CN106250667A - The monitoring method of a kind of landslide transition between states of paddling and device - Google Patents

Abstract

The invention provides monitoring method and the device of a kind of landslide transition between states of paddling, including: obtain land slide data and rainfall data；K mean algorithm and rainfall data are utilized to divide rain types；Obtain the evaporation capacity under every kind of rain types, infiltration capacity and run-off；According to particle group optimizing PSO algorithm, evaporation capacity, infiltration capacity and run-off are calculated, determine characteristics of rainfall vector；Build identification framework；The basic credibility being landslide transition between states of paddling described in evaluation index calculating with the reservoir level in landslide state, characteristics of rainfall vector and land slide data；According to identification framework and described basic credibility, utilize the total probability that landslide transition between states of paddling described in the calculating of Dempster composition rule occurs.

Description

The monitoring method of a kind of landslide transition between states of paddling and device

Technical field

The invention belongs to landslide disaster monitoring technical field, particularly relate to the monitoring of a kind of landslide transition between states of paddling Method and device.

Background technology

Along with use and the fast development of highway construction of reservoir in recent years, often come down in reservoir area, earthquake, mudstone The geological disasters such as stream, landslide；Communal facility, the people life property safety in this region are caused the biggest by the generation on landslide Threaten.

Affect a lot of because have of landslide transition between states of paddling, the most most importantly rainfall, reservoir level, subsoil water, fall Rain infiltrates, rainfall runoff etc..Landslide monitoring data corresponding to these principal elements are all time dependent data, are referred to as stream Data, the process of flow data is the very the key link of in the state research of landslide, and these data have numerical value at any time Between but or spatial variations, data volume is big, the quantity of information that comprises is many, difficult treatment, to landslide state is of crucial importance can not be direct The features such as use, therefore, prior art is when being monitored landslide transition between states, owing to can not effectively stream data carry out Process, cause accurately providing monitoring result, and then cause can not accurately providing early warning.

Based on this, monitoring method and the device of needing a kind of landslide transition between states of paddling at present badly are of the prior art to solve The problems referred to above.

Summary of the invention

The problem existed for prior art, embodiments provides the monitoring side of a kind of landslide transition between states of paddling Method and device, to solve to be effectively treated landslide monitoring data in prior art, cause can not accurately providing prison Survey result, and then the technical problem that lives and properties are threatened.

The present invention provides the monitoring method of a kind of landslide transition between states of paddling, and described method includes:

Obtain land slide data and rainfall data；

K mean algorithm and described rainfall data are utilized to divide rain types；

Obtain the evaporation capacity under every kind of rain types, infiltration capacity and run-off；

According to particle group optimizing (PSO, Particle Swarm Optimization) algorithm to described evaporation capacity, described Infiltration capacity and described run-off calculate, and determine characteristics of rainfall vector；

Build identification framework；

Reservoir level in and described land slide data vectorial with described landslide state, described characteristics of rainfall calculates for evaluation index The basic credibility of described landslide transition between states of paddling；

According to described identification framework and described basic credibility, utilize landslide of paddling described in the calculating of Dempster composition rule The total probability that transition between states occurs.

In such scheme, the described K of utilization mean algorithm and described rainfall data divide rain types and specifically include:

Predetermined interval cycle N, is divided into the rainfall number of times in the monitoring time M time with described cycle N；

Add up quantum of rainfall, rain time and duration during described M rainfall, constitutive characteristic data set；

Utilize K mean algorithm that described characteristic data set is carried out K mean cluster, determine K class rain types.

In such scheme, the described K of utilization mean algorithm carries out K mean cluster to described characteristic data set and specifically includes:

Preset cluster kind K；

K data point is randomly selected as initial cluster center at described characteristic data set；

Calculate the distance between all described data point and the initial cluster center in addition to described initial cluster center, and It is a nearest class that all described data point in addition to initial cluster center is classified as described distance；

When newly-increased described data point, update cluster centre, and calculate all described data in addition to current cluster centre Distance between point and current cluster centre；Until the square error convergence of all data points.

In such scheme, according to particle group optimizing PSO algorithm, described evaporation capacity, described infiltration capacity and described run-off are entered Row calculates, and determines that characteristics of rainfall vector specifically includes:

With the training error of multilayer feedforward neural network (BP, Back Propagation) as fitness function, utilize Described PSO algorithm calculates the weight coefficient that Landslide Stability is affected by described evaporation capacity, described infiltration capacity and described run-off；

Utilize described weight coefficient that the described evaporation capacity under identical rain types, described infiltration capacity and described run-off are entered Row weighting；

Described evaporation capacity, described infiltration capacity and described run-off after weighting under each rain types is separately summed, obtains Characteristics of rainfall；

Extract described characteristics of rainfall data, obtain described characteristics of rainfall vector.

In such scheme, described structure identification framework specifically includes:

Feature according to each state of described landslide sets up the decision problem of described identification framework；

Judge that described judgement is asked with described landslide state, described characteristics of rainfall vector and reservoir level respectively for evaluation index Topic, and obtain each judged result；

Carry out described each judged result comprehensively, constructing described identification framework.

Storehouse water in such scheme, in described vectorial with described landslide state, described characteristics of rainfall and described land slide data Position specifically includes for the basic credibility of landslide transition between states of paddling described in evaluation index calculating:

Statistical data legally constituted authority is utilized to count collection { x_kProbability P (the x of each evaluation index described in }_k)；

Utilize formulaBy the probability P (x of each evaluation index described_k) merge, obtain described Each evaluation index basic credibility m (C to described landslide transition between states of paddling_t)；Wherein, Described n, k are integer；Described t is evaluation index.

In such scheme, as the basic credibility m (C of described landslide transition between states of paddling_t) calculate after, described method Also include:

Utilize formula m ({ S_t)=CRE (X) × m (C_t) to described m (C_t) be modified；Wherein, described CRE (X) is institute State the definitiveness of X, described X={x_k}。

In such scheme, it is total general that landslide transition between states of paddling described in the calculating of the described Dempster of utilization composition rule occurs Rate specifically includes:

According to formulaCalculate total probability m (Y)；Wherein, described m₁(Y) According to the first basic probability assignment BPA of obtaining of described landslide state₁；Described m₂(Y) according to, described characteristics of rainfall is to measuring The the second basic probability assignment BPA arrived₂；Described m₃(Y) the 3rd basic probability assignment that according to, described Reservoir Water Level obtains BPA₃；Described K is normaliztion constant, and described Y is not empty set.

In such scheme, according to formulaCalculate K value.

The present invention also provides for the monitoring device of a kind of landslide transition between states of paddling simultaneously, and described device includes:

Acquiring unit, described acquiring unit is used for obtaining land slide data and rainfall data；Obtain under every kind of rain types Evaporation capacity, infiltration capacity and run-off；

Division unit, is used for utilizing K mean algorithm and described rainfall data to divide rain types；

First computing unit, is used for according to particle group optimizing PSO algorithm described evaporation capacity, described infiltration capacity and described footpath Flow calculates, and determines characteristics of rainfall vector；

Construction unit, for according to building identification framework；

Second computing unit, the storehouse in vectorial with described landslide state, described characteristics of rainfall and described land slide data Water level be evaluation index calculate described in paddle landslide shape body transition basic credibility；

3rd computing unit, for according to described identification framework and described basic credibility, utilizes Dempster synthesis rule The total probability that landslide transition between states of paddling described in then calculating occurs.

The invention provides monitoring method and the device of a kind of landslide transition between states of paddling, described method includes: obtain sliding Slope data and rainfall data；K mean algorithm and described rainfall data are utilized to divide rain types；Obtain under every kind of rain types Evaporation capacity, infiltration capacity and run-off；According to particle group optimizing PSO algorithm to described evaporation capacity, described infiltration capacity and described runoff Amount calculates, and determines characteristics of rainfall vector；Build identification framework；With described landslide state, described characteristics of rainfall vector and institute State the reservoir level in land slide data be evaluation index calculate described in paddle landslide transition between states basic credibility；According to described knowledge Other framework and described basic credibility, utilize the total general of landslide transition between states generation of paddling described in the calculating of Dempster composition rule Rate；So, in conjunction with particle group optimizing PSO algorithm and BP neutral net, the displacement monitoring data on landslide are carried out feature extraction, fortune By D-S evidence theory, data are merged, draw described in paddle the total probability that landslide transition between states occurs, and then can be to landslide Carry out accurate early warning.

Accompanying drawing explanation

The monitoring method schematic flow sheet of the landslide transition between states of paddling that Fig. 1 provides for the embodiment of the present invention one；

The state evolution curve signal of the landslide three phases under the action of gravity that Fig. 2 provides for the embodiment of the present invention one Figure；

The cluster result schematic diagram of the rain types that Fig. 3 provides for the embodiment of the present invention one；

The net rainfall that Fig. 4 provides for the embodiment of the present invention one and the corresponding relation schematic diagram of infiltration capacity；

The monitoring device overall structure schematic diagram of the landslide transition between states of paddling that Fig. 5 provides for the embodiment of the present invention two.

