CN106485327A - Under a kind of hazardous condition, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution - Google Patents

Abstract

The invention relates to a method for discovering knowledge about the evolution of crowd trampling behavior under disaster conditions, comprising the following steps: 1) extracting the features of crowd stampede scene elements, performing simulation in a swarm intelligent crowd evacuation model, obtaining the trampling evolution behavior characteristics during the evacuation process, and constructing 2) Rough set attribute discretization processing for the characteristics of the crowd's stampede scene elements and trampled evolution behavior characteristics; 3) Establish the knowledge of the trampled evolution mechanism of the "condition-decision" two-dimensional information model expression, and carry out knowledge reduction on the knowledge expression of the stepping evolution mechanism; 4) generate a meta-rule base according to the behavior characteristics of the stepping evolution; rule base. Compared with the prior art, the present invention realizes the automatic knowledge discovery of the evolution mechanism of crowd stampede behavior through classical rough set theory, and solves the problem of lack of knowledge of crowd stampede evolution mechanism.

Description

A Knowledge Discovery Method for the Evolution of Crowd Stampede Behavior under Disaster Conditions

technical field

The invention relates to the technical field of crowd stampede prevention, in particular to a knowledge discovery method for the evolution of crowd stampede behavior under disaster conditions.

Background technique

At present, the analysis and prevention of crowd stampede accidents are still in their infancy. The stampede during crowd evacuation may occur alone in the "rumor-panic" mode, or it may be a secondary disaster of sudden disasters (fires, earthquakes, explosions, poison gas leaks, etc.), coupled in the "disaster-panic" mode occur.

Most of the existing crowd evacuation models focus on the research of crowd kinematics and dynamics, mainly describing crowd evacuation behavior, and analyzing the relationship between building evacuation elements (minimum width of stairs and fire exits, etc.) and crowd capacity and evacuation time , some emergency management strategies and suggestions for preventing crowd stampede in buildings are given. Most of the existing models have not systematically studied the conditions and evolution mechanism of the stampede, and only consider the stampede as a typical instability phenomenon in the existing evacuation models.

Analyzing the conclusions of the existing research on crowd evacuation, it can be found that the research on the evolution mechanism of stampede is in the preliminary stage, and there are still problems such as dependence on knowledge and experience and lack of knowledge. For the prevention of stampedes, it mostly stays at the stage of contingency plan management based on experience. For the real-time control of stampedes, it mostly depends on the organizational experience of the emergency evacuation site commanders and the degree of cooperation of the crowd. It is urgent to use scientific methods to analyze the cause mechanism of crowd stampede and reveal the evolution law of crowd stampede.

Contents of the invention

The purpose of the present invention is to provide a knowledge discovery method for the evolution of crowd stampede behavior under disaster conditions in order to overcome the above-mentioned defects in the prior art, relying only on the actual domain record data itself, using the rough set theory to strictly sort out the logic, and automatically calculate and Discover knowledge about the evolution of crowd stampede behavior.

The purpose of the present invention can be achieved through the following technical solutions:

A knowledge discovery method for the evolution of crowd stampede behavior under disaster conditions, comprising the following steps:

1) Extracting the characteristics of crowd stampede scene elements, performing simulation in the swarm intelligence crowd evacuation model, obtaining the stampede evolution behavior characteristics in the evacuation process, and constructing the universe object set of the universe space layer, the stampede evolution behavior characteristics include stampede probability;

2) Carrying out rough set attribute discretization processing on described crowd stampede scene element feature and stampede evolution behavior feature;

3) Taking the discretized features of the stampede scene elements of the crowd as the condition attribute of the domain object, and using the corresponding trampled evolution behavior characteristics as the decision attribute of the domain object, to establish a "condition-decision" two-dimensional information model The stepping evolution mechanism knowledge expression of the stepping evolution mechanism, and performing knowledge reduction on the stepping evolution mechanism knowledge expression;

4) generating a meta-rule base according to the characteristics of the trampling evolution behavior;

5) Loading the meta-rule base and the evacuation instance to generate a generalization rule base of stepping evolution.

The characteristics of the crowd stampede scene elements include evacuated individual physiological factors, social factors, behavioral characteristics and environmental characteristics, the evacuated individual physiological factors include age, gender, disability degree, agility and weight, and the social factors include strangeness, so The behavioral characteristics include the degree of panic, and the environmental characteristics include disaster factors and space constraints.

