydrocarbon in the embodiment of the present invention and static parameter threshold monitor curve；

The structural representation generating device of a kind of refinery procedure parameter dynamic alert threshold value that Fig. 8 provides for the embodiment of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, the every other embodiment that those of ordinary skill in the art obtain under not making creative work premise, broadly fall into the scope of protection of the invention.

As it is shown in figure 1, the embodiment of the present invention provides a kind of generation method of refinery procedure parameter dynamic alert threshold value, including:

Step 101, obtain refinery process process units parameters history data.

Step 102, the sliding window length corresponding to historical data determining described parameter and step-length.

Step 103, according to described sliding window length and step-length, it is determined that the initial history training data of parameter, and update a historic training data every step-length time from described initial history training data.

Wherein, described initial history training data is the historical data of the parameter of the sliding window length before the initial time determining historic training data.

Step 104, this historic training data is arranged to normalized, formed normalization historic training data.

Step 105, according to this normalization historic training data, each parameter is carried out Density Estimator, generate the probability density function of each parameter.

Step 106, probability density function according to this each parameter, it is determined that the probability-distribution function of each parameter.

Step 107, according to the probability-distribution function of this each parameter and the alarm threshold value confidence level that pre-sets, it is determined that the alarm threshold value corresponding to each parameter.

The generation method of a kind of refinery procedure parameter dynamic alert threshold value that the embodiment of the present invention provides, the historic training data of the parameter by constantly updating carries out Density Estimator, generate the probability density function of each parameter, so that it is determined that the probability-distribution function of each parameter, and then determine that the alarm threshold value corresponding to each parameter, determined alarm threshold value are dynamic alarm threshold value.Present invention transient process suitable in refinery process, multiple operating modes process and operating mode be the malfunction monitoring of normal Drift Process slowly, significantly reduces false alarm rate and improves the susceptiveness of malfunction monitoring.The present invention can solve currently employed static models parameter threshold method, and by manual adjustment parameter alarm threshold value, it is easy to cause false alarm or fail to report police, and the problem being easily reduced the sensitivity of exception monitoring.

In order to make those skilled in the art be better understood by the present invention, an embodiment specifically is set forth below, as in figure 2 it is shown, the generation method of a kind of refinery procedure parameter dynamic alert threshold value of embodiment of the present invention offer, including:

Step 201, obtain the historical data of parameter of refinery process process units.

Step 202, the sliding window length corresponding to historical data determining described parameter and step-length.

Generally, it is possible to the degree of fluctuation according to parameter, it is determined that the sliding window length of historic training data renewal and step-length.Such as, sliding window length can choose the data length of production status even running as this sliding window length L according to refinery process-field experience for L.Step-length D refers to the data length that when each historic training data updates, window data adds or abandons, if working conditions change is relatively slow, can suitably increase step-length D, if working conditions change more frequently can suitably reduce step-length D.

Step 203, determine the initial history training data matrix of parameter.

Herein, the initial history training data matrix of parameter can be expressed as Wherein, l₀For historic training data length；M is the process variable number of the historic training data of parameter.Certainly, X herein₀For the historic training data after rejecting fault data.

Step 204, updating once this historic training data every a step-length D time.

Along with sliding window slides, update a historic training data when a step-length D, then the historic training data of the monitoring parameter after kth time renewal can be expressed as: Wherein, l_kHistoric training data length after updating for kth time.

Step 205, this historic training data is arranged to normalized, formed normalization historic training data.

Herein, it is possible to according to formula:

${x_{i,j}}^{\′}=\frac{x_{i,j}-x_{min}\left(j\right)}{x_{max}\left(j\right)-x_{min}\left(j\right)}$

Determine normalization historic training data；Wherein, x_i,j' in this normalization historic training data i-th row jth row data；X_i,jFor historic training data X_kIn i-th row jth row data；X_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Step 206, according to this normalization historic training data, each parameter is carried out Density Estimator, generate the probability density function of each parameter.

