WO2020076184A1

WO2020076184A1 - Method for assessing the time remaining until the parameters of a technical system exceed the thresholds for warning and fault signalling

Info

Publication number: WO2020076184A1
Application number: PCT/RU2019/000275
Authority: WO
Inventors: Михаил Валерьевич ЛИФШИЦ
Original assignee: Акционерное общество "РОТЕК" (АО "РОТЕК")
Priority date: 2018-10-08
Filing date: 2019-04-19
Publication date: 2020-04-16
Also published as: RU2018135274A3; RU2018135274A

Abstract

The invention relates to the field of engineering, and more particularly to a method for assessing the time remaining until the parameters of a technical system exceed the thresholds for warning and fault signalling. The present invention can be used in the creation, use, control and monitoring of a variety of different systems, including complex technical systems used in power generation, mechanical engineering, public utilities and other fields. The present invention addresses the problem of creating a method for assessing the time remaining until the parameters of a technical system exceed the thresholds for warning and fault signalling, which would make it possible to determine using manual and automated regression analysis methods the time remaining until individual parameters or groups of parameters exceed the thresholds for warning and fault signalling, and also to determine which parameters have minimal times remaining until they exceed threshold level. By comparison with methods known to the applicants, the claimed method provides maximum universality and flexibility and makes it possible to achieve better results. In particular, it makes it possible to objectively and automatically predict when individual parameters or groups of parameters will exceed the thresholds for warning and fault signalling, and also to determine which parameters have minimal times remaining until they exceed threshold level.

Description

METHOD FOR EVALUATING THE TIME FOR EXIT OF TECHNICAL SYSTEM PARAMETERS OUTSIDE WARNING AND EMERGENCY

ALARMS

DESCRIPTION

The invention relates to the field of technology, and more specifically to a method for evaluating the timing of the output of parameters of a technical system beyond the warning and alarm.

The present invention can find application in the creation, operation, management and monitoring of systems for various purposes, including complex technical systems used in energy, engineering, utilities and other industries.

The basis of the present invention is the creation of such a method for estimating the timing of the output of the parameters of the technical system beyond the warning and alarm, which would allow you to manually determine the methods of regression analysis and automatically exit individual parameters or their groups beyond the warning and alarm, as well as determine those parameters , the timing of which values are beyond these limits is minimal.

Closest to this invention is a method for evaluating the information content and priority parameters of the technical condition of a computer network (RF patent N ° 2439705), which can be taken as a prototype.

The technical solution proposed in the prototype relates to the field of computer technology, namely to the field of monitoring the technical condition (TS) of computer network elements, their monitoring and inspection, and can be used to adjust and reduce the number of controlled characteristics of the computer network TS.

The objective of the prototype is to assess the information content and priority of the monitored parameters of the CS CS, reduce the totality of the controlled characteristics by eliminating the least important of them, which will improve the efficiency of diagnostics of the CS CS, reduce the cost of acquiring diagnostic equipment, and carry out a diagnostic process that will increase the availability and operability of the enterprise CS . IN as the parameters of the CS CS, the number of subscribers in the network, throughput, subscriber intensity, average customer service time, failure rate, network data loading, conflict loading, etc. can be considered.

The technical result of the invention is to increase the efficiency of diagnostics of the TS of a computer network by reasonably reducing the number of monitored parameters of the TS CS during its continuous operation and monitoring the state of the CS in generalized control points that provide the user with maximum information about the diagnosed object, which is achieved by calculating additional probabilities of the output of the CS CS parameters beyond limits of acceptable values and probabilities of network failures. In addition, as a result of the transition to control of a smaller number of parameters of the CS CS, the time and financial costs of diagnosing and further monitoring the technical condition of the CS are reduced.