The degree of panic is mapped to the individual diameter of the evacuated individual according to Helbing, D. panic "psychology-behavior" fluctuation model, and the solution process of the individual diameter is specifically:

f _iw ＝{A _i exp[(r _i -d _iw )/B _i ]+kg(r _i -d _iw )}n _iw -γg(r _i -d _iw )v _i t _iw (3)

In the formula, m _i is the mass of the i-th evacuated individual, is the ideal velocity of the i-th evacuated individual, is the setting direction of the i-th evacuated individual, v _i is the actual speed of the i-th evacuated individual, τ _i is the characteristic time of the i-th evacuated individual, t is time, f _ij is the distance between evacuated individual i and evacuated individual j , f _iw is the interaction force between evacuated individual i and the boundary, A _i and B _i are constants, d _cij is the distance between the mass centers of two evacuated individuals, d _ij is the distance between two evacuated individuals, n _ij is the standard vector pointing to i from evacuated individual j, t _ij is the tangential direction of n _ij , is the vector difference of velocity at time t, kg(d _ij -d _cij ) represents mass force, Indicates the sliding friction force at time t, k and γ are parameters that determine the blocking effect of the interaction between evacuated individuals i and j, d _iw is the distance between evacuated individual i and the boundary, n _iw refers to the vertical direction, t _iw refers to the tangential direction, r _i is the individual diameter of the i-th evacuated individual, v _i is the individual velocity of the i-th evacuated individual, g(x) is a function, if the evacuated individual collides, g(x)=0 , otherwise g(x)=x.

The disaster factor mapping is expressed as the individual speed of the evacuated individual after the panic spread, specifically:

Define the rule θ as:

In the formula, μ _DA is the membership function of disaster loss degree DA, DA _max is the maximum disaster loss degree, is the membership function of the risk assessment intensity I ₀ , I _max is the maximum risk assessment intensity, the subscript t is the sort number of h _dis , s is the serial number of i value, i is the coordinate value on the risk assessment intensity I ₀ , n is the maximum value of i, rs _st is the fuzzy relationship matrix Elements in , using the inference formula:

DA＝I ₀ θR (5)

Distribute the risk assessment intensity I ₀ to the control points by means of information distribution, and finally obtain the value of the disaster risk index h _dis , R is the fuzzy relationship matrix;

Calculate the individual velocity of each evacuated individual after panic propagation:

h _dis =f(ρ) (6)

In the formula, ρ is the human flow density, f( ) represents the linear relationship function between the disaster risk index h _dis and the human flow density ρ, D _L =NA _P /W _A L _A =ρA _P , and D _L is the unit area in the horizontal projection plane is the number of evacuated individuals, N is the total number of people walking in the flow, _AP is the horizontal projected area of _a single person, _WA is the width of the flow, LA is the length of the flow, v _i is the individual velocity of the i-th evacuated individual.

The crowd stampede scene element features are divided into qualitative features and quantitative features, when performing step 2), for qualitative features, directly map it to rough set discrete attributes; for quantitative features, first map it to rough set continuous attributes, and then Using the heuristic SOM self-organizing clustering model, the rough set continuous attributes are automatically discretized, and the rough set continuous attribute values are transformed into distinguishable mathematical symbols of the rough set matrix.

The knowledge reduction includes the domain space dimensionality reduction reduction by using the full-distance dimensionality reduction model, and the attribute reduction and attribute value reduction by using the rough set theory reduction and the kernel calculation model.

Using rough set theory to carry out knowledge reduction is as follows:

For the knowledge system S=(U,A), U is the set of objects in the universe of discourse, U={x ₁ ,x ₂ ,…,x _n }, where the elements x _i are objects in the universe of discourse, n is the total number of objects, A is a non-empty attribute set, A=C∪D, C is a conditional attribute set, D is a decision attribute set, and C∩D=φ, calculate the distinction matrix M _DS (C) of the knowledge system:

The element m _ij in the resolution matrix is the set of all conditional attributes that distinguish objects x _i and x _j ;

Define the Boolean function f _DS as follows:

where the Boolean variable Corresponding to m conditional attributes The symbol ∨ represents the disjunction operation, and the symbol ∧ represents the conjunction operation;

Calculate the decision matrix M _DS (C) of the knowledge system S:

elements in the decision matrix is the set of all conditional attributes that distinguish objects x _i and x _j ;

Define the decision function f _DRDS as follows:

here

Compared with the prior art, the present invention has the following advantages:

A rough set knowledge discovery model of the evolution mechanism of crowd stampede is constructed, and the knowledge discovery problem of crowd stampede evolution is transformed into the problem of automatic generation of rough set decision rules.