Detailed description of the invention

In order to enable that landslide monitoring data are effectively treated, accurately provide monitoring result, it is to avoid lives and properties cause Loss, the invention provides monitoring method and the device of a kind of landslide transition between states of paddling, and described method includes: obtain landslide number According to and rainfall data；K mean algorithm and described rainfall data are utilized to divide rain types；Obtain the evaporation under every kind of rain types Amount, infiltration capacity and run-off；According to particle group optimizing PSO algorithm, described evaporation capacity, described infiltration capacity and described run-off are entered Row calculates, and determines characteristics of rainfall vector；Build identification framework；Vectorial and the described cunning with described landslide state, described characteristics of rainfall Reservoir level in the data of slope be evaluation index calculate described in paddle landslide transition between states basic credibility；According to described identification frame Frame and described basic credibility, utilize the total probability that landslide transition between states of paddling described in the calculating of Dempster composition rule occurs.

Below by drawings and the specific embodiments, technical scheme is described in further detail.

Embodiment one

The present embodiment provides the monitoring method of a kind of landslide transition between states of paddling, as it is shown in figure 1, described method includes following Step:

Step 110, obtains land slide data and rainfall data.

In this step, it is necessary first to obtain land slide data and the rainfall data of monitoring point from Historical Monitoring data, described Land slide data may include that monitoring point displacement and reservoir level.The most also need to obtain simulation artificial rainfall experiment and obtain experimentation In land slide data and rainfall data.

During simulation artificial rainfall experiment, need to increase landslide with the time m-displacement accumulative displacement curve of monitoring point The data of state, according to described landslide status data by the ground on landslide, vegetative coverage situation, landslide displacement, rain fall, storehouse Water level conditions etc. simulate.

Here, after getting land slide data, also to set up landslide state according to described land slide data definition landslide state Judgment criterion；Specifically, the Monitoring Data first passing through a large amount of landslides example shows, under gravity, and slope Rock And Soil Deformation evolution curve has three stage evolution features.With reference to Fig. 2, three phases is respectively as follows: landslide state and is divided into initial deformation rank Section, at the uniform velocity deformation stage and acceleration deformation stage.Wherein, the AB section in initial deformation stage, i.e. Fig. 2.Slope, landslide body deformability At the initial stage, along with deformation is grown out of nothing, there is rock fracture in surface, landslide.The slope of this stage deformation curve is relatively large, but The passage of deformation time at any time tends towards stability, and deformation curve slope diminishes, and deformation velocity reduces.

BC section in constant speed deformation stage, i.e. Fig. 2, constant speed deformation stage, on the basis of the initial deformation stage, comes down Deformation velocity is constant, and Landslide Deformation curve is substantially in an angled straight lines, and macroscopic deformation observed result is basically unchanged.

CD section in accelerating deformation stage, i.e. figure.After Landslide Deformation develops into a certain degree, deformation curve oblique Rate.Increase, until before coming down, deformation curve is bordering on and rises steeply.

Then according to the definition of three landslide states, the landslide displacement situation of every month is analyzed, draws quantitative Landslide state demarcation condition.Landslide reality is combined by landslide displacement, displacement acceleration, three aspects of the landslide displacement of last month Situation judges landslide state.

Step 111, utilizes K mean algorithm and described rainfall data to divide rain types.

In this step, after getting rainfall data, predetermined period N, with described cycle N by the rainfall in the monitoring time time Number is divided into M time；Add up average daily rainfall r, rainfall natural law d and duration D, constitutive characteristic data set during described M rainfall； Utilize K mean algorithm that described characteristic data set is carried out K mean cluster, according to similar maximum comparability, foreign peoples's maximum diversity Determine K class rain types.

Wherein, utilize K mean algorithm that described characteristic data set is carried out K mean cluster and specifically include: first, set final Cluster kind K；Then, K data point is randomly selected as initial cluster center at described characteristic data set；Furthermore, calculate All data points in addition to initial cluster center and the distance of initial cluster center, and by owning in addition to initial cluster center It is that nearest class that data point is classified as described distance；And when new data point adds certain class, it is required to recalculate Cluster centre；After cluster centre updates, and calculate all described data point in addition to current cluster centre and current cluster Distance between center, the most constantly repeats to ask distance to carry out sorting out and update these 2 steps of cluster centre, until square error Convergence.Wherein, described square error can calculate according to formula (1):

$J=\underset{j=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{n}{\Σ}}\left|\right|x_i^{\left(j\right)}-c_j|{|}^2---\left(1\right)$

Wherein, in formula (1), described J be all data points square error and, described x_iIt is characterized in data set Data point, described c_jCluster centre for jth class.

Here, when choosing initial cluster center, strive for that K initial cluster center can be maximum dissimilar, can with this Improve the accuracy of category division.Meanwhile, choose appropriate initial cluster center and can accelerate convergence of algorithm speed, therefore need The quadratic sum of K initial cluster center distance each other is maximum, on this basis, chooses and meets K the node work making L maximum For initial cluster center.

$L=\underset{j=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{k}{\Σ}}\left|\right|x_i-x_j|{|}^2---\left(2\right)$

Wherein, in formula (2), described L is initial cluster center average and distance, described x each other_jIt is characterized Data point in data set.

Finally, after utilizing K mean algorithm that described characteristic data set is carried out K mean cluster, finally give 4 kinds of rain types Cluster result, through analyze be respectively continuous rainfall, heavy rain, interrupted rainfall, fragmentary rainfall；Can be found in Fig. 3.

Wherein, due to numerical range and the unit property of there are differences, it is therefore desirable to the index of catchment is carried out pretreatment To guarantee that cluster result has in higher class discrimination between concordance and class.In order to the index of reconstruct catchment is special Levying, effective rainfall natural law d directly uses, and original catchment index is carried out conversion simultaneously and obtains r and T.

Specifically, described average daily rainfall r can calculate according to formula (3):

R=(R/d) * p₁ (3)

Wherein, in formula (3), described R is quantum of rainfall, p₁For coefficient, the numerical intervals of described p1 is [0.1,1], Step-length is 0.1.

Described rain time d can calculate according to formula (4):

D=d*p₂ (4)

Wherein, p₂For coefficient, in order to the index feature of reconstruct catchment, effective rainfall natural law d directly uses, p₂ It is taken as 1.

Here, it is also possible to calculate the ratio T of rainfall natural law d and duration D according to formula (5)；Described T is used for judging fall Whether rain is interrupted.

T=(d/D) * p₃ (5)

Wherein, p₃For coefficient, the numerical intervals of described p3 is [1,11], and step-length is 1.

After rain types determines, obtain the evaporation capacity under every kind of rain types, infiltration capacity and run-off.

Specifically, when determining evaporation capacity, if the rainfall on the same day is not more than the daily evaporation amount of this month, the then rainfall on the same day The all evaporation capacity of data, net rainfall is 0.

When determining infiltration capacity, double-ring infiltration experiment can be carried out, and combine artificial rainfall experiment and obtain net rainfall and enter The corresponding relation of milliosmolarity, and then obtain infiltration capacity.

When determining run-off, it is possible to use daily rainfall and daily evaporation amount, day infiltration capacity difference as diurnal courses amount.

Step 112, is carried out described evaporation capacity, described infiltration capacity and described run-off according to particle group optimizing PSO algorithm Calculate, determine characteristics of rainfall vector.

In this step, using the training error of BP neutral net as fitness function, described PSO algorithm is utilized to calculate described The weight coefficient that Landslide Stability is affected by evaporation capacity, described infiltration capacity and described run-off；Specifically, in conjunction with PSO algorithm and BP neural network algorithm asks for the weight coefficient of 12 dimensions of one group of evaporation capacity, described infiltration capacity and described run-off.BP god is set Being 30000 through the frequency of training of network, training precision is e^-5, learning rate is 0.05, using the mean square deviation of its training error as PSO Fitness function.The Population Size arranging PSO algorithm is 20, and inertia coeffeicent is S type, and scope is 1 to drop to 0.4, Studying factors Being 2, the 12 dimensional weight coefficients finally obtained are [0.307,0.522,0.171,0.371,0.395].

Here, the collection of potential solutions all in problem to be optimized is collectively referred to as " solution space " by particle swarm optimization algorithm PSO, one Potential solution is referred to as one " position ", is designated as Pos.It is one and there is no quality by abstract for every bird, there is no " particle " of size, grain Son with certain " speed " flight, is designated as V, and evaluates the excellent degree of particle position with fitness function in solution space.Logical Cross position and the speed of the more new particle that constantly flies, and chase personal best particle and colony's optimal location seeks optimization problem Optimal solution, is designated as P respectively_ibAnd P_gb.Remember that the speed after the renewal of each particle and position are respectively V_i+1, Pos_i+1, then:

V_i+1=ω * V_i+c₁*rand*(P_ib-Pos_i)+c₂*rand*(P_gb-Pos_i) (6)

Pos_i+1=Pos_i+V_i+1 (7)

Wherein, subscript i represents iterations；ω represents the inertia coeffeicent of particle, and value is a dull reduction of (0,1) Sequence, reflection particle inherits the degree of original speed；Rand represents the random number that a value is (0,1)；c₁And c₂For study because of Son, the degree that reflection particle is drawn close to personal best particle and colony's optimal location.

In units of monthly, utilize described weight coefficient to the described evaporation capacity under identical rain types, described infiltration capacity And described run-off is weighted；By described evaporation capacity, described infiltration capacity and described run-off after weighting under each rain types It is separately summed, obtains the characteristics of rainfall of this month, be designated as α, β, γ respectively；Characteristics of rainfall data described in feature extraction, obtain described Characteristics of rainfall vector [α, β, γ].