The technical result is achieved by the fact that in the method of evaluating the information content and priority of the parameters of the TS CS, consisting of sending a signal to the analyzer measuring the characteristics of the technical state of the computer network, storing values in the data storage unit, recording the measured values in the data processing unit, in which average values are sequentially calculated of each parameter for all controlled systems (similar to the purpose and operating conditions of the sections of the computer network), calculate the total average value parameters by the number of measurements taken, the elements of the covariance matrix are calculated and the correlation matrix is built on their basis, an additional parameter is calculated - the probability of the network not meeting the tasks by the end of the test (measurement) time, the average values of this parameter and the correlation coefficients are calculated by them, add the last row and the last a column of the obtained correlation matrix of probabilities as the last row and a column to the correlation matrix of parameters of the technical state of the computer network Form a transformation describing the internal connection parameters considered controlled network (or portion thereof), the product of which provides the corresponding columns of the resulting final element of the correlation matrix, the coefficients of each calculated priority indicator by solving the system of equations obtained on the basis of the elements of the constructed transformation, evaluate the information content of the initial set of characteristics using the Shannon information measure function, discard the parameters with the lowest priority coefficient, determine the information content of the reduced set of parameters, calculate the information loss of the system of the remaining parameters of the technical state compared to the original system of indicators, make the final selection of the most important pa ametrov technical condition in accordance with the user-specified maximum level of information loss. Calculation of additionally evaluated parameters is carried out according to the probability of going beyond the permissible values for each parameter and the probability of the network not meeting the tasks by the end of the test time. The calculation of the additional parameter is carried out in the data processing unit after calculating the correlation matrix of the controlled parameters of the technical state of the computer network, calculate the probability of exceeding the acceptable values for each parameter of the technical state of the computer network and the probability of the network failing to complete tasks by the end of the test time according to the number of network failures, calculation an additional parameter is carried out based on the average probability of network failures for all measurements for each characteristic the technical condition of the computer network and the average probability of network failures for all characteristics for each measurement.

However, the prototype considered has the following disadvantages:

- is not universal for various types of technical systems;

- does not allow to control the exit of individual parameters or their groups beyond the warning and alarm, as well as to determine those parameters, the timing of which values are beyond these limits is minimal.

The objectives of the invention are solved and the disadvantages of the prototype are eliminated in the method of estimating the timing of the output of the parameters of a technical system equipped with a warning or alarm, beyond the warning and alarm, comprising the following stages: 1) accumulate data on the functioning of the technical system, characterizing the indicators of the technological parameters of its functioning and transmit them to the expert;

2) the expert receives data on the functioning of the technical system and forms a selection of performance indicators based on the data received;

3) the expert, on the basis of a selection of performance indicators, determines by methods of regression analysis based on the parameters of the warning and alarm systems manually and automatically the exit of individual parameters or their groups beyond the limits of warning and alarm systems;

4) the expert, on the basis of a sample of performance indicators, determines by methods of regression analysis the parameters, the timing of which values go beyond the warning and alarm limits is minimal;

5) the results obtained by the expert at stages 3 and / or 4 are stored and used as signaling and / or control actions, allowing to achieve safe and optimal functioning of the technical system.

Due to the implementation of the claimed method, the following technical results are achieved:

- it is universal for various types of technical systems;

- allows you to manually and / or automatically control the exit of individual parameters or their groups beyond the warning and alarm, as well as determine those parameters, the timing of which values are beyond these limits is minimal.

The present invention will be disclosed in the following description of a system for estimating a time frame for a technical system to go beyond a warning and an alarm.

Preliminarily accumulate data on the functioning of the technical system, characterizing the indicators of the technological parameters of its functioning and transmit them to the expert; Suppose that n measurements are taken in the system. In the i-th test, the regressor takes the value Ci, and the test result (response) is a random variable 7,.

Test conditions do not change, so the test results are independent. The collection of pairs (X, ·,)) / = 1, ..., n is called a sample. If the tests are carried out and the results Y, =, are fixed, then the resulting set (x,,)) = 1, ..., n will be the implementation of the sample. Thus, a sample is a collection of independent random variables, and implementation is a collection of observations, that is, a collection of numbers.

The expert receives data on the functioning of the technical system and generates a selection of performance indicators based on the data received;

Thus, the initial information for the regression analysis is a selection of performance indicators, while there is no information on the form of the dependence of T on x. The selection of this dependence is the task of the expert. On how well the selection is made, the quality of the analysis results is largely determined.