For the research on the feature extraction and discretization of crowd stampede scene elements, based on the heuristic automatic clustering model of self-organizing map (SOM) network, the key scene elements and evolution behavior characteristics of crowd stampede are extracted, and discretized into rough set domain objects The conditional attributes and decision attributes of the crowd stampede are constructed to construct a two-dimensional information table containing the evolutionary knowledge of the crowd stampede, and a knowledge expression system for the evolution mechanism of the crowd stampede is proposed in a concise and complete manner. The meta-rules of knowledge expression are automatically generated through the rough set matrix calculation model, and the example of the stampede scene in the process of crowd evacuation is loaded, and the meta-rules can be rigorously restored into generalized rules with realistic background significance, forming explicit knowledge. The consistency and completeness of the above methods provide a scientific basis for the system to calculate and discover the valuable knowledge contained in the trampling evolution behavior of the crowd, and become the innovation point of the present invention.

Description of drawings

Fig. 1 is a schematic diagram of the principle of the present invention.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is carried out on the premise of the technical solution of the present invention, and detailed implementation and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

As shown in Figure 1, this embodiment provides a knowledge discovery method for the evolution of crowd stampede behavior under disaster conditions, including the following steps:

Step 1. Extract the characteristics of crowd stampede scene elements, simulate in the crowd evacuation model of swarm intelligence, obtain the evolution behavior characteristics of stampede during the evacuation process, and construct the set of universe objects in the space layer of universe. The evolution behavior characteristics of stampede include stampede probability.

The characteristics of the crowd stampede scene elements include evacuated individual physiological factors, social factors, behavioral characteristics and environmental characteristics, the evacuated individual physiological factors include age, gender, disability degree, agility and weight, etc., and the social factors include unfamiliarity, The behavioral characteristics include the degree of panic, and the environmental characteristics include disaster factors and space constraints.

In the present invention, the degree of panic is mapped to the individual diameter of the evacuated individual according to the Helbing, D. panic "psychology-behavior" fluctuation model, and the degree of panic is measured from the degree of tension (panic psychology) and speed change of the evacuated individual, and the degree of panic is mapped For the diameter of evacuated individuals, a panic quantification model is established.

The solution process of the individual diameter is specifically:

f _iw ＝{A _i exp[(r _i -d _iw )/B _i ]+kg(r _i -d _iw )}n _iw -γg(r _i -d _iw )v _i t _iw (3)

In the formula, m _i is the mass of the i-th evacuated individual, is the ideal velocity of the i-th evacuated individual, is the setting direction of the i-th evacuated individual, v _i is the actual speed of the i-th evacuated individual, τ _i is the characteristic time of the i-th evacuated individual, t is time, f _ij is the distance between evacuated individual i and evacuated individual j , f _iw is the interaction force between evacuated individual i and the boundary, A _i and B _i are constants, d _cij is the distance between the mass centers of two evacuated individuals, d _ij is the distance between two evacuated individuals, n _ij is the standard vector pointing to i from evacuated individual j, t _ij is the tangential direction of n _ij , is the vector difference of velocity at time t, kg(d _ij -d _cij ) represents mass force, Indicates the sliding friction force at time t, k and γ are parameters that determine the blocking effect of the interaction between evacuated individuals i and j, d _iw is the distance between evacuated individual i and the boundary, n _iw refers to the vertical direction, t _iw refers to the tangential direction, r _i is the individual diameter of the i-th evacuated individual, v _i is the individual velocity of the i-th evacuated individual, g(x) is a function, if the evacuated individual collides, g(x)=0 , otherwise g(x)=x.