Concrete, herein can according to formula:

$f\left(x\right)=\frac{1}{l_k{h}^d}\underset{i=1}{\overset{l_k}{\Σ}}k\left(\frac{x-{x_{i,j}}^{\′}}{h}\right)$

Each parameter is carried out Density Estimator, generates probability density function f (x) of each parameter.

Wherein,l_kThe length of the historic training data after updating for kth time；D is the space dimensionality pre-set, and in embodiments of the present invention, this space dimensionality can be 2, but is not only limited to this；For gaussian kernel function,Normalization historic training data average for jth parameter.

Step 207, probability density function according to this each parameter, it is determined that the probability-distribution function of each parameter.

Herein, it is possible to according to formula:

$F\left(x\right)=\underset{-\mathrm{\∞}}{\overset{x}{\∫}}f\left(x\right)dx$

Determine probability-distribution function F (x) of each parameter.

Step 208, according to the probability-distribution function of this each parameter and the alarm threshold value confidence level that pre-sets, it is determined that the first parameter value S corresponding to each parameter₁With the second parameter value S₂。

Herein, it is possible to according to formula:

F (x)=α

Determine the first parameter value S corresponding to each parameter₁.Wherein, α is the alarm threshold value confidence level pre-set；In embodiments of the present invention, α can be 0.05, but is not only limited to this；

Furthermore it is possible to according to formula:

F (x)=1-α

Determine the second parameter value S corresponding to each parameter₂。

Step 209, determine the alarm threshold value lower limit corresponding to jth parameter and the alarm threshold value upper limit.

Herein, it is possible to according to formula:

x₁=S₁×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value lower limit x corresponding to jth parameter₁；Wherein, x_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Furthermore it is possible to according to formula:

x₂=S₂×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value upper limit x corresponding to jth parameter₂。

In order to make those skilled in the art be better understood by the present invention, the application example of above-mentioned steps 201 a to step 209 is set forth below.At this, exemplary embodiment and the explanation thereof of the present invention are only used for explaining the present invention, but not as a limitation of the invention.Such as with the key parameter of common refinery process units gas separation unit for object of study, use collection in worksite data, carry out procedure parameter and monitor in real time.

Monitor in real time for propane tower raw material surge tank liquid level, 5 seconds each sampling intervals, gather 7445 samples of data, dynamic state of parameters threshold monitor is adopted from 5000 samplings, describe the invention process step in detail for first time threshold calculations, hereafter adopt slip window sampling to carry out a threshold calculations training data every 5 times (i.e. a step-length) sampling and update.

First time dynamic state of parameters threshold calculations adopts front 5000 sampled datas to be training data (shown in 5000 sampled datas as front in Fig. 3), by normal data normalization after rejecting fault data, historic training data curve after normalization is as shown in Figure 4, the probability-distribution function curve of parameter is obtained as shown in Figure 5 after Density Estimator, by confidence alpha=0.05, obtain the first parameter S₁=0.0060 and the second parameter S₂=0.9871, so calculate obtain parameter alarm threshold value bound respectively x₁=56.7918, x₂=63.3086.

As Fig. 6 (a) part show dynamic state of parameters threshold monitor curve, as Fig. 6 (b) part show static parameter threshold value of the prior art (alarm threshold value that DCS system is arranged) monitoring curve.All there is twice ultralow warning in Fig. 6 (a), (b) two parts, b () part exceeds DCS low alarm setting when the 761st sampling, and adopt dynamic threshold of (a) part to shift to an earlier date 1 sampling (5 seconds) when monitoring and trigger warning；B () part exceeds DCS low alarm setting when the 2055th sampling, and adopt dynamic threshold of (a) part to shift to an earlier date 7 samplings (35 seconds) when monitoring and trigger warning.Visible, this example illustrates that the procedure parameter exception monitoring sensitivity of the present invention is higher.

