CN104503441A - Process fault monitoring method based on improved dynamic visible graph - Google Patents

Abstract

The invention provides a process fault monitoring method based on an improved dynamic visible graph. With the help of a complicated network theory, the invention provides an improved dynamic visible graph algorithm, on the basis of the algorithm, time serial data is mapped into a complicated network structure, the time serial data with different variables can be distinguished and identified through network characteristics, in addition, whether production data generates a fault or not is judged. The method has the advantages that the fault false alarm rate and the missing report rate be reduced, in addition, the fault occurrence can be earlier monitored, and the real-time monitoring of the complicated industrial process can be more favorably realized.

Description

A Process Fault Monitoring Method Based on Improved Dynamic Visible Graph

technical field

The invention relates to the field of fault identification, in particular to a process fault monitoring method based on an improved dynamic visibility graph (MDVG).

Background technique

At present, the rapid development of information technology provides great convenience for the acquisition and processing of massive data. Complex network, as an emerging research field, has attracted extensive attention from many disciplines. In recent years, a large number of achievements have been made in extracting relevant information from data generated by complex systems in various fields, and using network knowledge to describe and analyze the attribute status of the system. Among them, mapping time-series data to complex networks and applying rich and advanced complex network analysis methods to analyze complex time-series data is particularly eye-catching.

Different from the relatively traditional distance-based and correlation-based network construction methods, Lacasa et al. creatively proposed a visibility algorithm, namely the natural visibility graph (NVG). Each data in the time series data is corresponding to a node of the complex network. For each node, if other nodes meet its visibility conditions, the two nodes are connected to form a visible graph. The constructed network inherits the structural attributes of the original sequence, such as periodic, random, and fractal sequences can be transformed into regular, random, and scale-free networks, respectively.

Subsequently, Fioriti et al. proposed a horizontal visibility graph (HVG) algorithm to calculate the maximum eigenvalue of the HVG correlation adjacency matrix obtained from multiple time series to distinguish sequence chaos and randomness. HVG and NVG are only slightly different in the determination of visible conditions. HVG is a subgraph of NVG, which converts a set of time series data into a unique network structure. The use of NVG and HVG algorithms can describe and explore time series of complex structures related to various phenomena, such as traffic fluctuations, stock indexes, heartbeat dynamics, random and chaotic sequences, and so on.

On the basis of the NVG algorithm, Bezsudnov et al. proposed a dynamical visibility graph (DVG) algorithm. By introducing the "viewpoint" parameter, changing the visible condition, and then affecting the change of the network structure, a time series is transformed into a set of networks, each DVG is a subgraph of NVG. At the same time, through the three network characteristics of node relative average degree, relative average connection length and number of disconnected groups, a new dynamic dimension is provided to distinguish, identify and describe various time series in detail

On the other hand, process monitoring methods can be mainly divided into three categories: model-based methods, knowledge-based methods and data-based methods. Compared with the former two more traditional methods, data-based process monitoring methods are widely used because they do not require process models, especially in complex industrial processes or systems where models and expert knowledge are difficult to establish and obtain in practice. Such as chemical process. Due to the widespread use of distributed control systems in modern industrial processes, large amounts of data can be recorded and collected. Process data usually has characteristics such as high dimensionality, nonlinearity, time-varying, multimodality, and autocorrelation, which are difficult to be fully processed by existing data-based process monitoring methods.

Therefore, it is necessary to develop a new data-based process detection method to solve the above-mentioned defects of the prior art.

Contents of the invention

In order to solve the above problems, the present invention proposes a process fault monitoring method based on Modified Dynamic Visible Graph (MDVG).

The invention provides a process fault monitoring method based on an improved dynamic visible graph. With the help of complex network theory, the present invention proposes an improved dynamic visible graph algorithm. On the basis of this algorithm, time series data is mapped to a complex network structure , distinguish and identify the time series data of different variables through network characteristics, and judge whether there is a failure in the production data. This method can reduce the false positive rate and false negative rate of the failure, and can detect the occurrence of the failure earlier, which is more conducive to complex Real-time monitoring of industrial processes.

The present invention claims a process fault monitoring method based on an improved dynamic visible graph, and the method includes the following steps:

S101, determine the monitoring variable, and after normalizing the historical data of each variable according to a certain moving pane length, use the MDVG algorithm to map into a complex network;

The process of mapping to a complex network using the MDVG algorithm is as follows:

Consider any two data (t _i , x _i ) and (t _k , x _k ) in a set of time series data, i<k, for all data (t _j , x _j ) between them, i<j< k, if the visible condition is satisfied

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\frac{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ik</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arctan</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Then, consider these two data-mapped nodes visible and connected in the network,

Calculated by formulas (1) and (3), determine the unique network structure corresponding to the time series data under the view angle α and the time interval constant h.