The model data of typical fires and gas leaks are exported from the disaster model library, and the key influencing factors of disasters on evacuation are extracted; the spatial structure data of large buildings related to evacuation are derived from the building model library, and the spatial constraints (such as exit width) are extracted. and wall position, etc.) are the key factors affecting evacuation. Hazard factors and spatial constraints are used as environmental characteristics and classified qualitatively and quantitatively. Disasters are the triggering conditions for the spread of panic among crowds. The disaster factors are mapped to the panic propagation model, and the disaster factors affect the speed of evacuating individuals.

In the present invention, the disaster factor mapping is expressed as the individual speed of the evacuated individual after the panic spread, specifically:

Define the rule θ as:

In the formula, μ _DA is the membership function of disaster loss degree DA, DA _max is the maximum disaster loss degree, is the membership function of the risk assessment intensity I ₀ , I _max is the maximum risk assessment intensity, the subscript t is the sort number of h _dis , s is the serial number of i value, i is the coordinate value on the risk assessment intensity I ₀ , n is the maximum value of i, rs _st is the fuzzy relationship matrix Elements in , using the inference formula:

DA＝I ₀ θR (5)

Distribute the risk assessment intensity I ₀ to the control points by means of information distribution, and finally obtain the value of the disaster risk index h _dis , R is the fuzzy relationship matrix;

Calculate the individual velocity of each evacuated individual after panic propagation:

h _dis =f(ρ) (6)

In the formula, ρ is the human flow density, f( ) represents the linear relationship function between the disaster risk index h _dis and the human flow density ρ, D _L =NA _P /W _A L _A =ρA _P , and D _L is the unit area in the horizontal projection plane is the number of evacuated individuals, N is the total number of people walking in the flow, _AP is the horizontal projected area of _a single person, _WA is the width of the flow, LA is the length of the flow, v _i is the individual velocity of the i-th evacuated individual.

Step 2: Perform rough set attribute discretization processing on the characteristics of crowd stampede scene elements and trampled evolution behavior features, use the feature attribute discretization algorithm based on SOM self-organizing map network to discretize the extracted evacuation features, and quantify and correlate them linearly map.

Described crowd stampede scene element feature is divided into qualitative feature and quantitative feature, when performing step 2), for qualitative feature (such as gender), it is directly mapped to rough set discrete attributes (such as 1 and 0 etc.); for quantitative feature ( Such as age, panic degree and stampede probability, etc.), first map it to rough set continuous attributes, and then use the heuristic SOM self-organizing clustering model to automatically discretize the rough set continuous attributes, and transform the rough set continuous attribute values It is a distinguishable mathematical symbol (such as 1, 2, ... etc.) for the rough set matrix.

Step 3. Using the discretized characteristics of the crowd stampede scene elements as the condition attribute of the domain object, and using the corresponding trampling evolution behavior characteristics as the decision attribute of the domain object, establish a "condition-decision" two-dimensional information The stepping evolution mechanism knowledge expression of the model, and the knowledge reduction is performed on the stepping evolution mechanism knowledge expression.

In terms of knowledge expression and reduction methods of stampede evolution, based on the discretized expression of the attributes of crowd stampede scene elements, the rough set theoretical knowledge expression model (domain, attribute, attribute value, information function) is used to analyze the scene elements (such as the number of people) , moving speed and degree of panic, etc.) the relationship and change characteristics between attributes form the attributes and attribute values of the knowledge expression system.

U＝{x ₁ ,x ₂ ,x ₃ ···x _n } (8)

Among them, U is the universe set, and each element x in the set, in this patent, x represents a single record of crowd evacuation results (including each characteristic value of crowd evacuation).

Taking the physiological, social, psychological and environmental characteristics of the situation elements as the condition attribute a _i of the rough set universe object, and taking the stampede probability in the simulation results of the swarm intelligence evacuation model as the decision attribute D of the rough set universe object, the "condition The information function expression of the two-dimensional information table mode of ——decision-making, forms the knowledge expression of the stepping evolution mechanism, which can be expressed as:

C＝{a ₁ ,a ₂ ,a ₃ ···a _n } (9)

Among them, C is a conditional attribute, and a ₁ can be set as age, a ₂ is gender, a ₃ is disability degree, a ₄ is agility, a ₅ is weight, a ₆ is the individual diameter mapped by panic level, and a ₇ is Individual velocities mapped by hazard factors. D is the decision attribute set.

