CN103577700B - Boat firefighting system interlock failure prediction method - Google Patents

Abstract

The invention belongs to the field of complicated system analysis and decision and particularly relates to a boat firefighting system interlock failure prediction method based on a grey correlation cluster and a BP artificial neural network. In a boat firefighting system, each cabin is set to be a firefighting unit and n firefighting system failure factors are summed up and obtained and serve as observation data in a boat fire accident, a reference data queue and N-1 comparison data queues are set up, the firefighting unit interlock failure finite threshold value serves as the reference data queue and the comparison data queues serve as actual observation values after failure factors of all the firefighting units are quantized. According to the boat firefighting system interlock failure prediction method based on the grey correlation cluster and the BP artificial neural network, the dynamic analysis technology of failure of the boat firefighting system is provided, defects of an existing static analysis technology are overcome, coupling relations among the cabins are taken into consideration, main interlock failure cabins and failure factors of the firefighting system are determined and the safety state of the firefighting system is determined according to the actual conditions of a boat.

Description

A kind of Forecasting Methodology of ship's fire fighting system interlock failure

Technical field

The invention belongs to analysis of complex system with decision domain and in particular to a kind of artificial based on Grey Correlation Cluster and bp The Forecasting Methodology of the ship's fire fighting system interlock failure of neutral net.

Background technology

Ship fire is to threaten one of main accident of safety of ship.Fire-fighting system is then related to the control intensity of a fire and goes out Fire, is extremely important complication system on ship.Ship's space is numerous, and running environment is unique and the source of trouble is more, projects and embodies The fragility of complication system, has chain ineffectiveness.In the past in prior art literature both domestic and external, it is all from ship fire kinetics Or fire risk sets out, only fire-fighting thrashing is carried out with static analysis, and have ignored the dynamic of ship's fire fighting thrashing State property can be the research of interlock failure.In order to overcome the above-mentioned deficiencies of the prior art, the invention provides a kind of be based on Lycoperdon polymorphum Vitt The Forecasting Methodology of the ship's fire fighting system interlock failure of association cluster and bp artificial neural network, the ship for setting up China disappears Anti- system security assessment provides technological means.

Content of the invention

It is an object of the invention to proposing one kind to overcome the various chain Failure Factors randomness of ship's fire fighting system, ambiguity Interlock failure Forecasting Methodology with irrelevance.

The object of the present invention is achieved like this:

The Forecasting Methodology of ship's fire fighting system interlock failure, comprises the steps:

(1) each cabin in ship's fire fighting system is set to a fire fighting unit, draws n according to ship fire event statistics The chain Failure Factors of individual fire-fighting system as observation data, set up 1 reference data array and n-1 compare data row, by fire-fighting The chain inefficacy of unit limits threshold values as reference data array, compares the reality after the Failure Factors that data sequence is each fire fighting unit quantify Border observation, wherein x₀={x₀ ⁽⁰⁾(i), (i=1,2...n) } it is reference data array, x_k={x_k ⁽⁰⁾(i),(k=1,2...n-1;i= 1,2...n) be } to compare data row, n-1 is the quantity in cabin, the element in each sequence represent respectively n chain inefficacy because Element；

(2) set up gm (1, n-1) model:Wherein $x_m^{\left(1\right)}\left(i\right)=\underset{i=1}{\overset{k}{\mathrm{\σ}}}{x}_m^{\left(0\right)}\left(i\right)(k=\mathrm{1,2}....n),z_0^{\left(1\right)}=(z_0^{\left(1\right)}\left(2\right),z_0^{\left(1\right)}\left(3\right).....z_0^{\left(1\right)}\left(n\right))$ ForNext-door neighbour's average generation sequence,-a For system development coefficient, b_mFor drive factor, solve by method of least square and can get a and b_mEstimated value；

(3) determine comparative sequences x_kMiddle Failure Factors are to reference sequences inefficacy x₀The gray relation grades r of factor_0k(k=1, 2...n-1)；

(4) according to degree of association r_0kNumerical values recited carry out ordered series of numbers sequence, draw fire-fighting system occur chain inefficacy cabin Primary and secondary sorts, and judges which cabin associates maximum with system interlock failure, and at screening, fire-fighting system is susceptible to linksystem mistake The cabin of effect, determines the chain Failure Factors in this cabin；

(5) calculate x_mWith x_jThe grey relational grade of (m≤j, and m, j=0,1...n-1), carries out grey cluster by real-time monitored Failure Factors carry out merger, n chain Failure Factors can be gathered for f class (f≤n)；

(6) according to the chain Failure Factors in step (4), each class of the f class of step (5) find out representative inefficacy because Element, total p representative Failure Factors (n >=p >=f), set up the forecast model of bp artificial neural network, the input of neutral net Vector is p representative Failure Factors, is designated as h=(h₁,h₂,....h_p), output is y=(y₁,y₂,y₃), output vector is drawn It is divided into safety, Generally Recognized as safe, the three kinds of states that lost efficacy to use (1,0,0), (0,1,0), (0,0,1) to represent respectively.

