CN106933097B

CN106933097B - Chemical process fault diagnosis method based on multi-layer optimization PCC-SDG

Info

Publication number: CN106933097B
Application number: CN201710339420.6A
Authority: CN
Inventors: 田文德; 董玉玺; 任玉佳; 贾旭清; 王雪
Original assignee: Qingdao University of Science and Technology
Current assignee: Qingdao University of Science and Technology
Priority date: 2017-05-15
Filing date: 2017-05-15
Publication date: 2021-07-20
Anticipated expiration: 2037-05-15
Also published as: CN106933097A

Abstract

The invention discloses a chemical process fault diagnosis method based on multi-layer optimization PCC-SDG. The method comprises the steps of primarily optimizing selected variables by using a Pearson Correlation Coefficient (PCC) statistical index by taking a network topological structure of the whole process as a reference point, then selecting special variables with larger weights from a multi-layer correlation coefficient set by using a PCA weight thought, establishing an optimal PCC-SDG network by combining a Sign Directed Graph (SDG), and finally performing fault diagnosis on a rule for establishing an aggregation weight coefficient Q aiming at the optimal PCC-SDG. The invention provides a new fault diagnosis method, perfects an SDG modeling method, improves the efficiency of detecting multivariable states by workers, avoids the influence of time lag and other non-information synchronization factors, more effectively reduces the false alarm rate, accurately identifies the type of fault and greatly reduces the occurrence of production safety accidents.

Description

Chemical process fault diagnosis method based on multi-layer optimization PCC-SDG

Technical Field

The invention belongs to the technical field of chemical process fault diagnosis, and particularly relates to a process fault diagnosis method based on multi-layer optimization PCC-SDG.

Background

With the increasing expansion of modern industrial scale and the increasing complexity of systems, the requirements of chemical process systems for identifying and detecting faults are higher and higher. At present, quantitative modeling of a Symbol Directed Graph (SDG) is difficult, the fault diagnosis effect of the whole process is not obvious based on a simple correlation rule, although simple mathematical model analysis can carry out certain diagnosis on the fault, the diagnosis rule is not deeply discussed, the locality is strong, the whole factor is deficient, the safety and the stability of the system cannot be guaranteed, and the application is difficult.

The normal operation of the chemical system process follows the first law, so that the process variables present complex correlation characteristics, the schema abstract characteristics of the SDG on the process and equipment can represent the complex causal influence relationship among the whole process variables, and the SDG theory has been successfully applied to the industrial process, the method is mature, but the research on the SDG quantitative modeling method is less. The correlation coefficient fault diagnosis is a new fault detection method in the chemical field, fault identification and detection are carried out by combining Pearson Correlation Coefficient (PCC) with a Sign Directed Graph (SDG), the internal influence process of the fault in a complex system can be excavated to the maximum extent, and the fault can be diagnosed more comprehensively to ensure the safety of the system.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides a qualitative (SDG) and quantitative (PCC) fault diagnosis method based on a full flow, which can replace the traditional fault diagnosis method and has better fault diagnosis effect.

In order to achieve the purpose, the invention provides the following technical scheme:

the chemical process fault diagnosis method based on the multi-layer optimization PCC-SDG comprises the steps of performing correlation analysis on PCC modeling, performing variable optimization on the SDG and the like, and comprises the following specific steps:

(1) selecting variables for a specific chemical engineering process by using a Symbol Directed Graph (SDG) to establish an SDG initial network;

(2) extracting real-time data segments of variables, constructing a data vector set for correlation analysis, formulating a correlation coefficient acceptance standard, and determining an initial threshold;

(3) selecting an initial characteristic variable by using a Pearson correlation coefficient for correlation analysis, and establishing a variable correlation coefficient group for weight analysis;

(4) determining a final optimization variable according to the weight analysis and the process analysis of the correlation coefficient groups, and selecting a correlation coefficient set with a larger weight from the multilayer correlation coefficient groups to construct a PCC-SDG optimization graph;

(5) faults are detected according to the aggregate weight factor Qrule and the relative degree of difference between states is used to identify which type of fault.

(6) When the parameter Q is larger than 1, namely the system has a fault, determining the fault type by using the relevant difference degree to search a fault source in one step, and if not, in a normal state, extracting data again to detect.

The method can completely identify the running state of the process system, search the fault source and solve the fault, and is a new idea for solving the system fault in the chemical process.

The invention has the following beneficial effects:

(1) the invention provides a chemical fault multilayer optimization method based on qualitative SDG and quantitative PCC, which can accurately judge the fault state and improve the SDG modeling method, and can provide a new thought in the field of chemical process fault diagnosis, deeply dig process flow information, improve the working efficiency of operators, and prevent the problems of unplanned shutdown and the like of the production process caused by fault hidden trouble.

(2) The method is determined by the variable detection value in a certain time period, can have a certain filtering effect on the abnormal measurement value of random interference, and can ensure the normal and effective operation of a chemical process system so as to reduce the economic loss caused by non-safety factors as much as possible.

