CN104317778A - Massive monitoring data based substation equipment fault diagnosis method - Google Patents

Abstract

The invention discloses a massive monitoring data based substation equipment fault diagnosis method. The method includes: structuring a process memory matrix D according to related monitoring information of equipment; performing normalization processing on measuring values of measure points according to extreme values of the measuring values; structuring a fault diagnosis model, and determining whether faults occur to the substation equipment or not according to residual errors of input and output monitoring amounts. Compared with an equipment fault diagnosis system based on the neural network technique, the massive monitoring data based substation equipment fault diagnosis method has the advantages that shortcomings that the neural network is directly used for learning when more variables are inputted and long time and massive learning samples are needed resulting in the fact that learning convergence cannot be guaranteed are overcome; meanwhile, the method is a parameter-free modeling method, no functions are assumed for model, and accuracy and rapidness in implementation are achieved.

Description

Based on the converting equipment method for diagnosing faults of magnanimity Monitoring Data

Technical field

The present invention relates to electric utility, particularly a kind of converting equipment method for diagnosing faults based on magnanimity Monitoring Data.

Background technology

Current power industry, multiple infosystems such as production management system, on-line monitoring system, lighting location are established, but because system is independent separately, fails to realize multi-field, multi-disciplinary, Multi-information acquisition, be unfavorable for the various dimensions of information, multidata analysis and displaying.And current infosystem, lays particular emphasis on the alarm of a certain index exceeding standard mostly, can only reflect the partial picture of equipment, also fail to make full use of multi-source information and realize dynamic evaluation.Many to the failure analysis report of single device at present, little to the reflection of overall condition, be unfavorable for that O&M appraiser understands the degree of depth of equipment state and evaluates, the requirement of equipment lean O&M cannot be met.

NETWORK STRUCTURE PRESERVING POWER SYSTEM experienced by development from simple to complex, and the model solution method that research adopts also analytically develops into artificial intelligence.Wherein, the important theme that quantification correlationship is modeling is found.And system is more complicated, correlationship is also more difficult to identify.Past, equipment fault diagnosis and early warning can only rely on limited experiment statistics or expertise modeling owing to lacking data, caused these models applying to there is very large limitation.Nowadays, along with the development of monitoring of equipment detection technique, the multi-source heterogeneous data such as equipment and environment constantly gather accumulation.But analytic model and intelligent algorithm are limited to again complex model and solve the problem with dimension calamity, be difficult to identify key factor in mass data and make accurate device diagnosis or early warning decision.

Therefore, in grid equipment operational monitoring with under detecting the ever-increasing situation of data, research can identify the algorithm of the complicated correlationship of non-linear nonmonotonicity between variable online fast, correlationship that is implicit or the unknown is found under mechanism is familiar with incomplete situation, for fault modeling provides new thread, be the important foundation work that under raising complex environment, grid equipment runs pre-alerting ability.

Summary of the invention

For solving the deficiency that prior art exists, the invention discloses a kind of converting equipment method for diagnosing faults based on magnanimity Monitoring Data, the method integrates historical failure record, converting equipment and power network monitoring, detection data, according to comparing of current operational monitoring information and process dot-blur pattern, judge that whether current device running status is abnormal, thus grid equipment potential faults can be identified rapidly, for scheduling and fortune inspection personnel provide decision-making foundation.

To achieve these goals, the present invention adopts following technical scheme:

Based on a converting equipment method for diagnosing faults for magnanimity Monitoring Data, comprising:

(1) suppose that equipment has n the monitoring information that is mutually related, the monitoring information in a certain moment is designated as vectorial X (i)=[x ₁x ₂... x _n] ^t, within the period that this process or equipment normally work, under different operating modes run, gather k the conception of history measure, construction process dot-blur pattern D;

Wherein, x ₁x ₂... x _nrepresent n the monitoring information that is mutually related;

(2) the monitoring information value of each measuring point is normalized according to respective extreme value, makes actual measured value be mapped to [01] interval;

(3) construct breakdown judge model, described model is with observation vector X _infor input, the output monitoring variable of described model is Xout=D*W=w ₁x (1)+w ₂x (2)+...+w _kx (k); Wherein, D is process dot-blur pattern D, W is weight vector W=[w ₁w ₂... w _n] ^t;

(4) according to the residual epsilon=X of constrained input monitoring variable _out-X _in, determine whether converting equipment breaks down; Concrete determination methods is: when | ε | during≤δ, converting equipment normally runs, when | ε | > δ converting equipment is abnormal; δ is the threshold value of setting.