(1) each cabin in ship's fire fighting system is set to a fire fighting unit, draws n according to ship fire event statistics The chain Failure Factors of individual fire-fighting system as observation data, set up 1 reference data array and n-1 compare data row, by fire-fighting The chain inefficacy of unit limits threshold values as reference data array, compares the reality after the Failure Factors that data sequence is each fire fighting unit quantify Border observation, wherein x₀={x₀ ⁽⁰⁾(i), (i=1,2...n) } it is reference data array, x_k={x_k ⁽⁰⁾(i),(k=1,2...n-1;i= 1,2...n) be } to compare data row, n-1 is the quantity in cabin, the element in each sequence represent respectively n chain inefficacy because Element；

(2) set up gm (1, n-1) model:Wherein $x_m^{\left(1\right)}\left(i\right)=\underset{i=1}{\overset{k}{\mathrm{\σ}}}{x}_m^{\left(0\right)}\left(i\right)(k=\mathrm{1,2}....n),z_0^{\left(1\right)}=(z_0^{\left(1\right)}\left(2\right),z_0^{\left(1\right)}\left(3\right).....z_0^{\left(1\right)}\left(n\right))$ ForNext-door neighbour's average generation sequence,- a is System development coefficient, b_mFor drive factor, solve by method of least square and can get a and b_mEstimated value；

(3) determine comparative sequences x_kMiddle Failure Factors are to reference sequences inefficacy x₀The gray relation grades r of factor_0k(k=1, 2...n-1)；

(4) according to degree of association r_0kNumerical values recited carry out ordered series of numbers sequence, draw fire-fighting system occur chain inefficacy cabin Primary and secondary sorts, and judges which cabin associates maximum with system interlock failure, and at screening, fire-fighting system is susceptible to linksystem mistake The cabin of effect, determines the chain Failure Factors in this cabin；

(5) calculate x_mWith x_jThe grey relational grade of (m≤j, and m, j=0,1...n-1), carries out grey cluster by real-time monitored Failure Factors carry out merger, n chain Failure Factors can be gathered for f class (f≤n)；

(6) according to the chain Failure Factors in step (4), each class of the f class of step (5) find out representative inefficacy because Element, total p representative Failure Factors (n >=p >=f), set up the forecast model of bp artificial neural network, the input of neutral net Vector is p representative Failure Factors, is designated as h=(h_1,h_2,....h_p), output is y=(y₁,y₂,y₃), output vector is divided (1,0,0), (0,1,0), (0,0,1) is used to represent respectively for safety, Generally Recognized as safe, the three kinds of states that lost efficacy.

The beneficial effects of the present invention is: the invention provides the dynamic analysing method of ship's fire fighting thrashing, make up The deficiency of existing Static Analysis Technology, it is contemplated that coupling contact between cabin, judges that main fire-fighting system is chain Inefficacy cabin and Failure Factors, and the safe condition of fire-fighting system is judged according to the actual state of ship.

Brief description

Fig. 1 is the flow chart of the present invention.

Specific embodiment

The present invention is further described with reference to the accompanying drawings and examples.

(1) each cabin in ship's fire fighting system is set to a fire fighting unit, from substantial amounts of ship fire event statistics Draw the chain Failure Factors of n fire-fighting system as observation data.Set up 1 reference data array and the individual data that compares of n-1 arranges.Will The chain inefficacy of fire fighting unit limits threshold values as reference data array, and comparative sequences are the reality after the Failure Factors quantization of each fire fighting unit Border observation.Make x₀={x₀ ⁽⁰⁾(i), (i=1,2...n) } it is reference data array, x_k={x_k ⁽⁰⁾(i),(k=1,2...n-1;i=1, 2...n it is) } to compare data row (n-1 is the quantity in cabin), the element in each sequence represents the smog of sense smoke sensor respectively Concentration, the temperature of heat detector, smog increase slope, time of fire alarming, the open circuit of circuit, the short circuit of circuit and personnel and patrol and examine N chain Failure Factors (n chain Failure Factors can be changed according to the difference of Ship Types).