If Fig. 7 is the monitoring curve (lasting about 5 days) of the propylene anti-tower temperature of bottom reboiler gas hydrocarbon, Fig. 7 (a) part adopts dynamic threshold monitoring, Fig. 7 (b) part adopts static threshold monitoring, there are slowly transition to 40000 sampling period production status 30000 samplings, during 36806 samplings, parameter is beyond static alarms lower limit, transitting in another operating mode, static parameter alarm threshold value original under new operating mode is no longer applicable, and false alarm rate is 46.5%.And dynamic state of parameters threshold value adjusts in real time according to historical data, significantly reducing the false alarm rate in variable working condition situation, false alarm rate is only 0.071%.

Can be seen that from examples detailed above analysis, the optimization method that the refinery process dynamics alarm threshold value of the present invention generates is adopted to improve abnormal parameters identification susceptiveness, relatively static threshold can accurately and in advance trigger warning, and effectively reduce false alarm rate for multi-state, transient process and the slow time-varying process present invention, there is the very strong suitability.Absolutely prove the feasibility of refinery procedure parameter dynamic alert threshold value generation method in the embodiment of the present invention.

The generation method of a kind of refinery procedure parameter dynamic alert threshold value that the embodiment of the present invention provides, the historic training data of the parameter by constantly updating carries out Density Estimator, generate the probability density function of each parameter, so that it is determined that the probability-distribution function of each parameter, and then determine that the alarm threshold value corresponding to each parameter, determined alarm threshold value are dynamic alert threshold value.Present invention transient process suitable in refinery process, multiple operating modes process and operating mode be the malfunction monitoring of normal Drift Process slowly, significantly reduces false alarm rate and improves the susceptiveness of malfunction monitoring.The present invention can solve currently employed static models parameter threshold method, and by manual adjustment parameter alarm threshold value, it is easy to cause false alarm or fail to report police, and the problem being easily reduced the sensitivity of exception monitoring.

Corresponding to the embodiment of the method shown in above-mentioned Fig. 1 and Fig. 2, as shown in Figure 8, the embodiment of the present invention provides the generation device of a kind of refinery procedure parameter dynamic alert threshold value, including:

Historical data acquiring unit 301, for obtaining the historical data of the parameter of refinery process process units.

Sliding window length and step size determination unit 302, for determining the sliding window length corresponding to historical data and the step-length of described parameter.

Historic training data updating block 303, for according to described sliding window length and step-length, it is determined that the initial history training data of parameter, and updates a historic training data from described initial history training data every a step-length time；Described initial history training data is the historical data of the parameter of the sliding window length before the initial time determining historic training data.

Normalized unit 304, for this historic training data arranged to normalized, form normalization historic training data.

Density Estimator unit 305, for each parameter being carried out Density Estimator according to this normalization history training evidence, generates the probability density function of each parameter.

Probability-distribution function determines unit 306, for the probability density function according to this each parameter, it is determined that the probability-distribution function of each parameter.

Alarm threshold value determines unit 307, is used for the probability-distribution function according to this each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter.

Additionally, this historic training data updating block 303, specifically for:

The initial history training data matrix determining this parameter is:Wherein, l₀For historic training data length；M is the process variable number of this historic training data.

Updating once this historic training data every a step-length D time, the historic training data after kth time renewal is:Lk is the length of the historic training data after kth time updates.

Additionally, this normalized unit 304, specifically for:

According to formula:

${x_{i,j}}^{\′}=\frac{x_{i,j}-x_{min}\left(j\right)}{x_{max}\left(j\right)-x_{min}\left(j\right)}$

Determine this normalization historic training data；Wherein, x_i,j' in this normalization historic training data i-th row jth row data；X_i,jFor historic training data X_kIn i-th row jth row data；X_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Additionally, this Density Estimator unit 305, specifically for:

According to formula:

$f\left(x\right)=\frac{1}{l_k{h}^d}\underset{i=1}{\overset{l_k}{\Σ}}k\left(\frac{x-{x_{i,j}}^{\′}}{h}\right)$

Each parameter is carried out Density Estimator, generates probability density function f (x) of each parameter.