S102. Calculate three characteristic parameters K(α), Λ(α), and Q(α) of the mapped network, and determine monitoring indicators and corresponding thresholds.

The three important characteristic parameters that characterize MDVG(α) are as follows:

(1) Node relative average degree K(α):

$\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}<span\; class="notranslate">\; /</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}}_{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

in, and are the average degrees of network nodes under the perspectives α and π, respectively.

(2) Relative average connection length Λ(α):

$\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; l</span>}}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}<span\; class="notranslate">\; /</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; l</span>}}}_{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

in, and are the average connection lengths of the network under views α and π, respectively.

(3) The number of non-connected groups Q(α): If a group of nodes arranged in sequence, there is at least one pair of connection relationship between each node, and each node has no connection relationship with any node in the non-group , then this set of nodes constitutes a disconnected clique. The number of disconnected cliques in each network is the number of disconnected cliques.

S103, implementing online process monitoring, using the same moving pane length as the historical data, monitoring the current data of each variable, and calculating the monitoring indicators of the current data;

S104, judging whether the monitoring index of the current data exceeds the threshold, if so, the system

Send an alarm so that the operator can find and determine the cause of the failure in time.

Further, the time interval constant h can be determined by the particle swarm optimization algorithm, so that the mean value of the mode occurrence times of K(α), Λ(α), and Q(α) is the smallest, and the mathematical model of the problem is expressed as:

min J(h)＝(M _K(α(h)) +M _Λ(α(h)) +M _Q(α(h)) )/3 (6)

Among them, M _K(α) , M _Λ(α) , and M _Q(α) are the occurrence times of the modes of K(α), Λ(α), and Q(α) respectively.

Further, h=0.15 is a better universal value that can obtain high-discrimination MDVG characteristics, and can be used as a reference value for calculation.

Description of drawings

Figure 1 is a legend of the visible graph algorithm;

Fig. 2 is the process fault monitoring method based on MDVG according to the embodiment of the present invention;

Fig. 3 is the TE process in the embodiment;

Figure 4 shows the relative average degree of nodes in the TE process;

Figure 5 is the relative average connection length of the TE process;

Figure 6 shows the number of disconnected groups in the TE process;

Figure 7 is a partial enlarged view of the number of disconnected cliques in the TE process;

Fig. 8 is the process monitoring of fault 3 by DVG-Q (85o);

Fig. 9 is the process monitoring of fault 3 by MDVG-Q (65o);

Figure 10 is the process monitoring of LKPCA-T2 to fault 3;

Figure 11 is the process monitoring of LKPCA-Q for fault 3.

Detailed ways

The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

The process failure monitoring method based on MDVG of the present invention, comprises the steps:

S101. Determine the monitoring variables, normalize the historical data of each variable according to a certain moving pane length, and then use the MDVG algorithm to map into a complex network.

The process of traditional DVG mapping time series data to complex networks is generally as follows:

Consider any two data (t _i , x _i ) and (t _k , x _k ) in a set of time series data, i<k, for all data (t _j , x _j ) between them, i<j< k, if the NVG visibility condition is met

${\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>\frac{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

and the limitation of viewing angle α

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ik</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arctan</span>}\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Then, consider the nodes of these two data maps to be visible and connected in the network. For example, as shown in Figure 1, line 1 belongs to all graphs—NVG, DVG (α=90°), HVG, line 2 belongs to NVG and HVG but not DVG (α=90°), and line 3 belongs to NVG and DVG (α =90°) but not HVG.

In the process of traditional DVG mapping time series data to complex networks, the influence of the relationship between time interval and numerical value on the construction of the network is not considered, which may lead to the network structure of the same group of data under different perspectives and different data under the same perspective. The difference is not large, which hinders further analysis.

For this reason, the improved dynamic visibility graph (modified dynamical visibility graph, MDVG) algorithm that the present invention proposes, under the premise of satisfying formula (1), introduces the time interval constant h to further limit the viewing angle, optimizes the formula (2) in the DVG algorithm:

$\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ik</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arctan</span>}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Calculated by formulas (1) and (3), determine the unique network structure corresponding to the time series data under the view angle α and the time interval constant h.

The time interval constant h can be obtained in two ways:

1. The time interval constant h is determined by the particle swarm optimization algorithm, so that the mean value of the mode occurrence times of K(α), Λ(α), and Q(α) is the smallest. The mathematical model of the problem is expressed as:

min J(h)＝(M _K(α(h)) +M _Λ(α(h)) +M _Q(α(h)) )/3 (6)

Among them, M _K(α) , M _Λ(α) , and M _Q(α) are the occurrence times of the modes of K(α), Λ(α), and Q(α) respectively.