Step 113, builds identification framework.

In this step, when building identification framework, comprehensively analyze according to the feature of each state that comes down, set up one The decision problem (decision problem of described identification framework) of " whether landslide transits to accelerate deformation state ".The history on comprehensive landslide Physical record and experimental record, analyze and determine the principal element of impact landslide transition between states, be evaluation index.Each evaluation Index has respective several result for decision problem, and all results being combined just has constructed identification framework.

Specifically, when landslide is in initial deformation stage or constant speed deformation stage, slope body deformability is relatively slow, and crack can not made big Impact, be now difficult to come down.But accelerating deformation stage, the rate of deformation of slope body quickly, drastically strengthen by crack, this Time it may happen that landslide.Therefore the decision problem set up is: " whether landslide transits to accelerate deformation state ".

Just there is different principal elements for different landslides, according to geologic feature and the influence factor on specific landslide, combine Close the history on this landslide, determine its impact landslide transition between states main factor to acceleration deformation stage, i.e. evaluation index, basis Evaluation index in embodiment is landslide state, described characteristics of rainfall vector and reservoir level.

Further, for " whether landslide transits to accelerate deformation state " this decision problem, each evaluation index has not With judged result, with described landslide state, described characteristics of rainfall vector and reservoir level for evaluation index judge respectively described in ask Topic, and obtain each judged result；Carry out described each judged result comprehensively, constructing described identification framework, described identification framework Subset is proposition.For example, if set Θ be a finite aggregate, and its element be the mutual exclusion about a certain problem domain and Comprehensively proposition is it is assumed that then gathering Θ is called identification framework.Identify and be meant that: for an enquirement, from relevant to this enquirement All possible answer in can only distinguish a correct answer.Any subset A of identification framework Θ all with a problem The proposition of answer is corresponding.This proposition is generally described as " answer of problem is in A ".

Step 114, the reservoir level in and described land slide data vectorial with described landslide state, described characteristics of rainfall is for evaluating Index calculate described in paddle landslide transition between states basic credibility.

In this step, after evaluation index determines, according to D-S evidence theory, with described landslide state, described rainfall spy Levy the reservoir level in vectorial and described land slide data be evaluation index calculate described in paddle landslide transition between states basic credibility, Specifically include:

In units of the moon, statistical data legally constituted authority is utilized to count collection { x_kProbability P (the x of each evaluation index described in }_k)；

According to formula (8) by probability P (x_k) merge, obtain each evaluation index described to described landslide state of paddling The basic credibility m (C of transition_t)；

$m\left(C_t\right)=\underset{x_k\∈\mathrm{\θ}}{\Σ}P\left(x_k\right)---\left(8\right)$

In formula (8),Described n, k are integer；Described t is evaluation index, described θ Set for each evaluation index probability.

Here, it is also possible to calculate the basic probability assignment of exceptional value according to formula (9), the unknown journey to test data is represented Degree:

$m\left({\mathrm{\δ}}_0\right)=1-\underset{t=1}{\overset{n}{\Σ}}m\left(C_t\right)---\left(9\right)$

Further, close the material composition of the formation condition, influence factor, evolution trend and the slip mass that consider landslide, with Time take into account landslide slip sampling, experimental apparatus, the impact of anthropic factor, it is necessary to analyze experimental data set X={x_kReliability, and Obtain deterministic quantized values CRE (E), revise basic credibility m (C further_t), it may be assumed that

m({S_t)=CRE (X) × m (C_t) (10)

Wherein, described CRE (X) is the definitiveness of described X.

Here, it is also possible to calculate { S according to formula (11)_tUnknown degree:

$m\left(\mathrm{\θ}\right)=1-{\mathrm{\Σ}}_{t=1}^{n}CRE\left(X\right)\×m\left(C_t\right)---\left(11\right)$

Step 115, according to described identification framework and described basic credibility, utilizes Dempster composition rule to calculate described The total probability that landslide transition between states of paddling occurs.

In this step, Dempster compositional rule is the rule reflecting multiple evidence combined effects, at same identification frame Under frame, having the confidence function of several different evidence, Dempster compositional rule just can confidence functions based on these evidences Calculate a final confidence level function.It is to say, in the present embodiment, landslide state be evaluation index draw the most general Rate assigns (basic credibility) BPA₁It is a confidence function m₁(Y)；With characteristics of rainfall vector for evaluation index draw the most general Rate assigns (basic credibility) BPA₂It is a confidence function m₂(Y)；The basic probability assignment drawn for evaluation index with reservoir level (basic credibility) BPA₃It is a confidence function m₃(Y), then landslide state of just can paddling according to formula (12) calculating The total probability that transition occurs:

$m\left(Y\right)=\frac{1}{K}\underset{Y_1\∩Y_2\∩Y_3=Y}{\Σ}{m}_1\left(Y\right)*m_2\left(Y\right)*m_3\left(Y\right)---\left(12\right)$

Wherein, in formula (12), described K is normaliztion constant, and described Y is not empty set.

Further, described normaliztion constant K can calculate according to formula (13):

The monitoring method of the landslide transition between states of paddling of the present embodiment offer, in conjunction with particle group optimizing PSO algorithm and BP god Through network, the displacement monitoring data on landslide are carried out feature extraction, use D-S evidence theory that data are merged, draw described The total probability that landslide transition between states of paddling occurs, and then landslide can be carried out accurate early warning.

Embodiment two

Corresponding to embodiment one, the present embodiment also provides for the monitoring device of a kind of landslide transition between states of paddling, such as Fig. 4 institute Showing, described device includes: acquiring unit 41, division unit the 42, first computing unit 43, construction unit the 44, second computing unit 45, the 3rd computing unit 46；Wherein,

Described acquiring unit 41 is used for obtaining land slide data and rainfall data；Specifically, first described acquiring unit 41 needs To obtain land slide data and the rainfall data of monitoring point from Historical Monitoring data, described land slide data may include that monitoring point Displacement and reservoir level.The most also need to obtain simulation artificial rainfall experiment and obtain the land slide data in experimentation and rainfall data.

During simulation artificial rainfall experiment, described acquiring unit 41 needs to add up position with the time m-displacement of monitoring point Move curve and increase the data of landslide state, according to described landslide status data by the ground on landslide, vegetative coverage situation, position, landslide Shifting, rain fall, reservoir level situation etc. simulate.

Here, after described acquiring unit 41 gets land slide data, also will be according to described land slide data definition landslide shape State, sets up landslide condition adjudgement criterion；Specifically, the Monitoring Data first passing through a large amount of landslides example shows, in action of gravity Under, the deformation evolution curve of slope Rock And Soil has three stage evolution features.With reference to Fig. 2, three phases is respectively as follows: landslide state It is divided into initial deformation stage, at the uniform velocity deformation stage and accelerates deformation stage.Wherein, the AB in initial deformation stage, i.e. Fig. 2 Section.At the landslide slope body deformability initial stage, along with deformation is grown out of nothing, there is rock fracture in surface, landslide.This stage deformation curve oblique Rate is relatively large, but the passage of deformation time at any time tends towards stability, and deformation curve slope diminishes, and deformation velocity reduces.

BC section in constant speed deformation stage, i.e. Fig. 2, constant speed deformation stage, on the basis of the initial deformation stage, comes down Deformation velocity is constant, and Landslide Deformation curve is substantially in an angled straight lines, and macroscopic deformation observed result is basically unchanged.

CD section in accelerating deformation stage, i.e. figure.After Landslide Deformation develops into a certain degree, deformation curve oblique Rate.Increase, until before coming down, deformation curve is bordering on and rises steeply.

The landslide displacement situation of every month, according to the definition of three landslide states, is carried out point by the most described division unit 42 Analysis, draws quantitative landslide state demarcation condition.By landslide displacement, displacement acceleration, three aspects of the landslide displacement of last month The situation actual in conjunction with landslide judges landslide state.

Further, after described acquiring unit 41 obtains rainfall data, described division unit 42 is additionally operable to utilize K average Algorithm and described rainfall data divide rain types；Specifically, described division unit 42 with default cycle N by the monitoring time Rainfall number of times be divided into M time；Add up average daily rainfall r, rainfall natural law d and duration D during described M rainfall, constitute spy Levy data set；Utilize K mean algorithm that described characteristic data set is carried out K mean cluster, according to similar maximum comparability, foreign peoples Big diversity determines K class rain types.

Wherein, utilize K mean algorithm that described characteristic data set is carried out K mean cluster and specifically include: first, set final Cluster kind K；Then, K data point is randomly selected as initial cluster center at described characteristic data set；Furthermore, calculate The distance between all data points and initial cluster center in addition to initial cluster center, and by addition to initial cluster center All data points are classified as distance for that nearest class；And when new data point adds certain class, it is required to recalculate Cluster centre；After cluster centre updates, and calculate all described data point in addition to current cluster centre and current cluster Distance between center, the most constantly repeats to ask distance to carry out sorting out and update these 2 steps of cluster centre, until square error Convergence.Wherein, described square error can calculate according to formula (1):

$J=\underset{j=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{n}{\Σ}}\left|\right|x_i^{\left(j\right)}-c_j|{|}^2---\left(1\right)$

Wherein, in formula (1), described J be all data points square error and, described x_iIt is characterized in data set Data point, described c_jCluster centre for jth class.

