CN111905396B

CN111905396B - Rectification process real-time monitoring method based on online sampling data driving

Info

Publication number: CN111905396B
Application number: CN202010570654.3A
Authority: CN
Inventors: 蓝艇; 其他发明人请求不公开姓名
Original assignee: Ningbo University
Current assignee: Dragon Totem Technology Hefei Co ltd; Tianjin Tianrun Chemical Technology Co ltd
Priority date: 2020-06-13
Filing date: 2020-06-13
Publication date: 2021-11-19
Anticipated expiration: 2040-06-13
Also published as: CN111905396A

Abstract

The invention discloses a real-time monitoring method for a rectification process based on online sampling data driving, which aims to drive characteristic transformation in real time according to online sampling data so as to realize real-time monitoring on the running state of a rectification tower by using the most representative characteristic. Specifically, the method does not adopt fixed projection transformation vectors to extract features, but utilizes online data to drive feature analysis and extraction in real time. Compared with the traditional method, the method is simple and clear to implement, and has few off-line modeling stages. Since the off-line modeling phase mainly involves a normalization process and the computation of basis matrices. Secondly, the method searches the projection transformation vector which can best distinguish the sample data from the normal working condition data aiming at the sample data measured on line. In this respect, the characteristic components extracted by the method are most beneficial to monitoring fault data. Finally, the superiority of the method in real-time monitoring of the rectifying tower process is verified through a specific embodiment.

Description

Rectification process real-time monitoring method based on online sampling data driving

Technical Field

The invention relates to a method for monitoring the running state of a chemical process, in particular to a method for monitoring the real-time rectification process based on-line sampling data drive.

Background

The rectifying tower is widely applied to the petrochemical industry, the main function of the rectifying tower is to realize gas-liquid separation of substances, and the control and optimization of the rectifying tower are greatly developed in recent years. From the aspect of ensuring the stability of production, the rectification tower realizes the separation of substances and plays a connected key role. Therefore, it is very important to maintain the distillation column operating under the desired normal conditions, since it directly affects the subsequent processing steps. In recent years, in the large background of intelligent manufacturing, the realization of intelligent chemical production by using chemical industry big data has received more and more attention from the business industry and the academia. The data-driven process monitoring method is to identify the fault working condition and the normal working condition by using the characteristic change of the sampled data. For the rectifying tower, the measuring instruments mainly involved comprise flow, temperature of each tower plate layer, cooling water feeding flow temperature and the like, and abnormal changes of data information can indicate that rectifying tower equipment deviates from normal working conditions and enters fault working conditions. It should be noted that the fault condition herein refers to a violent operation that is not in an expected state, and of course, includes a failure factor of the equipment.

Data-driven process monitoring methods have been widely used in the industries of machine manufacturing, biochemical engineering, medicine, and the like. A set of process technology frameworks based on Principal Component Analysis (PCA) has been established. The essence of the data-driven process monitoring technology is that useful hidden features are extracted by mining the features of the sampled data, and whether the process enters a fault working condition is reflected by monitoring the change of the hidden features. Therefore, how to extract useful latent features becomes a key to whether such a method technology can successfully detect the fault condition. In the prior patent and scientific research literature, the hidden characteristics of normal working condition data, such as variance, local neighbor structure, non-Gaussian characteristics and the like, are excavated through a specific algorithm. However, it is not known whether these data features play a role in process monitoring. This is mainly because normal condition data is abundant and fault data is seriously deficient, and we seem to be able to perform feature analysis and extraction only on normal data.

In view of the implementation idea of a data-driven process monitoring method, particularly a statistical process monitoring method, a projection transformation vector is generally used to transform sampling data to obtain corresponding feature components, and then a monitoring index is constructed by using a distance (such as a squared mahalanobis distance or a squared euclidean distance) to monitor changes in the feature components. If the monitoring index value exceeds a certain threshold value, judging that a fault occurs. From this perspective, the larger the value of the monitoring index, the more distinguishable the difference between normal and fault. Therefore, although normal working condition data are abundant and fault data are missing, corresponding feature analysis and extraction can be driven by data obtained through online real-time measurement, and therefore the monitoring of the process running state is achieved through the most favorable feature transformation.

Disclosure of Invention

The invention aims to solve the main technical problems that: and how to drive the characteristic transformation task in real time according to the online sampling data, thereby realizing the real-time monitoring of the running state of the rectifying tower by using the most representative characteristic. Specifically, the method does not adopt fixed projection transformation vectors to extract features, but utilizes online data to drive feature analysis and extraction in real time.