The mode is the value with the most occurrences in a set of data. The less the mode appears, the less the MDVG characteristics will be concentrated on a specific value as the perspective changes, and more abundant data can be obtained from a specific time series. It is easier to distinguish from other time series by processing and analyzing information.

2. By taking the average value of multiple calculations of different types and lengths of data, it is found that h=0.15 is a better universal value that can obtain high-discrimination MDVG characteristics, and can be used as a reference value for calculations.

S102. Calculate three characteristic parameters K(α), Λ(α), and Q(α) of the mapped network, and determine monitoring indicators and corresponding thresholds.

The three important characteristic parameters that characterize MDVG(α) are as follows:

(1) Node relative average degree K(α):

$\mathrm{<span\; class="notranslate">\; K</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}<span\; class="notranslate">\; /</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; k</span>}}}_{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

in, and are the average degrees of network nodes under the perspectives α and π, respectively.

(2) Relative average connection length Λ(α):

$\mathrm{<span\; class="notranslate">\; \&Lambda;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; l</span>}}}_{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}<span\; class="notranslate">\; /</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; l</span>}}}_{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

in, and are the average connection lengths of the network under views α and π, respectively.

(3) The number of non-connected groups Q(α): If a group of nodes arranged in sequence, there is at least one pair of connection relationship between each node, and each node has no connection relationship with any node in the non-group , then this set of nodes constitutes a disconnected clique. The number of disconnected cliques in each network is the number of disconnected cliques.

The monitoring index can include the maximum value, the average value, or the value under a specific perspective of the above three characteristic parameters. In this step, the threshold of the monitoring index can be determined. Within the threshold range, the monitoring index is considered normal.

S103, implementing online process monitoring, using the same moving pane length as the historical data, monitoring the current data of each variable, and calculating the monitoring index of the current data.

The monitoring index is consistent with that of step S102, and may include the maximum value, the average value, or the value under a specific perspective of the above three characteristics, and the calculation method of the three characteristics is as shown above.

S104, judging whether the monitoring index of the current data exceeds the threshold, and if so, the system sends out an alarm, so that the operator can find and determine the cause of the failure in time.

The method of the present application will be further described below through an embodiment.

The TE (Tennessee Eastman) process model is a real simulation program for a chemical plant, which is widely used in standard tests in control and monitoring research, and its process is shown in Figure 6. The TE process consists of 5 main units, namely reactor, condenser, compressor, separation tower and stripper, through 4 reactions to produce 2 products, and at the same time generate a total of 8 components of inert gas and by-products, which are denoted as A , B, C, D, E, F, G, and H. The whole system includes 12 operating variables and 41 process variables (including 22 direct measurement variables and 19 analysis variables), and 20 process faults are preset (among them, the type of fault 1-8 is a step change, fault 9-12 Random change, fault 13 is slow drift, fault 14, 15 is valve sticking, fault 16-20 is unknown). Since the MDVG process monitoring method is only based on process data, it is not necessary to know the mathematical model of the process in advance.

Select 22 directly measured variables as monitoring variables, the length of the moving pane is 50, and the sampling interval is 0.03h. The 1100 sampled data are normalized according to the variables, and mapped into a complex network with 1100 nodes using the MDVG algorithm. The relevant characteristic values of the network are then calculated. Four different types of fault data and normal data were selected for analysis and comparison, and five sets of data from different panes were selected for each type to reduce randomness. The calculation results are shown in Figure 3, Figure 4, and Figure 5.

It can be seen from Figure 3, Figure 4, and Figure 5 that the lines of the same color are gathered together, and the lines of different colors have a certain distinction, which shows that the MDVG algorithm can better distinguish different types of data, and the characteristic difference between the same type of data Basically not big.

By comparing the correlation characteristics between normal data and various types of fault data, it can be found that the distinction effect of the number of disconnected groups is relatively optimal, and it is easy to quantify. Q(α=65°) is used as a monitoring index to monitor the TE process online, and the upper and lower limits of the monitoring threshold are 540 and 520 respectively, that is, when the Q(α=65°) value of the data is between 520 and 540 , we consider the process to be in a normal state, otherwise the process malfunctions.

Fault 3 is a random change caused by D temperature in flow 2. It is a difficult fault to detect and has a certain representativeness. It will be analyzed in detail below. Collect 300 sets of data, introduce fault 3 from the 101st set, and compare the online monitoring results of DVG, MDVG methods and the current relatively advanced local kernel principal components analysis (LKPCA) method. The running results of each monitoring method are shown in Table 1.