Wherein,l_kThe length of the historic training data after updating for kth time；D is the space dimensionality pre-set；For gaussian kernel function,Normalization historic training data average for jth parameter.

Additionally, this probability-distribution function determines unit 306, specifically for:

According to formula:

$F\left(x\right)=\underset{-\mathrm{\∞}}{\overset{x}{\∫}}f\left(x\right)dx$

Determine probability-distribution function F (x) of each parameter.

It addition, this alarm threshold value determines unit 307, specifically for:

According to formula:

F (x)=α

Determine the first parameter value S corresponding to each parameter₁；Wherein, α is the alarm threshold value confidence level pre-set.

According to formula:

x₁=S₁×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value lower limit x corresponding to jth parameter₁；Wherein, x_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

Further, this alarm threshold value determines unit 307, it is also possible to be used for:

According to formula:

F (x)=1-α

Determine the second parameter value S corresponding to each parameter₂。

According to formula:

x₂=S₂×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value upper limit x corresponding to jth parameter₂。

What deserves to be explained is, the specific implementation generating device of a kind of refinery procedure parameter dynamic alert threshold value that the embodiment of the present invention provides may refer to embodiment of the method corresponding to above-mentioned Fig. 1 and Fig. 2, repeats no more herein.

The generation device of a kind of refinery procedure parameter dynamic alert threshold value that the embodiment of the present invention provides, the historic training data of the parameter by constantly updating carries out Density Estimator, generate the probability density function of each parameter, so that it is determined that the probability-distribution function of each parameter, and then determine the alarm threshold value corresponding to each parameter.Present invention transient process suitable in refinery process, multiple operating modes process and operating mode be the malfunction monitoring of normal Drift Process slowly, significantly reduces false alarm rate and improves the susceptiveness of malfunction monitoring.The present invention can solve currently employed static models parameter threshold method, and by manual adjustment parameter alarm threshold value, it is easy to cause false alarm or fail to report police, and the problem being easily reduced the sensitivity of exception monitoring.

Those skilled in the art are it should be appreciated that embodiments of the invention can be provided as method, system or computer program.Therefore, the present invention can adopt the form of complete hardware embodiment, complete software implementation or the embodiment in conjunction with software and hardware aspect.And, the present invention can adopt the form at one or more upper computer programs implemented of computer-usable storage medium (including but not limited to disk memory, CD-ROM, optical memory etc.) wherein including computer usable program code.

The present invention is that flow chart and/or block diagram with reference to method according to embodiments of the present invention, equipment (system) and computer program describe.It should be understood that can by the combination of the flow process in each flow process in computer program instructions flowchart and/or block diagram and/or square frame and flow chart and/or block diagram and/or square frame.These computer program instructions can be provided to produce a machine to the processor of general purpose computer, special-purpose computer, Embedded Processor or other programmable data processing device so that the instruction performed by the processor of computer or other programmable data processing device is produced for realizing the device of function specified in one flow process of flow chart or multiple flow process and/or one square frame of block diagram or multiple square frame.

These computer program instructions may be alternatively stored in and can guide in the computer-readable memory that computer or other programmable data processing device work in a specific way, the instruction making to be stored in this computer-readable memory produces to include the manufacture of command device, and this command device realizes the function specified in one flow process of flow chart or multiple flow process and/or one square frame of block diagram or multiple square frame.

These computer program instructions also can be loaded in computer or other programmable data processing device, make on computer or other programmable devices, to perform sequence of operations step to produce computer implemented process, thus the instruction performed on computer or other programmable devices provides for realizing the step of function specified in one flow process of flow chart or multiple flow process and/or one square frame of block diagram or multiple square frame.