Most often they use linear dependencies. There are several reasons for this. The class of linear functions leads to the most simple and deep analytical results. In the case of a normal distribution, the regression function is linear. Therefore, its use is justified when the distribution of quantities is close to the normal law, that is, in a wide range of practical situations. However, this fact (the distribution of the measured quantities of the technical system of quantities is close to the normal law) always in various cases of technical systems needs to be checked by an expert.

Moreover, the linear dependence can be considered in a certain interval as a good nonlinear approximation. The linear model for one regressor has the form

T = h + e, h = b o + bi x

Such a record assumes that the dependent variable h is measured with an error. U is the measured value, h is the exact but unknown value, e is error. The value of h linearly depends on the regressor x, b о, bi are the unknown coefficients of the linear function. Providing the quantities U, x, e with the subscript ί, we obtain n relations

which are conveniently written using vector-matrix notation.

Moreover, the vector b will have the size (i + l) x l, and the matrix X will be i x (s +1)

It should be noted that the model should be linear only in parameters b. The linearity across the regressors ci, ..., x _s , generally speaking, is not required.

Concerning errors e, the following assumptions are made.

Assumption 1. No bias

Me = 0 (i = 1, ..., ή), or else

MU = Tr.

Assumption 2. The variance is constant ϋe, = s ² (i = 1, ..., «).

Assumption 3. Independence. From independence, in particular, cov (

In order to obtain more general results, an assumption is also introduced on the distribution of errors according to the normal law [1].

Assumption 4. e, · e N (O, s ² ). If the covariance matrix of the random vector e is denoted by Όe, then the relations are presented in the form

Me = 0, G) e = s ² I,

where I is the identity matrix.

In the method under consideration, the parameters bo, bi are deterministic but unknown quantities. The expert should evaluate these values from the sample. The estimates will be sampling functions, i.e., random variables.

In mathematical statistics, certain requirements are introduced for estimates (points). For the estimates to be “good”, they must have the properties of consistency, non-bias, and a minimum of variance. All these properties are very natural. Consistency means that the more measurements in a sample, the better the score. More strictly: as n— »oo, the estimate tends to the true value of the parameter.

The requirement of non-bias is that the mathematical expectation of the estimate be equal to the estimated parameter. If the assessment is biased, it is corrected, corrected, achieving bias.

The minimum variance — the scatter around the true value of the parameter being evaluated — is also an obvious requirement. The smaller the spread, the smaller, generally speaking, the error in the estimation of the available sample.

The criterion for evaluating the unknown parameters bo, bi is the minimum of the sum of the squared errors min.

It is advisable to use this criterion or, in other words, the least squares method has been discussed previously. Note that under the assumption that the errors are normal, the criterion is a consequence of the maximum likelihood method, the most justified method for the point estimate of unknown parameters. Estimates of the parameters b make it possible to estimate the direct regression h = bo + bic.

The least squares method allows one to obtain estimates for other linear models using calculations similar to those performed above. Suppose, for example, we want to use a one-parameter model

Yi = b Xi + e _/ .

From the condition of the minimum of the sum of squares, we obtain the estimate of the parameter b

Thus, the expert, on the basis of a selection of performance indicators, determines the calculations described above using regression analysis based on the warning and alarm parameters manually and the individual parameters or their groups go beyond the warning and alarm limits automatically;

Further, the expert, on the basis of a sample of performance indicators, determines, by methods of regression analysis, parameters whose timing for the values to go beyond the warning and alarm limits is minimal;

The results obtained by the expert at the previous stages are saved and used as signaling and / or control actions, allowing to achieve safe and optimal functioning of the technical system.

Compared with the methods known to the authors, the inventive method has the maximum versatility and flexibility and allows you to achieve better results, in particular, objectively and automatically predict the exit of individual parameters or their groups beyond the warning and alarm, as well as to determine those parameters, the timing of which values are beyond these limits are minimal.

Literature

1. M.G. Sukharev Forecasting Methods - Series Applied Mathematics in Engineering M: 2009

Claims

CLAIM

A method for evaluating the timing of the output of parameters of a technical system equipped with a warning or alarm, beyond the warning and alarm, comprising the following stages:

1) accumulate data on the functioning of the technical system, characterizing the indicators of the technological parameters of its functioning and transmit them to the expert;

2) the expert receives data on the functioning of the technical system and generates a selection of performance indicators based on the received data;