(2) set up Lycoperdon polymorphum Vitt gm (1, n-1) model:Wherein $x_m^{\left(1\right)}\left(i\right)=\underset{i=1}{\overset{k}{\mathrm{\σ}}}{x}_m^{\left(0\right)}\left(i\right)(k=\mathrm{1,2}....n),z_0^{\left(1\right)}=(z_0^{\left(1\right)}\left(2\right),z_0^{\left(1\right)}\left(3\right).....z_0^{\left(1\right)}\left(n\right))$ ForNext-door neighbour's average generation sequence,-a For system development coefficient, b_mFor drive factor.Order $b=\left[\begin{array}{cccc}-z_0^{\left(1\right)}\left(2\right)& x_1^{\left(1\right)}\left(2\right)& .....& x_{n-1}^{\left(1\right)}\left(2\right)\\ -z_0^{\left(1\right)}\left(3\right)& x_1^{\left(1\right)}\left(3\right)& .....& x_{n-1}^{\left(1\right)}\left(3\right)\\ .....& .....& .....& .....\\ -z_0^{\left(1\right)}\left(n\right)& x_1^{\left(1\right)}\left(n\right)& .....& x_{n-1}^{\left(1\right)}\left(n\right)\end{array}\right],q=\left[\begin{array}{c}{x}_0^{\left(0\right)}\left(2\right)\\ x_0^{\left(0\right)}\left(3\right)\\ ..\\ x_0^{\left(0\right)}\left(n\right)\end{array}\right],$ Then parameter row a=[a, b₁,b₂,..,b_n-1]^tLeast-squares estimation meet a=(b^tb)^-1b^tQ, then solve and obtain a and b_mValue.

(3) calculate degree of association r_0k(k=1,2...n-1).For 0 < ε < 1, make $r(x_0\left(i\right),x_k\left(i\right))=\frac{\underset{k}{\min }\underset{i}{\min }|x_0\left(i\right)-x_k\left(i\right)|+\mathrm{\&epsiv;}\underset{k}{\max }\underset{i}{\max }|x_0\left(i\right)-x_k\left(i\right)|}{|x_0\left(i\right)-x_k\left(i\right)|+\mathrm{\&epsiv;}\underset{k}{\max }\underset{i}{\max }|x_0\left(i\right)-x_k\left(i\right)|}$ Then r_0kIt is comparative sequences x_kMiddle Failure Factors are to reference sequences inefficacy x₀The gray relation grades of factor, show that this factor is sent out with that factor Exhibition change situation degree of closeness, its change in value scope is 0～1.r_0kNumerical value is closer to 1, and influence degree is bigger.

(4) according to degree of association r_0kNumerical values recited carry out ordered series of numbers sequence, show that fire-fighting system occurs chain inefficacy accordingly Cabin primary and secondary sequence, judges which cabin associates maximum with system interlock failure, filters out the fire-fighting system in which cabin It is susceptible to interlock failure, investigate out chain Failure Factors.

(5) x is calculated according to step (3) formula_mWith x_jThe grey relational grade of (m≤j, and m, j=0,1...n-1), obtains Upper triangular matrix $t=\left[\begin{array}{cccc}{\mathrm{\&omega;}}_{00}& {\mathrm{\&omega;}}_{01}& {\mathrm{\&omega;}}_{02}& {\mathrm{\&omega;}}_{0n-1}\\ & {\mathrm{\&omega;}}_{11}& {\mathrm{\&omega;}}_{12}& {\mathrm{\&omega;}}_{1n-1}\\ & & ....& ....\\ & & & {\mathrm{\&omega;}}_{n-1n-1}\end{array}\right],$ Take and determine marginal value β ∈ (0,1), work as ω_mjDuring >=β (m ≠ j), then It is considered as x_mWith x_jFor similar, thus the Failure Factors of real-time monitored have been carried out merger by grey cluster, n chain inefficacy because Element can be gathered for f class (f≤n), for example can be with merger behaviour, machine, environment, management 4 classes.

(6) sequence according to the chain Failure Factors in step (4), finds out representativeness in each class of the f class of step (5) Failure Factors, total p representative Failure Factors (n >=p >=f), set up the forecasting sequence of bp artificial neural network.According to step (2) gm (1, n-1) model is succeeded in one's scheme and is counted according to reducing valueI () (k=1,2.....n-1, i=1,2...n), sets up residual errorIf to { e_k ⁽⁰⁾(i) } with the residual sequence that bp Neural Network model predictive goes out be The expected sequence of construction artificial neural networks built-up patternBp neutral net input to Measure as p representative Failure Factors, be designated as h=(h_1,h_2,....h_p).Output is y=(y₁,y₂,y₃), output vector is divided into Safety, Generally Recognized as safe, the three kinds of states that lost efficacy use (1,0,0), (0,1,0), (0,0,1) to represent respectively.