Here, when choosing initial cluster center, strive for that K initial cluster center can be maximum dissimilar, can with this Improve the accuracy of category division.Meanwhile, choose appropriate initial cluster center and can accelerate convergence of algorithm speed, therefore need The quadratic sum of K initial cluster center distance each other is maximum, on this basis, chooses and meets K the node work making L maximum For initial cluster center.

$L=\underset{j=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{k}{\Σ}}\left|\right|x_i-x_j|{|}^2---\left(2\right)$

Wherein, in formula (2), described L is initial cluster center average and distance, described x each other_jIt is characterized Data point in data set.

Finally, after utilizing K mean algorithm that described characteristic data set is carried out K mean cluster, finally give 4 kinds of rain types Cluster result, through analyze be respectively continuous rainfall, heavy rain, interrupted rainfall, fragmentary rainfall, can be found in Fig. 3.

Wherein, due to numerical range and the unit property of there are differences, it is therefore desirable to the index of catchment is carried out pretreatment To guarantee that cluster result has in higher class discrimination between concordance and class.In order to the index of reconstruct catchment is special Levying, effective rainfall natural law d directly uses, and original catchment index is carried out conversion simultaneously and obtains r and T.

Specifically, described average daily rainfall r can calculate according to formula (3):

R=(R/d) * p₁ (3)

Wherein, in formula (3), described R is quantum of rainfall, p₁For coefficient, the numerical intervals of described p1 is [0.1,1], Step-length is 0.1.

Described rain time d can calculate according to formula (4):

D=d*p₂ (4)

Wherein, p₂For coefficient, typically it is taken as 1.

Here, it is also possible to calculate the ratio T of rainfall natural law d and duration D according to formula (5)；Described T is used for judging fall Whether rain is interrupted.

T=(d/D) * p₃ (5)

Wherein, p₃For coefficient, the numerical intervals of described p3 is [1,11], and step-length is 1.

Here, after described division unit 42 is complete by division of rain types, described acquiring unit 41 is additionally operable to obtain Evaporation capacity, infiltration capacity and run-off under every kind of rain types.

Specifically, when determining evaporation capacity, if the rainfall on the same day is not more than the daily evaporation amount of this month, the then rainfall on the same day The all evaporation capacity of data, described acquiring unit 41 determines that net rainfall is 0.

Described acquiring unit 41, when determining infiltration capacity, can carry out double-ring infiltration experiment, and combine artificial rainfall experiment Obtain the corresponding relation of net rainfall and infiltration capacity, and then obtain infiltration capacity.

Described acquiring unit 41 is when determining run-off, it is possible to use daily rainfall and daily evaporation amount, the difference of day infiltration capacity Value is as diurnal courses amount.

After described acquiring unit 41 gets the evaporation capacity under every kind of rain types, infiltration capacity and run-off, described One computing unit 43 is for carrying out described evaporation capacity, described infiltration capacity and described run-off according to particle group optimizing PSO algorithm Calculate, determine characteristics of rainfall vector.Described first computing unit 43 using the training error of BP neutral net as fitness function, Described PSO algorithm is utilized to calculate the weight system that Landslide Stability is affected by described evaporation capacity, described infiltration capacity and described run-off Number；Specifically, described first computing unit 43 combine PSO algorithm and BP neural network algorithm ask for one group of evaporation capacity, described in enter The weight coefficient of 12 dimensions of milliosmolarity and described run-off.The frequency of training arranging BP neutral net is 30000, and training precision is e^-5, learning rate is 0.05, using the mean square deviation of its training error as the fitness function of PSO.The Population Size of PSO algorithm is set Being 20, inertia coeffeicent is S type, and scope is 1 to drop to 0.4, and Studying factors is 2, the 12 dimensional weight coefficients finally obtained for [0.307, 0.522,0.171,0.371,0.395].

Here, the collection of potential solutions all in problem to be optimized is collectively referred to as " solution space " by particle swarm optimization algorithm PSO, one Potential solution is referred to as one " position ", is designated as Pos.It is one and there is no quality by abstract for every bird, there is no " particle " of size, grain Son with certain " speed " flight, is designated as V, and evaluates the excellent degree of particle position with fitness function in solution space.Logical Cross position and the speed of the more new particle that constantly flies, and chase personal best particle and colony's optimal location seeks optimization problem Optimal solution, is designated as P respectively_ibAnd P_gb.Remember that the speed after the renewal of each particle and position are respectively V_i+1, Pos_i+1, then:

V_i+1=ω * V_i+c₁*rand*(P_ib-Pos_i)+c₂*rand*(P_gb-Pos_i) (6)

Pos_i+1=Pos_i+V_i+1 (7)

Wherein, subscript i represents iterations；ω represents the inertia coeffeicent of particle, and value is a dull reduction of (0,1) Sequence, reflection particle inherits the degree of original speed；Rand represents the random number that a value is (0,1)；c₁And c₂For study because of Son, the degree that reflection particle is drawn close to personal best particle and colony's optimal location.

In units of monthly, utilize described each weight coefficient to the described evaporation capacity under identical rain types, described in infiltrate Amount and described run-off are weighted；By described evaporation capacity, described infiltration capacity and described runoff after weighting under each rain types Amount is separately summed, and obtains the characteristics of rainfall of this month, is designated as α, β, γ respectively；Characteristics of rainfall data described in feature extraction, obtain institute State characteristics of rainfall vector [α, β, γ].

Here, after described first computing unit 43 determines characteristics of rainfall vector, described construction unit 44 is used for building knowledge Other framework.Specifically, described construction unit 44, when building identification framework, carries out total score according to the feature of each state that comes down Analysis, sets up " whether landslide transits to accelerate deformation state " decision problem (decision problem of described identification framework).Comprehensive landslide History physical record and experimental record, analyze and determine impact landslide transition between states principal element, be evaluation index.Often Individual evaluation index has respective several result for decision problem, and all results being combined just has constructed identification frame Frame, the subset of described identification framework is proposition.For example, if set Θ is a finite aggregate, and its element is about certain Mutual exclusion and the comprehensive proposition of one problem domain are it is assumed that then gathering Θ is called identification framework.Identify and be meant that: one is carried Ask, a correct answer can only be distinguished from all possible answer relevant to this enquirement.Identification framework Θ's is arbitrary Subset A is all corresponding with the proposition of a problem answers.This proposition is generally described as " answer of problem is in A ".

Specifically, when landslide is in initial deformation stage or constant speed deformation stage, slope body deformability is relatively slow, and crack can not made big Impact, be now difficult to come down.But accelerating deformation stage, the rate of deformation of slope body quickly, drastically strengthen by crack, this Time it may happen that landslide.Therefore the decision problem set up is: " whether landslide transits to accelerate deformation state ".

Just there is different principal elements for different landslides, according to geologic feature and the influence factor on specific landslide, combine Close the history on this landslide, determine its impact landslide transition between states main factor to acceleration deformation stage, i.e. evaluation index, basis Evaluation index in embodiment is landslide state, described characteristics of rainfall vector and reservoir level.

Further, for " whether landslide transits to accelerate deformation state " this decision problem, each evaluation index has not With judged result, with described landslide state, described characteristics of rainfall vector and reservoir level for evaluation index judge respectively described in ask Topic, and obtain each judged result；Described each judged result is carried out comprehensively, constructing described identification framework by described construction unit 44.

Further, described second computing unit 45 is for vectorial and described with described landslide state, described characteristics of rainfall Reservoir level in land slide data be evaluation index calculate described in paddle landslide transition between states basic credibility；Specifically, when commenting After valency index determines, described second computing unit 45 is according to D-S evidence theory, with described landslide state, described characteristics of rainfall Reservoir level in vectorial and described land slide data is the basic credibility that evaluation index calculates landslide transition between states of paddling, and specifically wraps Include:

In units of the moon, statistical data legally constituted authority is utilized to count collection { x_kProbability P (the x of each evaluation index described in }_k)；

According to formula (8) by probability P (x_k) merge, obtain each evaluation index described to described landslide state of paddling Transition basic credibility m (C_t)；

$m\left(C_t\right)=\underset{x_k\∈\mathrm{\θ}}{\Σ}P\left(x_k\right)---\left(8\right)$

In formula (8),Described n, k are integer；Described t is evaluation index, described θ Set for each evaluation index probability.

Here, it is also possible to calculate the basic probability assignment of exceptional value according to formula (9), the unknown journey to test data is represented Degree:

$m\left({\mathrm{\δ}}_0\right)=1-\underset{t=1}{\overset{n}{\Σ}}m\left(C_t\right)---\left(9\right)$

Further, close the material composition of the formation condition, influence factor, evolution trend and the slip mass that consider landslide, with Time take into account landslide slip sampling, experimental apparatus, the impact of anthropic factor, it is necessary to analyze experimental data set X={x_kReliability, and Obtain deterministic quantized values CRE (E), revise basic credibility m (C further_t), it may be assumed that

m({S_t)=CRE (X) × m (C_t) (10)

Wherein, described CRE (X) is the definitiveness of described X.