Table 1 Operation results of various monitoring methods

Through the simulation results of the TE fault process, it can be found that under the relatively optimal monitoring indicators selected respectively, the MDVG method improves the monitoring effect compared with the DVG method. At the same time, compared with the T ² and Q statistics of LKPCA, the MDVG process monitoring method proposed by the present invention has a higher accuracy rate, and can detect faults earlier.

The description of the present invention has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and changes will be apparent to those of ordinary skill in the art. The embodiment was chosen and described in order to better explain the principles of the invention and the practical application, and to enable others of ordinary skill in the art to understand the invention and design various embodiments with various modifications as are suited to the particular use.

Claims (4)

1. A method for monitoring process faults based on an improved dynamic visibility diagram, the method comprising the steps of:

s101, determining monitoring variables, respectively normalizing historical data of the variables according to a certain moving pane length, and mapping the variables to a complex network by using an MDVG algorithm;

the process of mapping to a complex network using the MDVG algorithm is as follows:

consider any two data (t) in a set of time series data_i,x_i) And (t)_k,x_k)，i<k，For all data (t) between them_j,x_j)，i<j<k, if the visible condition is satisfied

$x_j<x_i+(x_k-x_i)\frac{t_j-t_i}{t_k-t_i}---\left(1\right)$

<math> <mrow> <mi>&alpha;</mi> <mrow> <mo>(</mo> <mi>h</mi> <mo>)</mo> </mrow> <mo>&lt;</mo> <msub> <mi>&alpha;</mi> <mi>ik</mi> </msub> <mo>=</mo> <mi>arctan</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <msub> <mi>x</mi> <mi>k</mi> </msub> <mo>-</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> </mrow> <mrow> <msub> <mi>t</mi> <mi>k</mi> </msub> <mo>-</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> </mrow> </mfrac> <mo>*</mo> <mi>h</mi> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow> </math>

Then, the nodes to which the two data maps are considered to be visible and connected in the network,

and (3) determining the corresponding unique network structure of the time series data under the viewing angle alpha and the time interval constant h through the calculation of the formulas (1) and (3).

S102, calculating three characteristic parameters K (alpha), Lambda (alpha) and Q (alpha) of the mapped network, and determining a monitoring index and a corresponding threshold value.

Three important characteristic parameters characterizing MDVG (α) are as follows:

(1) node relative average degree K (α):

<math> <mrow> <mi>K</mi> <mrow> <mo>(</mo> <mi>&alpha;</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mover> <mi>k</mi> <mo>&OverBar;</mo> </mover> <mi>&alpha;</mi> </msub> <mo>/</mo> <msub> <mover> <mi>k</mi> <mo>&OverBar;</mo> </mover> <mi>&pi;</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow> </math>

wherein,andthe average of the network nodes at the viewing angles alpha and pi, respectively.

(2) Relative average connection length Λ (α):

<math> <mrow> <mi>&Lambda;</mi> <mrow> <mo>(</mo> <mi>&alpha;</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mover> <mi>l</mi> <mo>&OverBar;</mo> </mover> <mi>&alpha;</mi> </msub> <mo>/</mo> <msub> <mover> <mi>l</mi> <mo>&OverBar;</mo> </mover> <mi>&pi;</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow> </math>

wherein,andthe average link lengths of the network at viewing angles alpha and pi, respectively.

(3) Number of non-connected groups Q (α): if a group of nodes arranged in sequence has at least one pair of connection relations between each node and any node in the non-group has no connection relation, the group of nodes form a non-connected group, and the number of the non-connected groups in each network is the number of the non-connected groups.

S103, carrying out online process monitoring, monitoring the current data of each variable by adopting the length of a movable pane which is the same as the historical data, and calculating the monitoring index of the current data;

and S104, judging whether the monitoring index of the current data exceeds a threshold value, and if so, giving an alarm by the system so as to facilitate an operator to search and determine the fault reason in time.

2. The method of claim 1, wherein the time interval constant h can be determined by a particle swarm optimization algorithm such that the mean of the number of occurrences of the modes of K (α), Λ (α), Q (α) is minimized, and the mathematical model of the problem is represented as:

minJ(h)＝(M_K(α(h))+M_Λ(α(h))+M_Q(α(h)))/3 (6)

wherein M is_K(α)、M_Λ(α)、M_Q(α)The number of occurrences of the K (α), Λ (α), Q (α) modes, respectively.

3. The method of claim 1, wherein h-0.15 is a good general value for obtaining the high resolution MDVG characteristic, and can be used as a reference value for the operation.

4. The method of claim 1, wherein the monitoring index of step S103 is consistent with the monitoring index of step S102, and may include the most value, the average value or the value at a specific viewing angle of the three characteristic parameters.