In terms of knowledge reduction, under the condition of keeping the classification ability of the domain object set unchanged, the full-distance dimensionality reduction model is used to merge the domain objects, reduce the dimensionality of the domain space, and simplify the subsequent calculation of the domain space. Use the rough set theory reduction (Reduct) and core (Core) calculation model to delete redundant attributes of domain objects, and perform attribute reduction and attribute value reduction to simplify stepping on evolutionary knowledge. Let Q be independent, and Q∈C, if any

IND(Q)＝IND(C) (10)

Then Q is a reduction of the equivalence relation family C, and the set of all irreducible relations in C is the core of the equivalence relation family C, denoted as Core(C). There are multiple reductions, and Red(C) represents the set of all reductions of C.

In the present invention, knowledge reduction includes domain space dimensionality reduction reduction by using full-distance dimensionality reduction model, and attribute reduction and attribute value reduction by using rough set theory reduction and kernel calculation model.

Step 4. Rough set knowledge discovery of trampling evolution mechanism. A meta-rule base is generated according to the behavior characteristics of the stepping evolution, and the meta-rule base and the evacuation instance are loaded to generate a generalization rule base of the stepping evolution.

Using the classic Skowron matrix calculation method, construct a rough set matrix calculation model; for the knowledge system S=(U,A,V,f), U is the domain of discourse, x _i is the object in the domain of discourse, U={x ₁ ,x ₂ ,...,x _n }. A is a non-empty attribute set, A=C∪D, C is a conditional attribute, D is a decision attribute, and C∩D=φ. V represents the attribute value, and f is the information function. For system S, only conditional attributes are considered, and an information system about conditional attribute C is formed. The system’s distinction matrix DM, whose order is related to the number of objects in the domain of discourse, is n×n order, denoted as M _DS (C):

The element m _ij in the resolution matrix is the set of all condition attributes that distinguish objects x _i and x _j . For the variable j, the range of i is defined as 1≤j≤i≤n, and the lower triangular matrix is obtained, but when the variable j=i, that is, M _DS (C), the diagonal elements are obtained, and the result of the object’s own comparison is positive is the empty set φ. Therefore, in order to reduce n comparisons, the range of j and i is defined as: 1≤j<i≤n, which reduces the computational complexity of the original definition.

Introduce a Boolean function f _DS , called the discrimination function, as follows:

where the Boolean variable Corresponding to m conditional attributes The symbol ∨ represents the disjunction operation, and the symbol ∧ represents the conjunction operation;

To calculate the decision matrix of the knowledge system S, its order is related to the number of objects in the domain of discourse, that is, n×n order, denoted as M _DS (C):

With the calculation of the above matrix, it is found that this definition only reflects the comparison of objects other than the same value of the decision attribute, but does not describe the conflicting object (inconsistency). The matrix values of conflicting objects with the same condition attribute value but different decision attribute value are still empty sets, which cannot reflect the existence of conflict phenomenon, let alone the degree of conflict. For this reason, an improved definition of the matrix is made to eliminate the limitations.

elements in the decision matrix is the set of all conditional attributes that distinguish objects x _i from x _j .

Define the decision function f _DRDS as follows:

here

The purpose of rough set theory research is to simplify calculation, reduce computational complexity, optimize algorithm structure, and solve practical engineering problems more effectively. Based on the above research results, in the data preprocessing stage, MAWD dimensionality reduction calculations are performed on the space of discourse; for continuous attributes, the SOM network method based on improved IMDV is used for clustering calculations; for qualitative attributes or little change (ie The value is relatively fixed), and the direct discrete method is used here.

In this way, the optimization results of the above-mentioned links are integrated into an organic whole, and the MS-VPRS knowledge discovery model with improved overall performance is obtained. In the following chapters, we will study how to apply the IMSRS model to several typical diagnostic cases in the fields of engineering and medicine, to solve practical problems for enterprise users and medical researchers, and to study the effectiveness of the model from the aspects of knowledge discovery and decision support. Engineering verification for reliability and versatility.