The technical scheme adopted by the method for solving the problems is as follows: a real-time monitoring method for a rectification process based on online sampling data driving comprises the following steps:

step (1): by utilizing a measuring instrument arranged on the rectifying tower equipment, N sample data x are collected when the rectifying tower is in a normal operation state₁，x₂，…，x_NWherein the sample data x at the ith sampling time_i∈R^m×1The method is composed of m sampling data, and specifically comprises the following steps: liquid level in the column bottom, pressure in the column bottom, product flow at the bottom of the column bottom, feed flow, feed temperature, top reflux flow, condenser liquid level, and temperature of each column plate layer, R^m×1Representing a real vector of dimension m x 1, i e {1, 2, …, N }.

Step (2): for N sample data x₁，x₂，…，x_NPerforming normalization to obtain N m × 1-dimensional data vectors

And build it into a matrix

Post-computation basis matrix C ═ (XX)^T)^-1/2Wherein the upper index T represents the transpose of a matrix or vector.

By this point, the offline modeling phase has been completed. That is, the offline modeling phase of the method of the present invention involves only two tasks: and (5) data acquisition and standardized processing under normal working conditions. And then driving a characteristic analysis and extraction task in real time according to the data obtained by the on-line new measurement.

And (3): at the latest sampling time t, a data vector x consisting of m sampling data is obtained by measuring with a measuring instrument arranged on the rectifying tower equipment_t∈R^m×1And carrying out the same standardization processing as the step (2) to obtain a new data vector

And (4): according to the formula

After the matrix G is calculated, solving the eigenvector p belonging to the maximum eigenvalue of G and belonging to R^m×1。

It is worth noting that solving the eigenvalue problem in step (4) is actually the optimal solution to solve the optimization problem as shown below:

in the above formula, after the normal working condition data is transformed by the projection transformation vector w, the variance or length is 1. And the objective function aims to ensure that the data vectors sampled on line are better as far as the far point after the same projective transformation. In other words, the online data is separated from the normal working condition data as much as possible, so that the characteristic components most suitable for monitoring the fault can be extracted.

The optimization solution of the formula I can be completed by a classical Lagrange multiplier method, and an intermediate quantity p (XX) needs to be introduced first^T)^1/2w makes a transition, and pay attention to

Where tr () represents finding the trace of the matrix within the parenthesis.

In addition, the calculation of the eigenvector corresponding to the maximum eigenvalue of the matrix G can be realized by a numerical solution, which is specifically described below.

Step (4.1): the initialization feature vector p is a real number vector of arbitrary m × 1 dimensions.

Step (4.2): after the feature vector p is updated according to the formula p ═ Gp, the feature vector p is normalized, wherein p ═ p/| | | | | p | | | represents the length of the calculated feature vector p.

Step (4.3): judging whether the characteristic vector p is converged; if so, obtaining the characteristic vector p ∈ R corresponding to the maximum characteristic value of G^m×1(ii) a If not, returning to the step (4.2).

And (5): calculating a projective transformation vector w epsilon R according to the formula w-Cp^m×1Then, the monitoring index vector D ═ diag { X is calculated^Tww^TX, and determining the maximum value D in D_max。

And (6): according to the formula

Calculating a monitoring index D_tAnd judging whether D is satisfied_t≤D_maxIs there a If so, the rectifying tower operates normally at the current sampling moment, and the step (3) is returned to continue to monitor the operating state of the sample data at the next latest sampling moment; if not, step (7) is executed to decide whether to identify the fault.

And (7): returning to the step (3) to continue to monitor the running state of the sample data at the next latest sampling moment, and if the monitoring indexes at the continuous 3 sampling moments do not meet the judgment condition in the step (6), enabling the rectifying tower to enter a fault working condition and triggering a fault alarm; otherwise, returning to the step (3) to continue to monitor the running state of the next latest sampling moment.

The advantages and features of the method of the present invention are shown below.

First, the method of the present invention is very straightforward to implement and has few off-line modeling stages. Since the off-line modeling phase mainly involves a normalization process and the computation of basis matrices. Secondly, the method searches the projection transformation vector which can best distinguish the sample data from the normal working condition data aiming at the sample data measured on line. In this respect, the characteristic components extracted by the method are most beneficial to monitoring fault data. Finally, the superiority of the method in monitoring the rectifying tower is verified through the following specific implementation case.

Drawings

FIG. 1 is a schematic flow chart of the method of the present invention.

Fig. 2 is a schematic view of a real view and a schematic structural diagram of a rectifying tower device.