Described process dot-blur pattern D is specially:

$D=[X\left(1\right),X\left(2\right),...X\left(k\right)]=\left[\begin{array}{ccc}{x}_1\left(1\right)& ...& x_1\left(k\right)\\ x_2\left(1\right)& ...& x_2\left(k\right)\\ x_n\left(1\right)& ...& x_n\left(k\right)\end{array}\right]$

Wherein, k is the number in monitoring moment, and n is the number of synchronization monitoring variable.

The circular of described weight vector W is:

$W={(D^T\⊗D)}^{-1}\·(D^T\⊗X_{\mathrm{in}})$

Wherein, for nonlinear operation symbol, $D^T\⊗D=\⊗(D^T,D)=\sqrt{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{({X_i}^T-X_i)}^2},$

$D^T\⊗X_{\mathrm{in}}=\⊗(D^T,X_{\mathrm{in}})=\sqrt{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{({X_i}^T-x_i)}^2}.$

Beneficial effect of the present invention:

The inventive method, compared with the Fault Diagnosis of Mechanical Equipment based on nerual network technique, overcomes when input variable is a lot, directly learns by neural network, needs long time and huge learning sample, and can not ensure the shortcoming of Learning Convergence.The method is a kind of parameterless modeling method simultaneously, is not any function of model hypothesis, so have more the rapidity of accuracy and enforcement.

Accompanying drawing explanation

Fig. 1 is converting equipment method for diagnosing faults process flow diagram of the present invention.

Embodiment:

Below in conjunction with accompanying drawing and embodiment, the present invention is described in detail:

Based on a converting equipment method for diagnosing faults for magnanimity Monitoring Data, as shown in Figure 1, comprising:

Suppose that equipment has n the monitoring information that is mutually related, the monitoring information in a certain moment is designated as a vector, X (i)=[x ₁x ₂... x _n] ^t, within the period that this process or equipment normally work, under different operating modes run, gather k the conception of history measure, construction process dot-blur pattern:

$\begin{array}{c}D=[X\left(1\right),X\left(2\right),...X\left(k\right)]\\ =\left[\begin{array}{ccc}{x}_1\left(1\right)& ...& x_1\left(k\right)\\ x_2\left(1\right)& ...& x_2\left(k\right)\\ x_n\left(1\right)& ...& x_n\left(k\right)\end{array}\right]\end{array}$ Wherein k is monitoring moment number, and n is synchronization monitoring variable number.

Wherein, x ₁x ₂... x _nrepresent that n to be mutually related monitoring information, the measurement informations such as the electric current and voltage load in such as equipment a certain moment and equipment is the environmental information such as temperature, wind speed, humidity at that time.

Because the be correlated with dimension of monitoring variable measuring point of equipment is different, and different measuring points data absolute value differs greatly, for ensureing that use nonlinear operator correctly weighs the distance between different observation vector, need to be normalized according to respective extreme value the measured value of each measuring point, make actual measured value be mapped to [01] interval.

Each row observation vector in process dot-blur pattern represents a normal operating conditions of equipment.K history observation vector in the process dot-blur pattern of choose reasonable the subspace of opening can represent the whole dynamic process that process or equipment normally runs.Therefore, the structure essence of process dot-blur pattern is exactly the learning and memory process to process or the normal operation characteristic of equipment.