Here, it is also possible to calculate { S according to formula (11)_tUnknown degree:

$m\left(\mathrm{\θ}\right)=1-{\mathrm{\Σ}}_{t=1}^{n}CRE\left(X\right)\×m\left(C_t\right)---\left(11\right)$

Further, after basic credibility determines, described 3rd computing unit 46 is for according to described identification framework And the basic credibility on described landslide, utilizing Dempster composition rule calculate described in paddle that landslide transition between states occurs total general Rate；Specifically, Dempster compositional rule is the rule reflecting multiple evidence combined effects, under same identification framework, has The confidence function of several different evidences, Dempster compositional rule just can calculate one by confidence functions based on these evidences Individual final confidence level function.It is to say, in the present embodiment, landslide state is the basic probability assignment that evaluation index draws (basic credibility) BPA₁It is a confidence function m₁(Y)；The basic probability assignment drawn for evaluation index with characteristics of rainfall vector (basic credibility) BPA₂It is a confidence function m₂(Y)；(basic with the basic probability assignment that reservoir level draws for evaluation index Credibility) BPA₃It is a confidence function m₃(Y), then described 3rd computing unit 46 just can calculate institute according to formula (12) Stating paddle landslide transition between states occur total probability:

$m\left(Y\right)=\frac{1}{K}\underset{Y_1\∩Y_2\∩Y_3=Y}{\Σ}{m}_1\left(Y\right)*m_2\left(Y\right)*m_3\left(Y\right)---\left(12\right)$

Wherein, in formula (12), described K is normaliztion constant, and described Y is not empty set.

Further, described normaliztion constant K can calculate according to formula (13):

During actual application, described acquiring unit 41, division unit the 42, first computing unit 43, construction unit 44, second are counted Calculate unit the 45, the 3rd computing unit 46 can by the central processing unit (CPU, Central Processing Unit) of this device, Digital signal processor (DSP, Digtal Signal Processor), programmable logic array (FPGA, Field Programmable Gate Array), micro-control unit (MCU, Micro Controller Unit) realize.

The monitoring device of the landslide transition between states of paddling of the present embodiment offer, in conjunction with particle group optimizing PSO algorithm and BP god Through network, the displacement monitoring data on landslide are carried out feature extraction, use D-S evidence theory that data are merged, draw described The total probability that landslide transition between states of paddling occurs, and then landslide can be carried out accurate early warning.

Embodiment three

The present embodiment, as a example by coming down in plain boiled water river, reservoir area of Three Gorges, utilizes method and embodiment two offer that embodiment one provides Device to paddle landslide transition between states occur total probability calculate, detailed process is as follows:

First five monitoring points are set on landslide, plain boiled water river, obtain Monitoring Data；Wherein it is desired to from Historical Monitoring data Obtaining land slide data and the rainfall data of monitoring point, described land slide data may include that monitoring point displacement and reservoir level.Then also Simulation artificial rainfall experiment need to be obtained and obtain the land slide data in experimentation and rainfall data.

During simulation artificial rainfall experiment, need to increase landslide with the time m-displacement accumulative displacement curve of monitoring point The data of state, according to described landslide status data by the ground on landslide, vegetative coverage situation, landslide displacement, rain fall, storehouse Water level conditions etc. simulate.

Here, the specific practice of simulation rain making is: landslide, plain boiled water river is carried out geologic structure and generally changes and environmental condition Generalization, tests the size of model groove in existing Landslide Model experimental system according to this, determines train length and prototype length The likelihood ratio be 1:150.Field experimentation, chooses an on-the-spot 5*5m²Region be rain area, Rainfall height 4m.Select the east of Sichuan The rainfall simulation system of large-scale field trial field, including portable rain controller, soil moisture and flow of water automatic monitoring system, The instruments such as soil erosion detector, piezometer.Monitoring system transmit after soil moisture content, the flow of water, temperature can be gathered automatically to Terminal.Experiment landslide displacement is reduced to the 0.88% of true landslide displacement, and experiment rain time is reduced to truly drop The 1.81% of rain time, experiment rainfall is reduced to the 0.67% of true rainfall, and experiment reservoir level is reduced to true reservoir level 0.43%.Diverse location on landslide arranges 4 observation sections, and lays 1 surface displacement on each is domatic respectively The change in displacement situation on sensor observation landslide.By to landslide rainfall, the simulation of reservoir level, record landslide accumulative displacement is bent Line.

Here, after getting land slide data, also to set up landslide state according to described land slide data definition landslide state Judgment criterion；Specifically, the Monitoring Data first passing through a large amount of landslides example shows, under gravity, and slope Rock And Soil Deformation evolution curve has three stage evolution features.With reference to Fig. 2, three phases is respectively as follows: landslide state and is divided into initial deformation rank Section, at the uniform velocity deformation stage and acceleration deformation stage.Wherein, the AB section in initial deformation stage, i.e. Fig. 2.Slope, landslide body deformability At the initial stage, along with deformation is grown out of nothing, there is rock fracture in surface, landslide.The slope of this stage deformation curve is relatively large, but The passage of deformation time at any time tends towards stability, and deformation curve slope diminishes, and deformation velocity reduces.

BC section in constant speed deformation stage, i.e. Fig. 2, constant speed deformation stage, on the basis of the initial deformation stage, comes down Deformation velocity is constant, and Landslide Deformation curve is substantially in an angled straight lines, and macroscopic deformation observed result is basically unchanged.

CD section in accelerating deformation stage, i.e. figure.After Landslide Deformation develops into a certain degree, deformation curve oblique Rate.Increase, until before coming down, deformation curve is bordering on and rises steeply.

Then according to the definition of three landslide states, the landslide displacement situation of every month is analyzed, draws quantitative Landslide state demarcation condition.Landslide reality is combined by landslide displacement, displacement acceleration, three aspects of the landslide displacement of last month Situation judges landslide state.

During actual application, according to plain boiled water river in land slide data in July, 2003 to the accumulative displacement of in December, 2008, obtain white The moon displacement of water river every month, deformation velocity, deformation acceleration, using deformation velocity and deformation acceleration as the state of every month Characteristic vector, is denoted as [v, a].The state characteristic vector of 66 months is clustered by utilization K average, can obtain three class results, Its cluster centre is respectively as follows: [14.45,1.32], [68.47 ,-191.6], [263.67,164.8].Cluster result represents respectively The three phases on landslide: initial deformation stage, constant speed deformation stage, acceleration deformation stage.

After getting rainfall data, predetermined period N, with described cycle N, the rainfall number of times in the monitoring time is divided into M Secondary；Add up average daily rainfall r, rainfall natural law d and duration D, constitutive characteristic data set during described M rainfall；Utilize K equal Value-based algorithm carries out K mean cluster to described characteristic data set, determines K class according to similar maximum comparability, foreign peoples's maximum diversity Rain types.

Wherein, utilize K mean algorithm that described characteristic data set is carried out K mean cluster and specifically include: first, set final Cluster kind K；Then, K data point is randomly selected as initial cluster center at described characteristic data set；Furthermore, calculate All data points in addition to initial cluster center and the distance of initial cluster center, and by owning in addition to initial cluster center Data point is classified as that nearest class；And when new data point adds certain class, it is required to recalculate cluster centre；When After cluster centre updates, and calculate all described data point in addition to current cluster centre and between current cluster centre away from From, the most constantly repeat to ask distance to carry out sorting out and update these 2 steps of cluster centre, until square error convergence.Wherein, institute State square error to calculate according to formula (1):

$J=\underset{j=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{n}{\Σ}}\left|\right|x_i^{\left(j\right)}-c_j|{|}^2---\left(1\right)$

Wherein, in formula (1), described J be all data points square error and, described x_iIt is characterized in data set Data point, described c_jCluster centre for jth class.

Here, when choosing initial cluster center, strive for that K initial cluster center can be maximum dissimilar, can with this Improve the accuracy of category division.Meanwhile, choose appropriate initial cluster center and can accelerate convergence of algorithm speed, therefore need The quadratic sum of K initial cluster center distance each other is maximum, on this basis, chooses and meets K the node work making L maximum For initial cluster center.

$L=\underset{j=1}{\overset{k}{\Σ}}\underset{i=1}{\overset{k}{\Σ}}\left|\right|x_i-x_j|{|}^2---\left(2\right)$

During actual application, come down for plain boiled water river, use K mean algorithm that 211 rainfall data are carried out cluster analysis.If Putting cluster numbers K=4, the attribute arranging one rainfall event is: average daily rainfall r, rainfall natural law d, rainfall natural law and continuous days it Compare T.Wherein, average daily rainfall is the ratio of single quantum of rainfall and rainfall natural law.Stop condition is: twice adjacent iteration In, the change of cluster centre is less than threshold value.First concentrate from characteristic at random and choose 4 points as initial cluster center, Then calculate the distance to each initial cluster center of the data point in addition to initial cluster center, and be included into distance for minimum class In.Complete one take turns calculating after, using the meansigma methods of all data of every class as such new cluster centre.Whether judge stop condition Meet, if it is satisfied, then stop iteration, otherwise enter next iteration.

Wherein, due to numerical range and the unit property of there are differences, it is therefore desirable to the index of catchment is carried out pretreatment To guarantee that cluster result has in higher class discrimination between concordance and class.In order to the index of reconstruct catchment is special Levying, effective rainfall natural law d directly uses, and original catchment index is carried out conversion simultaneously and obtains r and T.

Specifically, described average daily rainfall r can calculate according to formula (3):

R=(R/d) * p₁ (3)

Wherein, in formula (3), described R is quantum of rainfall, p₁For coefficient, p in the present embodiment₁It is 0.1.