Due to uncertain between the randomness of the various chain Failure Factors of ship's fire fighting system, ambiguity, each factor Dependency and the imperfection of statistics, the mutual relation between therefore each Failure Factors is unintelligible.The invention discloses A kind of Forecasting Methodology of the ship's fire fighting system interlock failure based on Grey Correlation Cluster and bp artificial neural network.The present invention Ship's fire fighting system is regarded as a gray system, using chain for fire-fighting system Failure Factors as fire-fighting system behavior characteristicss amount Process.Chain for ship's fire fighting system Failure Factors regard random quantity, within the specific limits be all grey colo(u)r specification in certain time With grey process, set up Lycoperdon polymorphum Vitt gm (1, n-1) model, wherein n-1 is cabin quantity.Carry out chain inefficacy using gray relation analysis method to send out The quantitative analysiss of exhibition situation, have obtained Failure Factors and the size of ship's fire fighting security of system energy influence degree have been sorted, clearly Main chain Failure Factors, thus be conducive to fire-fighting system fire prevention, the design of extinguishing property and inspection.By grey cluster pair N Failure Factors carry out merger and are processed as f class, and filter out representative Failure Factors in each apoplexy due to endogenous wind.Application bp artificial neuron Network carries out data sequence matching to gm (1, n-1) model, and the input and output problem of sample is converted into a nonlinear optimization Problem, the model that network is set up, after training, can get the non-linear relation between representative Failure Factors and cabin, to even Lock inefficacy cabin is predicted, for taking Prevention and control strategy to provide quantitative analyses further.The method is applicable not only to Ship's fire fighting system, the prediction to the interlock failure behavior of other complication systems of similar structures has universality.

(1) with each cabin for a fire fighting unit in ship's fire fighting system, total n-1 fire fighting unit, each Fire fighting unit as object of observation, draws the chain Failure Factors conduct of n fire-fighting system from substantial amounts of ship fire event statistics Observation data, sets up Lycoperdon polymorphum Vitt gm (1, n-1) model.This model comprises 1 reference data array and the individual data that compares of n-1 arranges.Fire-fighting The chain inefficacy of unit limits threshold values as reference data array, and comparative sequences are the actual sight after the Failure Factors quantization of each fire fighting unit Measured value.If x₀={x₀(i), (i=1,2...n) } it is reference data array, x_k={x_k(i),(k=1,2...n-1;I=1,2...n) } it is Relatively data row.Element in each sequence represents the sense smokescope of smoke sensor, the temperature of heat detector, smog respectively Increase slope, the n chain Failure Factors such as time of fire alarming, the open circuit of circuit, the short circuit of circuit and personnel patrol and examine.Set up gm (1, N-1) model.This model is $x_0^{\left(0\right)}\left(i\right)+{\mathrm{az}}_0^{\left(1\right)}\left(i\right)=\underset{m-1}{\overset{n-1}{\mathrm{\&sigma;}}}{b}_m{x}_m^{\left(1\right)}\left(i\right),$ Wherein $x_m^{\left(1\right)}\left(i\right)=\underset{i=1}{\overset{k}{\mathrm{\&sigma;}}}{x}_m^{\left(0\right)}\left(i\right)(k=\mathrm{1,2}....n),$ $z_0^{\left(1\right)}=(z_0^{\left(1\right)}\left(2\right),z_0^{\left(1\right)}\left(3\right).....z_0^{\left(1\right)}\left(n\right))$ ForNext-door neighbour's average generation sequence, $z_0^{\left(1\right)}\left(i\right)=\frac{1}{2}(x_0^{\left(1\right)}\left(i\right)+x_0^{\left(1\right)}(i-1)),$ -a For system development coefficient, b_mFor drive factor, solve by method of least square and can get a and b_mEstimated value.