The rough set matrix calculation model can generate the meta-rules of the attribute kernel set and input them into the meta-rule library; the meta-rules can be expressed as:

Rule 1:

if a ₃ = 0 and a ₄ = 0 and a ₇ = 1. then d = 0

Rule 2:

if a ₃ = 1 and a ₄ = 0 and a ₇ = 1. then d = 1

Rule 3:

if a ₃ = 0 and a ₄ = 1 and a ₇ = 0. then d = 1

Rule 4:

if a ₃ = 1 and a ₄ = 0 and a ₇ = 1. then d = 2

For meta-rule loading, the reversible process analysis method of attribute discretization is used to form a reversible mapping between the domain object attributes and the stampede scene elements; reload the domain object feature values, and the discrete mathematical symbols (such as the gender attribute value is 0) are restored become the element of the evacuation scenario (such as gender is male); in the rule generation part of the stepping evolution, the evacuation instance will be loaded, and the rough set meta-rules will be generalized into generalization rules for specific evacuation instances, forming generalization rules with realistic guiding significance repository to form explicit knowledge. For example, in rule 1, in the event of a severe disaster, the probability of stampede is 0 for a non-disabled, overweight female elderly without panic.

An example of Shanghai Hongqiao transportation hub is used to verify the above method.

The first step is to analyze and extract the environmental characteristics by extracting the building example data of Shanghai Hongqiao Comprehensive Transportation Hub, simulate the real situation of the site, and carry out the simulation. The total number of people is N=2109 (according to the population ratio data of the National Bureau of Statistics, there are 723 men, 686 women, 348 children, 221 elderly people and 131 disabled people).

The second step is to extract 4 records from the whole simulation to form the domain set U, and U={x ₁ ,x ₂ ,x ₃ ,x ₄ }. a ₁ is age, a ₂ is gender, a ₃ is disability degree, a ₄ is agility, a ₅ is body weight, a ₆ is individual diameter mapped by panic degree, a ₇ is individual speed mapped by disaster factors, C = {a ₁ , a ₂ , a ₃ , a ₄ , a ₅ , a ₆ , a ₇ }. Among them, 2 is the old man, 1 is the child, 0 is the middle-aged youth; 1 is the man, 0 is the woman; 1 is the handicapped, 0 is the non-disabled; 2, below normal weight is 1, normal weight is 0; panic is 2, nervousness is 1, normal is 0; disaster situation is serious is 1, disaster situation is not serious is 0, as shown in Table 1.

Table 1

u a ₁ a ₂ a ₃ a ₄ a ₅ a ₆ a ₇ D. _x1 2 0 0 0 2 0 1 0 _x2 0 0 1 0 0 1 1 1 _x3 0 1 0 1 0 2 0 1 _x4 1 1 1 0 1 2 1 2

The third step is to obtain the distinction matrix, discrimination function and decision function.

According to formula (11), the discrimination matrix is obtained:

The resolution function is:

f _DS ＝(a ₁ ∨a ₂ ∨a ₃ ∨a ₅ ∨a ₆ )(a ₁ ∨a ₂ ∨a ₄ ∨a ₅ ∨a ₆ ∨a ₇ )

(a ₂ ∨a ₃ ∨a ₄ ∨a ₆ ∨a ₇ )(a ₁ ∨a ₂ ∨a ₃ ∨a ₅ ∨a ₆ )

(a ₁ ∨a ₂ ∨a ₅ ∨a ₆ )(a ₁ ∨a ₃ ∨a ₄ ∨a ₅ ∨a ₇ )

= a ₁ a ₂ a ₄ a ₅ a ₇

According to formula (14), the resolution matrix is obtained:

Get the decision function:

f _DRDS ＝(a ₁ ∨a ₂ ∨a ₃ ∨a ₅ ∨a ₆ )(a ₁ ∨a ₂ ∨a ₄ ∨a ₅ ∨a ₆ ∨a ₇ )

(a ₂ ∨a ₃ ∨a ₄ ∨a ₆ ∨a ₇ )(a ₁ ∨a ₂ ∨a ₅ ∨a ₆ )

(a ₁ ∨ a ₃ ∨ a ₄ ∨ a ₅ ∨ a ₇ )

= a ₃ a ₄ a ₇

After reduction, it is shown in Table 2.

Table 2

u a ₃ a ₄ a ₇ D. _x1 0 0 1 0 _x2 1 0 1 1 _x3 0 1 0 1 _x4 1 0 1 2

According to the meta-rules, the value of the elements in the decision attribute set D (D is the probability of stampede, and its range is 0-1) is obtained; and then through the generalization rules, the decision attributes are translated into explicit knowledge, d ₁ =0 The probability of trampling is 0; d ₂ =1 means the probability of trampling is 0.1; d ₃ =1 means the probability of trampling is 0.1; d ₄ =2 means the probability of trampling is 0.2.