Described rain time d can calculate according to formula (4):

D=d*p₂ (4)

Wherein, p₂For coefficient, in order to the index feature of reconstruct catchment, effective rainfall natural law d directly uses, p₂ It is taken as 1.

Here, it is also possible to the ratio T, described T that calculate rainfall natural law d and duration D according to formula (5) are used for judging fall Whether rain is interrupted.

T=(d/D) * p₃ (5)

Wherein, p₃For coefficient, the present embodiment value is 11.

In the present embodiment, rain types includes four classes, sees Fig. 3, rain types be respectively as follows: fragmentary rainfall, interrupted rainfall, Heavy rain and rainfall continuously.

After rain types determines, obtain the evaporation capacity under every kind of rain types, infiltration capacity and run-off.

Specifically, when determining evaporation capacity, if the rainfall on the same day is not more than the daily evaporation amount of this month, the then rainfall on the same day The all evaporation capacity of data, net rainfall is 0.

When determining infiltration capacity, double-ring infiltration experiment can be carried out, and combine artificial rainfall experiment and obtain net rainfall and enter The corresponding relation of 16 groups of numerical value of milliosmolarity, as shown in table 1:

Table 1

Net rainfall/mm	Infiltration capacity/mm	Net rainfall/mm	Infiltration capacity/mm
				0	0	40	28.72
5	1.25	45	32.07
				10	3.9	50	32.95
15	7.57	55	32.75
				20	12.62	60	29.77
25	15.61	65	22.01
				30	21.59	70	15.05
35	26.09	75	2.43

Then utilize method of least square that 16 groups of numerical fittings become a function:

Y=-0.0005 × x²+0.0371×x²+0.0510×x+0.1640 (6)

Wherein, described x is net rainfall, and described net rainfall and infiltration capacity corresponding relation see Fig. 5.

After evaporation capacity, described infiltration capacity and described run-off draw, according to particle group optimizing PSO algorithm to described steaming Send out amount, described infiltration capacity and described run-off to calculate, determine characteristics of rainfall vector.Specifically, with the instruction of BP neutral net White silk error, as fitness function, utilizes described PSO algorithm to calculate described evaporation capacity, described infiltration capacity and described run-off to cunning The weight coefficient of slope stability impact；Specifically, one group of evaporation capacity, described is asked in conjunction with PSO algorithm and BP neural network algorithm The weight coefficient of 12 dimensions of infiltration capacity and described run-off.The frequency of training arranging BP neutral net is 30000, and training precision is e^-5, learning rate is 0.05, using the mean square deviation of its training error as the fitness function of PSO.The Population Size of PSO algorithm is set Being 20, inertia coeffeicent is S type, and scope is 1 to drop to 0.4, and Studying factors is 2, the 12 dimensional weight coefficients finally obtained for [0.307, 0.522,0.171,0.371,0.395].

Here, the collection of potential solutions all in problem to be optimized is collectively referred to as " solution space " by particle swarm optimization algorithm PSO, one Potential solution is referred to as one " position ", is designated as Pos.It is one and there is no quality by abstract for every bird, there is no " particle " of size, grain Son with certain " speed " flight, is designated as V, and evaluates the excellent degree of particle position with fitness function in solution space.Logical Cross position and the speed of the more new particle that constantly flies, and chase personal best particle and colony's optimal location seeks optimization problem Optimal solution, is designated as P respectively_ibAnd P_gb.Remember that the speed after the renewal of each particle and position are respectively V_i+1, Pos_i+1, then:

V_i+1=ω * V_i+c₁*rand*(P_ib-Pos_i)+c₂*rand*(P_gb-Pos_i) (7)

Pos_i+1=Pos_i+V_i+1 (8)

Wherein, subscript i represents iterations；ω represents the inertia coeffeicent of particle, and value is a dull reduction of (0,1) Sequence, reflection particle inherits the degree of original speed；Rand represents the random number that a value is (0,1)；c₁And c₂For study because of Son, the degree that reflection particle is drawn close to personal best particle and colony's optimal location.

In units of monthly, utilize described each weight coefficient to the described evaporation capacity under identical rain types, described in infiltrate Amount and described run-off are weighted；By described evaporation capacity, described infiltration capacity and described runoff after weighting under each rain types Amount is separately summed, and obtains the characteristics of rainfall of this month, is designated as α, β, γ respectively；Extract described characteristics of rainfall data, obtain described fall Rain characteristic vector [α, β, γ].

Specifically, come down as a example by August, 2008 by plain boiled water river, have 4 rainfalls this moon, including three kinds of rain types (days The unit of rainfall and characteristics of rainfall parameter is all mm), it respectively is: being fragmentary rainfall for the first time, its daily rainfall is [15,2.4]；Second time is interrupted rainfall, and daily rainfall is [2.1,44.1,3.4,16.4,0,14.1,23.1,0,0,4.8], Third time is fragmentary rainfall, and daily rainfall is [3.1]；4th time is continuous rainfall, daily rainfall be [2.2,11.1,4.8, 17.5,2.5,59.6].When daily rainfall is not more than 4mm, the daily evaporation amount on the same day is equal to daily rainfall, otherwise daily evaporation amount etc. In 4mm.Day net rainfall is the difference of daily rainfall and daily evaporation amount, utilizes formula (6) to ask for a day infiltration capacity.Day net rainfall It is diurnal courses amount with the difference of day infiltration capacity.In August, 2008, the evaporation capacity of continuous rainfall, infiltration capacity, run-off be [20.7, 40.58,36.42]；Discontinuously rainfall [35.1,44.03,28.87]；Fragmentary rainfall [11.7,3.15,5.65].12 Wei Te of this month Levy parameter for [20.7,40.58,36.42,0,0,0,35.1,44.03,28.87,11.7,3.15,5.65].According to aforesaid way Statistics obtains the moon characteristics of rainfall parameter of 2003.8 2008.12.

So every daily rainfall is become characteristic component [α, beta, gamma] monthly, three numerical value by after feature extraction.Number Being only 1/10th of initial data according to amount, and contain more characteristic information, the displacement that can be applied directly to landslide is pre- In survey.

Be multiplied the characteristics of rainfall component after being weighted by the weight coefficient of the 12 characteristics of rainfall parameters tieed up and 12 dimensions, and 2008 The weighted feature component in August in year is: [6.36,21.18,6.23,0,0,0,11.13,10.26,12.99,6.236,0.287, 2.12].Evaporation capacity under 4 class rain fall, infiltration capacity, run-off are separately summed, obtain three-dimensional feature vector, represent with this The characteristics of rainfall that this month is overall.The characteristics of rainfall vector in August, 2003 is [23.72,31.73,21.34].Then, according to permissible Above-mentioned statistical obtains 2003.8 2008.12 characteristics of rainfall vectors monthly.

Further, identification framework is built；When building identification frame framework, according to coming down, each state feature carries out total score Analysis, sets up the decision problem (decision problem of described identification framework) of " whether landslide transits to accelerate deformation state ".Combine Close history physical record and the experimental record on landslide, analyze and determine the principal element of impact landslide transition between states, be evaluation Index.Each evaluation index has respective several result for decision problem, and all judged results being combined just constructs Having gone out identification framework, the subset of described identification framework is proposition.For example, if set Θ is a finite aggregate, and its yuan Element for about the mutual exclusion of a certain problem domain and comprehensively proposition it is assumed that then gathering Θ is called identification framework.Identify and be meant that: For an enquirement, a correct answer can only be distinguished from all possible answer relevant to this enquirement.Identify frame Any subset A of frame Θ is corresponding with the proposition of a problem answers.This proposition is generally described as that " answer of problem is at A In ".

Specifically, when landslide is in initial deformation stage or constant speed deformation stage, slope body deformability is relatively slow, and crack can not made big Impact, be now difficult to come down.But accelerating deformation stage, the rate of deformation of slope body quickly, drastically strengthen by crack, this Time it may happen that landslide.Therefore the decision problem set up is: " whether landslide transits to accelerate deformation state ".

Just there is different principal elements for different landslides, according to geologic feature and the influence factor on specific landslide, combine Close the history on this landslide, determine its impact landslide transition between states main factor to acceleration deformation stage, i.e. evaluation index, basis Evaluation index in embodiment is landslide state, described characteristics of rainfall vector and reservoir level.

Further, for " whether landslide transits to accelerate deformation state " this decision problem, each evaluation index has not With judged result, with described landslide state, described characteristics of rainfall vector and reservoir level for evaluation index judge respectively described in ask Topic, and obtain each judged result；Carry out described each judged result comprehensively, constructing described identification framework.

Here, after identification framework builds, also need to described landslide state, described characteristics of rainfall vectorial and described cunning Reservoir level in the data of slope be evaluation index calculate described in paddle landslide transition between states basic credibility.Specifically, with the moon it is Unit, utilizes statistical data legally constituted authority to count collection { x_kProbability P (the x of each evaluation index described in }_k)；

Specifically, in this example, statistical data legally constituted authority is utilized to count collection { x_kLandslide state in }, characteristics of rainfall vector And the probability P (x of reservoir level evaluation index_k) step is as follows:

First with described landslide state for evaluation index calculating probability P1 (x_k):

By using the decision criteria of landslide state, plain boiled water river during in December, 2008 in July, 2003 to can be added up The landslide state of one monitoring point every month.In like manner, according to this decision criteria, the landslide displacement situation in artificial rainfall experiment is entered Row analyzing and processing, determines which state landslide is in.Being A by initial deformation phased markers, constant speed deformation stage is labeled as B, Accelerate deformation stage and be labeled as C, early stage landslide state S_i-1To current landslide state S_iTransformation be referred to as state changes delta S.According to The definition of state change, can produce and show all of landslide state changes delta S, as shown in table 2.For decision problem " whether landslide transits to accelerate deformation stage ", it is believed that no matter which state early stage landslide is in, transit to accelerate deformation Phased markers is → C.