FIG. 3 is a comparative monitoring chart of the method of the present invention in monitoring the operation state of the rectifying tower.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

As shown in FIG. 1, the present invention discloses a real-time monitoring method for rectification process based on-line sampling data driving, and the following describes a specific implementation of the method of the present invention with reference to a specific application example.

As can be seen from the real view of the rectifying tower in FIG. 2, the rectifying tower is not only a single rectifying device, but also a matched reboiler at the bottom and a condenser at the top. As can be seen from the schematic structural diagram in fig. 2, the rectification column apparatus measuring instrument includes: flow meter, temperature instrument, three types of liquid level meter, there are 17 corresponding measured variables, specifically include: column bottom liquid level, column bottom pressure, column bottom product flow, feed temperature, top reflux flow, condenser liquid level, and temperature of each layer of trays (10 layers in total).

Step (1): by using a measuring instrument installed on the rectifying tower equipment, 1000 sample data x are acquired when the rectifying tower is in a normal operation state₁，x₂，…，x₁₀₀₀。

Step (2): for 1000 sample data x₁，x₂，…，x₁₀₀₀Performing normalization to obtain 1000 data vectors of 17 × 1 dimension

And build it into a matrix

Then, the base matrix C ═ XX is calculated^T)^-1/2。

After the off-line modeling stage is completed, the on-line state monitoring of the rectifying tower can be continuously carried out according to the following steps, and sample data at each new sampling moment is required to be utilized.

And (4): according to the formula

And (6): according to the formula

The conventional Principal Component Analysis (PCA) and Independent Component Analysis (ICA) were used to monitor the operating conditions of the rectification column, and the method of the present invention is illustrated in fig. 3 in comparison with the monitoring of PCA and ICA. When on-line monitoring is implemented, the rectifying tower equipment operates normally at the beginning of a period of time, and then a valve of cooling water of a condenser at the top of the rectifying tower has viscous fault, and as can be found from the graph in FIG. 3, the method is more sensitive to monitoring of fault working condition data. This is mainly because the method of the present invention can extract the characteristic components most beneficial to fault monitoring on line.

The above embodiments are merely illustrative of specific implementations of the present invention and are not intended to limit the present invention. Any modification of the present invention within the spirit of the present invention and the scope of the claims will fall within the scope of the present invention.

Claims

1. A real-time monitoring method for a rectification process based on online sampling data driving is characterized by comprising the following steps:

step (1): by utilizing a measuring instrument arranged on the rectifying tower equipment, N sample data x are collected when the rectifying tower is in a normal operation state₁，x₂，…，x_NWherein the sample data x at the ith sampling time_i∈R^m×1The method is composed of m sampling data, and specifically comprises the following steps: liquid level in the column bottom, pressure in the column bottom, product flow at the bottom of the column bottom, feed flow, feed temperature, top reflux flow, condenser liquid level, and temperature of each column plate layer, R^m×1Representing a real number vector of dimension m × 1, i ∈ {1, 2, …, N };

And build it into a matrix

Post-computation basis matrix C ═ (XX)^T)^-1/2Wherein, the upper label T represents the transposition of a matrix or a vector;

And (4): according to the formula

After the matrix G is calculated, solving the eigenvector p belonging to the maximum eigenvalue of G and belonging to R^m×1；

And (5): calculating a projective transformation vector w epsilon R according to the formula w-Cp^m×1Then, the monitoring index vector D ═ diag { X is calculated^Tww^TX, and determining the maximum value D in D_max；

And (6): according to the formula

Calculating a monitoring index D_tAnd judging whether D is satisfied_t≤D_maxIs there a If so, the rectifying tower operates normally at the current sampling moment, and the step (3) is returned to continue to monitor the operating state of the sample data at the next latest sampling moment; if not, executing the step (7) so as to decide whether to identify the fault;

2. The real-time rectification process monitoring method based on online sampling data driving according to claim 1, wherein the specific implementation process of solving the eigenvector corresponding to the maximum eigenvalue of the matrix G in the step (4) is as follows:

step (4.1): initializing a characteristic vector p as a real number vector of any dimension of m multiplied by 1;

step (4.2): after updating the feature vector p according to the formula p ═ Gp, performing normalization processing on the feature vector p ═ p/| | p | |, wherein | | | | p | | | represents the length of the calculated feature vector p;

step (4.3): judging whether the characteristic vector p is converged; if so, obtaining the characteristic vector p ∈ R corresponding to the maximum characteristic value of G^m×1(ii) a If not, go back toGo back to step (4.2).