(2) calculate degree of association r_0k(k=1,2...n-1), r_0kIt is comparative sequences x_kMiddle Failure Factors lose to reference sequences Effect x₀The gray relation grades of factor, its change in value scope is 0～1.r_iNumerical value is closer to 1, and influence degree is bigger.According to association Degree r_0kNumerical values recited carry out ordered series of numbers sequence, show that fire-fighting system occurs the cabin primary and secondary sequence of chain inefficacy accordingly, judge Which cabin associates maximum with system interlock failure, and the fire-fighting system filtering out which cabin is susceptible to interlock failure, Investigate out chain Failure Factors

(3) x is calculated according to step (2)_mWith x_jThe grey relational grade of (m≤j, and m, j=0,1...n-1), carries out Lycoperdon polymorphum Vitt The Failure Factors of real-time monitored are carried out merger by cluster.N chain Failure Factors can be gathered for f class (f≤n) so that losing A number of factors can be represented with the comprehensive average index of these Failure Factors or some Failure Factors therein in effect assessment And make information not be subject to heavy losses.The behavior of the chain inefficacy of ship's fire fighting system is thus compactly described, for simplifying chain mistake Effect assessment indicator system provides theoretical foundation.

(4) combine the mainly chain Failure Factors in inventive step (2), find out in each class of the f class of inventive step (3) Representative Failure Factors, total p representative Failure Factors (n >=p >=f), set up the forecast model of bp artificial neural network.God Input vector through network is p representative Failure Factors, is designated as h=(h_1,h_2,....h_p).Output is y=(y₁,y₂,y₃), Output vector is divided into safety, Generally Recognized as safe, the three kinds of states that lost efficacy use (1,0,0), (0,1,0), (0,0,1) to represent, in advance respectively Surveyed cabin fire fighting state and the behavior of the chain inefficacy of fire-fighting system, be ship's fire fighting system security decision provide with reference to according to According to.

The invention provides a kind of ship's fire fighting system interlock based on Grey Correlation Cluster and bp artificial neural network The Forecasting Methodology losing efficacy, does not only give and occurs the primary and secondary in chain inefficacy cabin to sort and mainly chain Failure Factors, and in advance Survey the behavior of ship's fire fighting system interlock failure, its thinking and method can be extended to other complexity systems with similar structures The chain failure prediction of system.

Claims (2)

1. a kind of Forecasting Methodology of ship's fire fighting system interlock failure is it is characterised in that comprise the steps:

(1) each cabin in ship's fire fighting system is set to a fire fighting unit, show that n disappears according to ship fire event statistics The chain Failure Factors of anti-system as observation data, set up 1 reference data array and n-1 compare data row, by fire fighting unit Chain inefficacy limits threshold values as reference data array, compares the actual sight after the Failure Factors that data sequence is each fire fighting unit quantify Measured value, wherein x₀={ x₀ ⁽⁰⁾(i), (i=1,2...n) } it is reference data array, x_k={ x_k ⁽⁰⁾(i), (k=1,2...n-1；I= 1,2...n) be } to compare data row, n-1 is the quantity in cabin, the element in each sequence represent respectively n chain inefficacy because Element；

(2) set up gm (1, n-1) model:Wherein

ForNext-door neighbour's average generation sequence Row,

A is system development coefficient, b_mFor drive factor, solved and can be obtained by method of least square To a and b_mEstimated value；

(3) determine comparative sequences x_kMiddle Failure Factors are to reference sequences inefficacy x₀The gray relation grades r of factor_0k(k=1,2...n-1),

(4) according to degree of association r_0kNumerical values recited carry out ordered series of numbers sequence, draw fire-fighting system occur chain inefficacy cabin primary and secondary Sequence, judges which cabin associates maximum with system interlock failure, filters out fire-fighting system and be susceptible to interlock failure Cabin, determines the chain Failure Factors in this cabin；

(5) calculate x_mWith x_jThe grey relational grade of (m≤j, and m, j=0,1...n-1), carries out grey cluster by real-time monitored Failure Factors carry out merger, and n chain Failure Factors can be gathered for f class (f≤n)；

(6) according to the chain Failure Factors in step (4), representative Failure Factors are found out in each class of the f class of step (5), Total p representative Failure Factors (n >=p >=f), sets up the forecast model of bp artificial neural network, the input of neutral net to Measure as p representative Failure Factors, be designated as h=(h₁,h₂,....h_p), output is y=(y₁,y₂,y₃), output vector is drawn It is divided into safety, Generally Recognized as safe, the three kinds of states that lost efficacy to use (1,0,0), (0,1,0), (0,0,1) to represent respectively.

2. a kind of ship's fire fighting system interlock failure according to claim 1 Forecasting Methodology it is characterised in that: described Chain Failure Factors include feeling the smokescope of smoke sensor, when the temperature of heat detector, smog increase slope, report to the police Between, the open circuit of circuit, the short circuit of circuit and personnel patrol and examine parameter.