The invention proposes a knowledge discovery method for the evolution mechanism of crowd stampede, which extracts relevant characteristic elements from evacuation scene elements and crowd stampede evolution behavior under disaster conditions, uses the existing rule base for calculation, and obtains decision-making attribute values to prevent vicious stampede accidents. Scientific basis and theoretical support have important theoretical value and social significance.

Claims (7)

1. under a kind of hazardous condition, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that comprise the following steps：

1) extraction crowd tramples scenario factors feature, is emulated, during being evacuated in colony intelligence crowd evacuation model Evolved behavior feature is trampled, the domain object set of domain space layer is built, the evolved behavior feature of trampling includes to trample generally Rate；

2) trampling scenario factors feature to the crowd and trampling evolved behavior feature carries out rough set attribute sliding-model control；

3) scenario factors feature is trampled as the conditional attribute of domain object using the crowd after the sliding-model control, with corresponding The evolved behavior feature of trampling sets up trampling for " conditional decision " two-dimensional signal pattern as the decision attribute of domain object Evolution mechanism knowledge expression, and Reduction of Knowledge is carried out to the evolution mechanism knowledge expression of trampling；

4) meta-rule storehouse is generated according to the evolved behavior feature of trampling；

5) load the meta-rule storehouse and example is evacuated, generate the abstraction rule storehouse for trampling evolution.

2. under hazardous condition according to claim 1, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that The crowd tramples scenario factors feature to be included to evacuate individual physiologic factor, social factor, behavioural characteristic and environmental characteristic, described Evacuating individual physiologic factor includes age, sex, disability degree, agility and body weight, and the social factor includes strange degree, The behavioural characteristic includes panic degree, and the environmental characteristic includes disaster factors and space constraint.

3. under hazardous condition according to claim 2, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that The fear degree is mapped as evacuating individual individual diameter, institute according to Helbing, D. fear " Psychology and behavior " volatility model The solution procedure for stating individual diameter is specially：

$m_i=\frac{{\mathrm{dv}}_i}{dt}=m_i\frac{v_i^0\left(t\right)e_i^0\left(t\right)-v_i\left(t\right)}{{\mathrm{\&tau;}}_i}+\underset{j\left(\&NotEqual;i\right)}{\&Sigma;}{f}_{ij}+\underset{w}{\&Sigma;}{f}_{iw}---\left(1\right)$

$f_{ij}=\{A_i\exp \&lsqb;(d_{ij}-d_{cij})/B_i\&rsqb;+kg(d_{ij}-d_{cij})\}{n}_{ij}+\mathrm{\&gamma;}g(d_{ij}-d_{cij}){\mathrm{\&Delta;v}}_{ji}^t{t}_{ij}---\left(2\right)$

f_iw={ A_iexp[(r_i-d_iw)/B_i]+kg(r_i-d_iw)}n_iw-γg(r_i-d_iw)v_it_iw(3)

In formula, m_iIt is the individual quality of i-th evacuation,It is the individual ideal velocity of i-th evacuation,It is that i-th evacuation is individual Direction initialization, v_iIt is the individual actual speed of i-th evacuation, τ_iIt it is i-th evacuation individual characteristic time, t is the time, f_ij It is to evacuate the interaction force between individuality i and evacuation individuality j, f_iwIt is to evacuate interaction force of the individuality between i and border, A_i、B_iFor constant, d_cijIt is the individual mass centre's distance of two evacuations, d_ijDistance between individuality, n are evacuated for two_ijIt is individual by evacuating Body j points to the standard vector of i, t_ijIt is n_ijTangential direction,It is the phasor difference of t speed, kg (d_ij-d_cij) Represent mass force,Represent t force of sliding friction, k and γ is mutual between decision evacuation individuality i and j The parameter of the blocking effect of effect, d_iwIt is to evacuate the distance between individuality i and border, n_iwRefer to vertical direction, t_iwRefer to tangential Direction, r_iIt is the individual individual diameter of i-th evacuation, v_iIndividual individual speed is evacuated for i-th, g (x) is a function, such as Fruit is evacuated individuality and collides, g (x)=0, otherwise g (x)=x.