Description of the drawings

FIG. 1 is a schematic flow diagram of a multi-layer optimized PCC-SDG process;

figure 2SDG initial network schematic;

FIG. 3 is a schematic diagram of an optimal PCC-SDG network;

FIG. 4 is a graph comparing the normal state and fault state aggregate weight coefficients Q;

TABLE 1TE Process 22 consecutive measured variables (M)_st)；

TABLE 2 optimization variables M_ovContinuously measuring a set of variable related coefficients;

table 3Q parameter for fault 1;

table 4Q parameter for fault 8;

table 5Q parameter for fault 13;

table 6Q parameter for fault 16.

Detailed Description

In order to make the technical solutions of the present invention better understood, the following description is provided clearly and completely, and other similar embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present application based on the embodiments in the present application.

The first embodiment is as follows:

as shown in fig. 1, a fault diagnosis method for a chemical process of a multi-layer optimization PCC-SDG is described in a flow diagram form, including deep mining of process information by the SDG, quantitative optimization fault diagnosis of PCC, and the like, and specifically includes the following steps:

(1) analyzing the TE (Tennessee Eastman) process, selecting 22 continuous measurement variables, and establishing a Symbol Directed Graph (SDG) initial network;

(2) extracting real-time data segments of TE process variables to construct a data vector set, analyzing correlation coefficients, establishing a correlation coefficient acceptance standard, and determining an initial threshold;

(4) determining a final optimization variable through weight analysis and process analysis of the TE process correlation coefficient group, and selecting a correlation coefficient set with a larger weight from the multilayer correlation coefficient groups to construct a PCC-SDG optimization graph;

(5) and detecting faults according to an aggregation weight coefficient Q rule established by a process correlation rule, and identifying the type of the faults by using the correlation difference between the states.

The variables in the steps (1) and (2) are selected, and data acquisition and the like are as follows:

the TE process includes 12 operating variables, 41 measured variables, 20 preset faults, and 6 control modes. Taking 22 continuous measurement variables from the measurement variables as an example, dividing the TE process into three parts for analysis: the device comprises a reactor, a condenser, a gas-liquid separator, a circulating compressor and a desorption tower. Table 1 includes information of 22 continuous measurement variables, and normal values thereof are values under basic conditions. Each training data includes 480 sets of sample data, and test data 960 sets of sample data. Fig. 2 is a Symbol Directed Graph (SDG) initial network.

The weight analysis result of the variable correlation coefficient group obtained in the step (3) is shown in table 2;

the PCC-SDG optimization graph constructed in the step (4) is shown in FIG. 3;

the fault detection information required in the steps (5) and (6) is shown in tables 3-6, and the detection result is shown in fig. 4;

the analysis of the results in steps (5) and (6) is specifically as follows:

as shown in tables 3-6, the normal state and four fault state optimal variables M_fnThe Q parameter has obvious difference, the normal state and the fault state are effectively separated from each other by comparing the Q value in the figure 4, and the fault state is above a fault tolerance numerical value line, and the correlation difference degree of the state is further determined, for example, as can be seen from

faults

1 and 8, the correlation difference degrees of the normal state and the

faults

1 and 8 are respectively 6 and 7, and the correlation difference degrees of the faults 1 and the faults 8 are 4, so that the types of the faults can be accurately distinguished, and the fault source can be confirmed. The results show that the optimum variable M formed by the PCC-SDG method_fnThe normal state and the fault state can be accurately distinguished, and the effect is obvious.

Tables 1-6 are as follows:

TABLE 1TE Process continuous measurement variables

TABLE 2 optimization variables M_ovContinuously measuring a set of variable related coefficients

TABLE 3Q parameter for Fault 1

TABLE 4Q parameter for Fault 8

TABLE 5Q parameter for Fault 13

TABLE 6Q parameter for Fault 16

And at this point, the multi-layer optimization PCC-SDG fault diagnosis method is successfully finished.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims

1. A chemical process fault diagnosis method based on multi-layer optimization PCC-SDG is characterized by comprising the following steps:

(1) analyzing the TE process, selecting 22 continuous measurement variables, and establishing a symbol directed graph initial network;

(5) detecting faults according to an aggregation weight coefficient Q rule established by a correlation rule of the process, and identifying the type of the faults by using the correlation difference between the states;

(6) when the parameter Q is larger than 1, namely the system has a fault, determining the fault type by using the relevant difference degree to search a fault source in one step, and if not, determining the fault type to be in a normal state and extracting data again to detect;

variables in the steps (1) and (2) are selected, and data acquisition is specifically as follows:

in step (1), the TE process is divided into three parts for analysis: the method comprises the following steps of (1) a reactor, a condenser, a gas-liquid separator, a circulating compressor and a desorption tower, wherein the table 1 comprises information of 22 continuously measured variables, and normal values of the continuously measured variables are numerical values under basic working conditions;

the fault detection information required in steps (5) (6) is shown in tables 3-6;

tables 1-6 are as follows:

TABLE 1TE Process continuous measurement variables

TABLE 2 set of correlation coefficients for continuous measurement variables of the optimized variables Mov

TABLE 3Q parameter for Fault 1

TABLE 4Q parameter for Fault 8

TABLE 5Q parameter for Fault 13

TABLE 6Q parameter for Fault 16

。