Table 2

Do well the statistics of changes delta S to the every month in July, 2003 in December, 2008, including actual landslide Landslide state change in state change and artificial rainfall experiment.Draw 9 kinds of respective quantity n of state changes delta S, such as table 3 institute Show.

Table 3

In the case of not considering rainfall and reservoir level, only can judge " whether landslide jumps from early stage landslide state aspect Adjourn acceleration deformation stage ".Wherein, total quantity n (A) when early stage landslide state is in initial acceleration phase A is all early stages State S_i-1For summation during A, can be calculated by formula (9).Total when early stage landslide state is in initial acceleration phase B Quantity n (B) is all preneoplastic state S_i-1For summation during B, can be calculated by formula (10).Early stage landslide state is in Total quantity n (C) during initial acceleration phase C is all preneoplastic state S_i-1For summation during C, can be calculated by formula (11) Go out.

N (A)=n (AA)+n (AB)+n (AC) (9)

N (B)=n (BA)+n (BB)+n (BC) (10)

N (C)=n (CA)+n (CB)+n (CC) (11)

Here, by preneoplastic state S_i-1Transit to accelerate the probability of deformation stage C, it is simply that preneoplastic state S_i-1Transit to accelerate The quantity of deformation stage C is divided by the summation of corresponding landslide state.Such as: as preneoplastic state S_i-1It is in initial deformation stage A Time, the probability P (A → C) transitting to acceleration deformation stage C is the ratio of n (AC) and n (A), can be calculated by formula (12) Go out.The probability being transitted to accelerate deformation stage C by preneoplastic state n (AC) thus can be calculated according to formula (12).With Reason, when can be calculated at the uniform velocity deformation stage B by formula (13), transits to accelerate the probability P (B → C) of deformation stage C；Can Accelerate deformation stage remain the probability P (C → C) of stage C to be calculated by formula (14).

P (A → C)=n (AC)/n (A) (12)

P (B → C)=n (BC)/n (B) (13)

P (C → C)=n (CC)/n (C) (14)

Early stage landslide state S finally calculated_i-1And the transition probability P1 (x of correspondence_k) as shown in table 4:

Table 4

Early stage landslide state Si-1	Transit to state C probability	Do not transit to state C probability
			A	0.0494	0.9506
B	0.4300	0.5700
			C	0.2797	0.7203

Secondly, the probability P 2 being landslide transition between states of paddling described in evaluation index calculating with described characteristics of rainfall vector (x_k):

By K mean algorithm, the characteristics of rainfall vector clusters of 2003.7 to 2008.12 is become four classes.Four Lei Lei centers are for dividing Not Wei [13.95,32,14.34], [11.42,8.16,6.95], [13.87,19.44,11.64], [5.95,9.68,6.12]. Under each class characteristics of rainfall vector, three kinds of landslide state quantity n (ABC) altogether in statistics priori data, accelerate deformation rank Quantity n (C) of section, as shown in table 5.

Table 5

Here, the transition probability under certain class characteristics of rainfall vector, it is simply that accelerate the quantity of deformation stage C and all landslides The ratio of state sum.Such as: under the characteristics of rainfall vector of the first kind, its probability P transitting to accelerate deformation stage C₁(→C) It is the total n of all landslides state₁(ABC) with quantity n accelerating deformation stage C₁(C) ratio, sees formula (15).Successively Analogize, do not consider early stage landslide state and current reservoir level situation, only calculate under four class characteristics of rainfall, transition between states of coming down To the probability of acceleration deformation stage, can calculate under the characteristics of rainfall vector of Equations of The Second Kind according to formula (16), it transits to add The probability P of speed deformation stage C₂(→C)；Can calculate under the characteristics of rainfall vector of the 3rd class according to formula (17), its transition To the probability P accelerating deformation stage C₃(→C)；Can calculate under the characteristics of rainfall vector of the 4th class according to formula (18), its Transit to accelerate the probability P of deformation stage C₄(→C)。

P₁(→ C)=n₁(C)/n₁(ABC) (15)

P₂(→ C)=n₂(C)/n₂(ABC) (16)

P₃(→ C)=n₃(C)/n₃(ABC) (17)

P₄(→ C)=n₄(C)/n₄(ABC) (18)

Finally calculate every kind of transition probability P2 (x corresponding to characteristics of rainfall vector_k) as shown in table 6:

Table 6

Characteristics of rainfall vector type	Transit to state C probability	Do not transit to state C probability
			1	0.9608	0.0392
2	0.4667	0.5333
			3	0.0588	0.9412
4	0	1

Probability P 3 (the x being finally landslide transition between states of paddling described in evaluation index calculating with reservoir level_k):

Specifically, it is analyzed the reservoir level situation of existing in July, 2003 to of in December, 2008 processing.Current storehouse water Position R_iWith early stage reservoir level R_i-1Difference be referred to as Reservoir Water Level Δ R.By cluster kind is carried out substantial amounts of experiment, analysis Operation result.Use K mean algorithm to be clustered by the Reservoir Water Level Δ R of every month, obtain four classes.

The first kind is that reservoir level rises on a small quantity, and Equations of The Second Kind is that reservoir level rises in a large number, and the 3rd class is that reservoir level declines on a small quantity, 4th class is that reservoir level declines in a large number.Cluster centre is: [4.7,10 ,-2].Under each class Reservoir Water Level, statistics reality is sliding A in ramp shaped state and artificial rainfall experiment, B, C tri-kinds come down state quantity n (ABC) altogether, and are in acceleration deformation stage Quantity n (C), as shown in table 7.

Table 7

Here, the transition probability under certain class Reservoir Water Level Δ R, it is simply that accelerate the quantity of deformation stage C and all cunnings The ratio of ramp shaped state sum.Such as: under the Reservoir Water Level Δ R1 of the first kind, its probability P transitting to accelerate deformation stage C_ΔR1 (→ C) is the total n of all landslides state_ΔR1(ABC) with quantity n accelerating deformation stage C_ΔR1(C) ratio, sees formula (19).The like, do not consider early stage landslide state and current characteristics of rainfall vector, only calculate in four class Reservoir Water Level Under, landslide transition between states is to the probability accelerating deformation stage；The Reservoir Water Level at Equations of The Second Kind can be calculated according to formula (20) Under Δ R2, its probability P transitting to accelerate deformation stage C_ΔR2(→C)；The storehouse water in the 3rd class can be calculated according to formula (21) Under the change of position, its probability P transitting to accelerate deformation stage C_ΔR3(→C)；Can calculate in the storehouse of the 4th class according to formula (22) Under SEA LEVEL VARIATION, its probability P transitting to accelerate deformation stage C_ΔR4(→C)。

P_ΔR1(→ C)=n_ΔR1(C)/n_ΔR1(ABC) (19)

P_ΔR2(→ C)=n_ΔR2(C)/n_ΔR2(ABC) (20)

P_ΔR3(→ C)=n_ΔR3(C)/n_ΔR3(ABC) (21)

P_ΔR4(→ C)=n_ΔR4(C)/n_ΔR4(ABC) (22)

Finally calculate every kind of transition probability P3 (x corresponding to Reservoir Water Level_k) as shown in table 8:

Table 8

Reservoir Water Level Δ R	Transit to state C probability	Do not transit to state C probability
			ΔR1	0.9608	0.0392
ΔR2	0.4667	0.5333
			ΔR3	0.0588	0.9412
ΔR4	0	1

When respectively will according to collection { x_kProbability P (the x of each evaluation index described in }_k) calculate after, according to formula (23) will Probability P (x_k) merge, obtain each the evaluation index described basic credibility m (C to described landslide transition between states of paddling_t)；

$m\left(C_t\right)=\underset{x_k\∈\mathrm{\θ}}{\Σ}P\left(x_k\right)---\left(23\right)$

In formula (23),Described n, k are integer；Described t is evaluation index, described θ is the set of each evaluation index probability.

Here, it is also possible to calculate the basic probability assignment of exceptional value according to formula (24), the unknown to test data is represented Degree:

$m\left({\mathrm{\δ}}_0\right)=1-\underset{t=1}{\overset{n}{\Σ}}m\left(C_t\right)---\left(24\right)$

Further, close the material composition of the formation condition, influence factor, evolution trend and the slip mass that consider landslide, with Time take into account landslide slip sampling, experimental apparatus, the impact of anthropic factor, it is necessary to analyze experimental data set X={x_kReliability, and Obtain deterministic quantized values CRE (E), revise basic credibility m (C further_t), it may be assumed that

m({S_t)=CRE (X) × m (C_t) (25)

Wherein, described CRE (X) is the definitiveness of described X.