Applying specific embodiment in the present invention principles of the invention and embodiment are set forth, the explanation of above example is only intended to help to understand method and the core concept thereof of the present invention；Simultaneously for one of ordinary skill in the art, according to the thought of the present invention, all will change in specific embodiments and applications, in sum, this specification content should not be construed as limitation of the present invention.

Claims (14)

1. the generation method of a refinery procedure parameter dynamic alert threshold value, it is characterised in that including:

Obtain the historical data of the parameter of refinery process process units；

Determine the sliding window length corresponding to the historical data of described parameter and step-length；

According to described sliding window length and step-length, it is determined that the initial history training data of parameter, and update a historic training data from described initial history training data every a step-length time；Described initial history training data is the historical data of the parameter of the sliding window length before the initial time determining historic training data；

Described historic training data is arranged to normalized, formed normalization historic training data；

According to described normalization historic training data, each parameter is carried out Density Estimator, generate the probability density function of each parameter；

Probability density function according to described each parameter, it is determined that the probability-distribution function of each parameter；

Probability-distribution function according to described each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter.

2. the generation method of refinery procedure parameter dynamic alert threshold value according to claim 1, it is characterized in that, according to described sliding window length and step-length, determine the initial history training data of parameter, and update a historic training data from described initial history training data every a step-length time, including:

The initial history training data matrix determining described parameter is:Wherein, l₀For historic training data length；M is the process variable number of described historic training data；

Updating once described historic training data every a step-length D time, the historic training data after kth time renewal is:l_kThe length of the historic training data after updating for kth time.

3. the generation method of refinery procedure parameter dynamic alert threshold value according to claim 2, it is characterised in that described historic training data is arranged to normalized, formed normalization historic training data, including:

According to formula:

${x_{i,j}}^{\′}=\frac{x_{i,j}-x_{min}\left(j\right)}{x_{max}\left(j\right)-x_{min}\left(j\right)}$

Determine described normalization historic training data；Wherein, x_i,j' in described normalization historic training data i-th row jth row data；X_i,jFor historic training data X_kIn i-th row jth row data；X_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

4. the generation method of refinery procedure parameter dynamic alert threshold value according to claim 3, it is characterised in that according to described normalization historic training data, each parameter is carried out Density Estimator, generate the probability density function of each parameter, including:

According to formula:

$f\left(x\right)=\frac{1}{l_k{h}^d}\underset{i=1}{\overset{l_k}{\Σ}}k\left(\frac{x-{x_{i,j}}^{\′}}{h}\right)$

Each parameter is carried out Density Estimator, generates probability density function f (x) of each parameter；

Wherein,l_kThe length of the historic training data after updating for kth time；D is the space dimensionality pre-set；For gaussian kernel function, Normalization historic training data average for jth parameter.

5. the generation method of refinery procedure parameter dynamic alert threshold value according to claim 4, it is characterised in that the probability density function according to described each parameter, it is determined that the probability-distribution function of each parameter, including:

According to formula:

$F\left(x\right)=\underset{-\mathrm{\∞}}{\overset{x}{\∫}}f\left(x\right)dx$

Determine probability-distribution function F (x) of each parameter.

6. the generation method of refinery procedure parameter dynamic alert threshold value according to claim 5, it is characterized in that, probability-distribution function according to described each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter, including:

According to formula:

F (x)=α

Determine the first parameter value S corresponding to each parameter₁；Wherein, α is the alarm threshold value confidence level pre-set；

According to formula:

x₁=S₁×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value lower limit x corresponding to jth parameter₁；Wherein, x_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

7. the generation method of refinery procedure parameter dynamic alert threshold value according to claim 6, it is characterized in that, probability-distribution function according to described each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter, also include:

According to formula:

F (x)=1-α

Determine the second parameter value S corresponding to each parameter₂；

According to formula:

x₂=S₂×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value upper limit x corresponding to jth parameter₂。