Structure breakdown judge model: to any one input observation vector X _in, determine a weight vector W=[w ₁w ₂... w _n] ^t, then monitoring variable Xout=D*W=w is exported ₁x (1)+w ₂x (2)+...+w _kx (k).

Weight vector $W={(D^T\⊗D)}^{-1}\·(D^T\⊗X_{\mathrm{in}}),$ Wherein for nonlinear operation symbol, Euclidean distance is adopted to weigh, when two vectorial same or similar time, distance is 0 or close to 0; Two vectorial differences are larger, and the result of its nonlinear operation is larger.This weight vector has reacted the similarity of each vector in Input Monitor Connector amount and process dot-blur pattern.

According to the residual epsilon=X of constrained input monitoring variable _out-X _in, determine whether converting equipment breaks down; Concrete determination methods is: when | ε | during≤δ, converting equipment normally runs, when | ε | > δ converting equipment is abnormal; δ is the threshold value of setting, and the present embodiment gets 0.07.

When equipment is working properly, new Input Monitor Connector vector is positioned at the normal working space representated by process dot-blur pattern, and survey vector distance comparatively closely with some conception of history in D matrix, its predicted value corresponding exports monitoring variable X _outthere is very high precision.When equipment working state change break down hidden danger time, due to the change of dynamic perfromance, Input Monitor Connector vector will depart from normal working space, cannot construct the accurately predicting value of its correspondence by the combination of history observation vector in D matrix, cause precision of prediction to decline, residual error increases.Therefore judge that converting equipment is normal or malfunction.

By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various amendment or distortion that creative work can make still within protection scope of the present invention.

Claims (3)

1., based on a converting equipment method for diagnosing faults for magnanimity Monitoring Data, it is characterized in that, comprising:

(1) suppose that equipment has n the monitoring information that is mutually related, the monitoring information in a certain moment is designated as vectorial X (i)=[x ₁x ₂... x _n] ^t, within the period that this process or equipment normally work, under different operating modes run, gather k the conception of history measure, construction process dot-blur pattern D;

Wherein, x ₁x ₂... x _nrepresent n the monitoring information that is mutually related;

(2) the monitoring information value of each measuring point is normalized according to respective extreme value, makes actual measured value be mapped to [01] interval;

(3) construct breakdown judge model, described model is with observation vector X _infor input, the output monitoring variable of described model is Xout=D*W=w ₁x (1)+w ₂x (2)+...+w _kx (k); Wherein, D is process dot-blur pattern D, W is weight vector W=[w ₁w ₂... w _n] ^t;

(4) according to the residual epsilon=X of constrained input monitoring variable _out-X _in, determine whether converting equipment breaks down; Concrete determination methods is: when | ε | during≤δ, converting equipment normally runs, when | ε | > δ converting equipment is abnormal; δ is the threshold value of setting.

2. a kind of converting equipment method for diagnosing faults based on magnanimity Monitoring Data as claimed in claim 1, it is characterized in that, described process dot-blur pattern D is specially:

$D=[X\left(1\right),X\left(2\right),...X\left(k\right)]=\left[\begin{array}{ccc}{x}_1\left(1\right)& ...& x_1\left(k\right)\\ x_2\left(1\right)& ...& x_2\left(k\right)\\ x_n\left(1\right)& ..& x_n\left(k\right)\end{array}\right]$

Wherein, k is the number in monitoring moment, and n is the number of synchronization monitoring variable.

3. a kind of converting equipment method for diagnosing faults based on magnanimity Monitoring Data as claimed in claim 1, it is characterized in that, the circular of described weight vector W is:

$W={(D^T\⊗D)}^{-1}\·(D^T\⊗X_{\mathrm{in}})$

Wherein, for nonlinear operation symbol, $D^T\⊗D=\⊗(D^T,D)=\sqrt{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{({X_i}^T-X_i)}^2},$ $D^T\⊗X_{\mathrm{in}}=\⊗(D^T,X_{\mathrm{in}})=\sqrt{\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{({X_i}^T-x_i)}^2}.$