Here, it is also possible to calculate { S according to formula (26)_tUnknown degree:

$m\left(\mathrm{\θ}\right)=1-{\mathrm{\Σ}}_{t=1}^{n}CRE\left(X\right)\×m\left(C_t\right)---\left(26\right)$

After basic credibility calculates, according to described identification framework and the basic credibility on described landslide, utilize The total probability that landslide transition between states of paddling described in the calculating of Dempster composition rule occurs；Wherein, described decision problem is specially Whether landslide transits to accelerate deformation stage.Specifically, Dempster compositional rule is the method reflecting multiple evidence combined effects Then, under same identification framework, having the confidence function of several different evidence, Dempster compositional rule just can be several based on this The confidence function of individual evidence calculates a final confidence level function.It is to say, in the present embodiment, landslide state is for evaluating Basic probability assignment (basic credibility) BPA that index draws₁It is a confidence function m₁(Y)；With characteristics of rainfall vector for evaluating The basic probability assignment BPA that index draws₂It is a confidence function m₂(Y)；The elementary probability drawn for evaluation index with reservoir level Assign BPA₃It is a confidence function m₃(Y), then landslide transition between states of just can paddling according to formula (27) calculating occurs Total probability:

$m\left(Y\right)=\frac{1}{K}\underset{Y_1\∩Y_2\∩Y_3=Y}{\Σ}{m}_1\left(Y\right)*m_2\left(Y\right)*m_3\left(Y\right)---\left(27\right)$

Wherein, in formula (27), described K is normaliztion constant, and described Y is not empty set.

Further, described normaliztion constant K can calculate according to formula (28):

Specifically, as a example by the Monitoring Data of landslide, plain boiled water river in JIUYUE, 2008, formula (23) is utilized to evaluate to described each The index basic credibility m (C to described landslide transition between states of paddling_t), calculate this month landslide by formula (27) and transit to add The probability of speed deformation stage is 0.9690.

Just can be shown that by above method in JIUYUE, 2008 transits to accelerate the probability of deformation stage.And according to this Method can ask for the probability transitting to accelerate deformation stage of any one month.

Further, the data using 2003.7 to 2007.12, as Monitoring Data, obtain each evaluation index every month Basic probability assignment.It is analyzed the data of 2008.1 to 2008.12 processing, finally carries out D-S evidence theory synthesis, obtain Transition probability, then compare with actual situation, the feasibility of checking context of methods and effectiveness.

Utilize the basic reliability distribution of every month, with reference to the method for D-S evidence theory information fusion, be calculated During 2008.1 to 2008.12 every month come down transition between states to accelerate deformation stage probability as shown in table 10.

Table 10

Date	Transit to accelerate deformation stage C probability	Do not transit to accelerate deformation stage C probability
			In January, 2008	0	1
In February, 2008	0	1
			In March, 2008	0	1
In April, 2008	0	1
			In May, 2008	0	1
In June, 2008	0	1
			In July, 2008	0	1
In August, 2008	0	1
			In JIUYUE, 2008	0.9690	0.0310
In October, 2008	0	1
			In November, 2008	0	1
In December, 2008	0	1

By the probability after the D-S evidence theory information fusion of 2008.1 to 2008.12 in table 10, determine that landslide transits to Accelerate the month of deformation stage C.The most only transitting to accelerate the probability of deformation stage in JIUYUE, 2008 landslide is 0.9690, and the probability of other 11 months is all 0.Simultaneously in actual 2008, the most only JIUYUE is from the beginning of early stage Beginning deformation stage transits to accelerate deformation stage, and actual has landslide to have part slumping.Result shows to use D-S evidence theory to grind Study carefully landslide and transit to accelerate deformation stage be effective

The present embodiment utilizes monitoring method and embodiment two offer of the landslide transition between states of paddling that embodiment one provides Device, for slide area, plain boiled water river data, the displacement prison to landslide in conjunction with particle group optimizing PSO algorithm and BP neutral net Survey data and carry out feature extraction, use D-S evidence theory that data are merged, draw described in landslide transition between states of paddling occur Total probability, and then can accurately early warning.

The above, only presently preferred embodiments of the present invention, be not intended to limit protection scope of the present invention, all Any amendment, equivalent and the improvement etc. made within the spirit and principles in the present invention, should be included in the protection of the present invention Within the scope of.

Claims (10)

1. the monitoring method of a landslide transition between states of paddling, it is characterised in that described method includes:

Obtain land slide data and rainfall data；

K mean algorithm and described rainfall data are utilized to divide rain types；

Obtain the evaporation capacity under every kind of rain types, infiltration capacity and run-off；

According to particle group optimizing PSO algorithm, described evaporation capacity, described infiltration capacity and described run-off are calculated, determine rainfall Characteristic vector；

Build identification framework；

Reservoir level in and described land slide data vectorial with described landslide state, described characteristics of rainfall calculates described for evaluation index Paddle landslide transition between states basic credibility；

According to described identification framework and described basic credibility, utilize landslide state of paddling described in the calculating of Dempster composition rule The total probability that transition occurs.

2. the method for claim 1, it is characterised in that described utilize K mean algorithm and described rainfall data to divide fall Rain type specifically includes:

Predetermined interval cycle N, is divided into the rainfall number of times in the monitoring time M time with described cycle N；

Add up quantum of rainfall, rain time and duration during described M rainfall, constitutive characteristic data set；

Utilize K mean algorithm that described characteristic data set is carried out K mean cluster, determine K class rain types.

3. method as claimed in claim 2, it is characterised in that the described K of utilization mean algorithm carries out K to described characteristic data set Mean cluster specifically includes:

Preset cluster kind K；

K data point is randomly selected as initial cluster center at described characteristic data set；

Calculate the distance between all described data point and the initial cluster center in addition to described initial cluster center, and will remove It is a nearest class that all described data point outside initial cluster center is classified as described distance；

When newly-increased described data point, update cluster centre, and calculate all described data point in addition to current cluster centre with Distance between current cluster centre；Until the square error convergence of all data points.

4. the method for claim 1, it is characterised in that according to particle group optimizing PSO algorithm to described evaporation capacity, described Infiltration capacity and described run-off calculate, and determine that characteristics of rainfall vector specifically includes:

Using the training error of multi-layer feed-forward BP neutral net as fitness function, described PSO algorithm is utilized to calculate described evaporation The weight coefficient that Landslide Stability is affected by amount, described infiltration capacity and described run-off；

Utilize described weight coefficient that the described evaporation capacity under identical rain types, described infiltration capacity and described run-off are added Power；

Described evaporation capacity, described infiltration capacity and described run-off after weighting under each rain types is separately summed, obtains rainfall Feature；

Extract described characteristics of rainfall data, obtain described characteristics of rainfall vector.

5. the method for claim 1, it is characterised in that described structure identification framework specifically includes:

Feature according to each state of described landslide sets up the decision problem of described identification framework；

Described decision problem is judged respectively for evaluation index with described landslide state, described characteristics of rainfall vector and reservoir level, and Obtain each judged result；

Carry out described each judged result comprehensively, constructing described identification framework.

6. method as claimed in claim 5, it is characterised in that described with described landslide state, described characteristics of rainfall vector and Reservoir level in described land slide data be evaluation index calculate described in paddle landslide transition between states basic credibility specifically include:

Statistical data legally constituted authority is utilized to count collection { x_kProbability P (the x of each evaluation index described in }_k)；

Utilize formulaBy the probability P (x of each evaluation index described_k) merge, obtain described each The evaluation index basic credibility m (C to described landslide transition between states of paddling_t)；Wherein, m (C_t) >=0,Described N, k are integer；Described t is evaluation index.

7. method as claimed in claim 6, it is characterised in that as the basic credibility m (C of described landslide transition between states of paddling_t) After calculating, described method also includes:

Utilize formula m ({ S_t)=CRE (X) × m (C_t) to described m (C_t) be modified；Wherein, described CRE (X) is described X's Definitiveness, described X={x_k}。

8. the method for claim 1, it is characterised in that cunning of paddling described in the calculating of the described Dempster of utilization composition rule The total probability that slope transition between states occurs specifically includes:

According to formulaCalculate total probability m (Y)；Wherein, described m₁(Y) according to The first basic probability assignment BPA that described landslide state obtains₁；Described m₂(Y) that according to, described characteristics of rainfall vector obtains Two basic probability assignment BPA₂；Described m₃(Y) the 3rd basic probability assignment BPA that according to, described Reservoir Water Level obtains₃；Institute Stating K is normaliztion constant, and described Y is not empty set.

9. method as claimed in claim 8, it is characterised in that according to formula Calculate K value.

10. the monitoring device of a landslide transition between states of paddling, it is characterised in that described device includes:

Acquiring unit, is used for obtaining land slide data and rainfall data；Obtain the evaporation capacity under every kind of rain types, infiltration capacity and footpath Flow；

Division unit, is used for utilizing K mean algorithm and described rainfall data to divide rain types；

First computing unit, is used for according to particle group optimizing PSO algorithm described evaporation capacity, described infiltration capacity and described run-off Calculate, determine characteristics of rainfall vector；

Construction unit, for according to building identification framework；

Second computing unit, the reservoir level in vectorial with described landslide state, described characteristics of rainfall and described land slide data Basic credibility for landslide shape body transition of paddling described in evaluation index calculating；

3rd computing unit, described 3rd computing unit, for according to described identification framework and described basic credibility, utilizes The total probability that landslide transition between states of paddling described in the calculating of Dempster composition rule occurs.