8. the generation device of a refinery procedure parameter dynamic alert threshold value, it is characterised in that including:

Historical data acquiring unit, for obtaining the historical data of the parameter of refinery process process units；

Sliding window length and step size determination unit, for determining the sliding window length corresponding to historical data and the step-length of described parameter；

Historic training data updating block, for according to described sliding window length and step-length, it is determined that the initial history training data of parameter, and updates a historic training data from described initial history training data every a step-length time；Described initial history training data is the historical data of the parameter of the sliding window length before the initial time determining historic training data；

Normalized unit, for described historic training data arranged to normalized, form normalization historic training data；

Density Estimator unit, for each parameter being carried out Density Estimator according to described normalization historic training data, generates the probability density function of each parameter；

Probability-distribution function determines unit, for the probability density function according to described each parameter, it is determined that the probability-distribution function of each parameter；

Alarm threshold value determines unit, is used for the probability-distribution function according to described each parameter and the alarm threshold value confidence level pre-set, it is determined that the alarm threshold value corresponding to each parameter.

9. the generation device of refinery procedure parameter dynamic alert threshold value according to claim 8, it is characterised in that described historic training data updating block, specifically for:

The initial history training data matrix determining described parameter is:Wherein, l₀For historic training data length；M is the process variable number of described historic training data；

Updating once described historic training data every a step-length D time, the historic training data after kth time renewal is:l_kThe length of the historic training data after updating for kth time.

10. the generation device of refinery procedure parameter dynamic alert threshold value according to claim 9, it is characterised in that described normalized unit, specifically for:

According to formula:

${x_{i,j}}^{\′}=\frac{x_{i,j}-x_{min}\left(j\right)}{x_{max}\left(j\right)-x_{min}\left(j\right)}$

Determine described normalization historic training data；Wherein, x_i,j' in described normalization historic training data i-th row jth row data；X_i,jFor historic training data X_kIn i-th row jth row data；X_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

11. the generation device of refinery procedure parameter dynamic alert threshold value according to claim 10, it is characterised in that described Density Estimator unit, specifically for:

According to formula:

$f\left(x\right)=\frac{1}{l_k{h}^d}\underset{i=1}{\overset{l_k}{\Σ}}k\left(\frac{x-{x_{i,j}}^{\′}}{h}\right)$

Each parameter is carried out Density Estimator, generates probability density function f (x) of each parameter；

Wherein,l_kThe length of the historic training data after updating for kth time；D is the space dimensionality pre-set；For gaussian kernel function, Normalization historic training data average for jth parameter.

12. the generation device of refinery procedure parameter dynamic alert threshold value according to claim 11, it is characterised in that described probability-distribution function determines unit, specifically for:

According to formula:

$F\left(x\right)=\underset{-\mathrm{\∞}}{\overset{x}{\∫}}f\left(x\right)dx$

Determine probability-distribution function F (x) of each parameter.

13. the generation device of refinery procedure parameter dynamic alert threshold value according to claim 12, it is characterised in that described alarm threshold value determines unit, specifically for:

According to formula:

F (x)=α

Determine the first parameter value S corresponding to each parameter₁；Wherein, α is the alarm threshold value confidence level pre-set；

According to formula:

x₁=S₁×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value lower limit x corresponding to jth parameter₁；Wherein, x_minJ () is the minima of jth row historic training data；X_maxJ () is the maximum of jth row historic training data.

14. the generation device of refinery procedure parameter dynamic alert threshold value according to claim 13, it is characterised in that described alarm threshold value determines unit, is additionally operable to:

According to formula:

F (x)=1-α

Determine the second parameter value S corresponding to each parameter₂；

According to formula:

x₂=S₂×(x_max(j)-x_min(j))+x_min(j)

Determine the alarm threshold value upper limit x corresponding to jth parameter₂。