4. under hazardous condition according to claim 2, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that The disaster factors are mapped as the individual individual speed of evacuating after panic propagation and are expressed, specially：

Defining rule θ is：

${\mathrm{\&mu;}}_{DA}\left({\left(h_{dis}\right)}_t\right)=min\{1,\underset{s=1}{\overset{n}{\&Sigma;}}{\mathrm{\&mu;}}_{t_0}\left(i_s\right)r_{st}\}---\left(4\right)$

In formula, μ_DAFor the membership function of casualty loss degree DA,DA_maxFor maximum casualty loss degree,For wind Danger assessment intensity I₀Membership function,I_maxIntensity is assessed for greateset risk, subscript t is h_disSequence number, s It is the sequence number of i value, i is risk assessment intensity I₀On coordinate value, n for i maximum, r_stIt is the unit in fuzzy relation matrix Element, using inference formula：

DA=I₀θR (5)

By risk assessment intensity I₀It is assigned on control point in the method for information distribution, finally obtains calamity source index h_dis's Value, R are fuzzy relation matrix；

Calculate each individual speed for individuality being evacuated after panic propagation：

h_dis=f (ρ) (6)

$v_i=1.867D_L^4-6.333D_L^3+7.233D_L^2-3.617D_L+0.95---\left(7\right)$

In formula, ρ is density of stream of people, and f () represents calamity source index h_disWith density of stream of people ρ institute linear relationship function, D_L= NA_P/W_AL_A=ρ A_P, D_LIt is the evacuation individual amount of unit area in horizontal plane, N is the total number of persons that walks in the stream of people, A_PFor The horizontal projected area of single people, W_AFor the width of the stream of people, L_AFor the length of the stream of people, v_iIndividual individuality speed is evacuated for i-th Degree.

5. under hazardous condition according to claim 1, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that The crowd tramples scenario factors feature and is divided into qualitative features and quantitative characteristic, execution step 2) when, for qualitative features, directly It is mapped as rough set Category Attributes；For quantitative characteristic, rough set connection attribute is first mapped as, then using inspiration Formula SOM self-organizing clustering model, carries out automatic sliding-model control to rough set connection attribute, and rough set continuous property is converted For the distinguishable mathematic sign of rough set matrix.

6. under hazardous condition according to claim 1, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that The Reduction of Knowledge is included the domain space dimensionality reduction yojan carried out using full distance dimensionality reduction model and uses rough set theory about Attribute reduction and Value reduction that letter and core computation model are carried out.

7. under hazardous condition according to claim 6, crowd tramples the Methods of Knowledge Discovering Based of behavior evolution, it is characterised in that Reduction of Knowledge is carried out with rough set theory to be specially：

For knowledge system S=(U, A), U is domain object set, U={ x₁,x₂,…,x_n, element x therein_iFor in domain Object, n be object sum, A for non-NULL community set, A=C ∪ D, C are conditional attribute set, and D is decision kind set, and C ∩ D=φ, calculates the differentiation matrix M of the knowledge system_DS(C)：

Element m in discrimination matrix_ijIt is to discriminate between object x_iAnd x_jAll conditions attribute set；

Define Boolean function f_DSAs follows：

Wherein, Boolean variableCorresponding to m conditional attribute a₁,...,a_m,Symbol ∨ represents Extract computing, symbol ∧ represents conjunction computing；

The decision matrix M of calculation knowledge system S_DS(C)：

$M_{DRDS}\left(C\right)=\&lsqb;m_{{\mathrm{DS}}_{ij}}\left(x_i,x_j\right)\&rsqb;=\left\{\begin{array}{c}\left\{a\&Element;C|a\left(x_i\right)\&NotEqual;a\left(x_j\right)andd\left(x_i\right)\&NotEqual;d\left(x_j\right)\right\}\\ -1,a\left(x_i\right)=a\left(x_j\right)andd\left(x_i\right)\&NotEqual;d\left(x_j\right)\\ \mathrm{\&phi;},d\left(x_i\right)=d\left(x_j\right)\end{array}\right.---\left(10\right)$

Element in decision matrixIt is to discriminate between object x_iAnd x_jAll conditions attribute set；

Define decision function f_DRDSAs follows：

Here