CN102496077B - Harmful disaster prediction system and method - Google Patents

Harmful disaster prediction system and method Download PDF

Info

Publication number
CN102496077B
CN102496077B CN201110400419.2A CN201110400419A CN102496077B CN 102496077 B CN102496077 B CN 102496077B CN 201110400419 A CN201110400419 A CN 201110400419A CN 102496077 B CN102496077 B CN 102496077B
Authority
CN
China
Prior art keywords
cellular
model
state
harmful
disaster
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201110400419.2A
Other languages
Chinese (zh)
Other versions
CN102496077A (en
Inventor
张晓丽
谢芳毅
王昆
张凝
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Forestry University
Original Assignee
Beijing Forestry University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Forestry University filed Critical Beijing Forestry University
Priority to CN201110400419.2A priority Critical patent/CN102496077B/en
Publication of CN102496077A publication Critical patent/CN102496077A/en
Application granted granted Critical
Publication of CN102496077B publication Critical patent/CN102496077B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A kind of system and method for predicting Harmful disaster, mainly comprise: Harmful hazard prediction model module, it comprises geography cellular automata model, artificial nerve network model and multiple agent model, wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, its state transition rules is obtained by training of human artificial neural networks model, artificial nerve network model be input as disaster factor of influence, export as disaster degree, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, the analysis result of intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading.

Description

Harmful disaster prediction system and method
Technical field
The present invention relates to Harmful hazard prediction field, be specifically related to a kind of Harmful disaster prediction system and method.
Background technology
Biological epidemics, comprise the disasters such as insect, disease, mouse (rabbit) evil, noxious plant, cause tremendous influence and loss to the Sustainable Operation of forest resourceies and sustainable development.Therefore, large quantifier elimination is carried out to the prediction of Harmful disaster both at home and abroad, and propose a lot of Forecasting Methodology and forecast model, comprise discriminatory analysis model, method observed and predicted in principal component analysis (PCA), time series models, statistical model, gray prediction, markoff chain, space regression Forecasting Methodology etc.
There is following subject matter in current Harmful forecast model and method:
Current great majority prediction just rests on numerical prediction basis, carries out simple numerical prediction in conjunction with data such as generating capacity in the past, generation times.But for the prediction of Harmful disaster, not only need the prediction of the aspects such as generating capacity, plague grade, time of origin, more need spatially for the prediction of disaster occurrence and distribution, more comprehensively, three-dimensionally could grasp that Harmful disaster occurs, spreading trend like this, so make accurately, the control of economy, actual effect, control decision.
The kinetic models such as cellular automaton (CA), artificial neural network (ANN), multiple agent (MAS) can realize non-linear space prediction.But the simple CA model that uses also also exists limitation.The neighborhood rule of cellular only considered neighborhood cellular around cellular to its impact, but ignores the impact of other Macroscopic Factors; The state transition rules of cellular is determined in cellular space very much, but in real world, the otherness of the rule domain of the existence that things occurs; The position of cellular and state are too static, do not consider that dynamic, variable, unordered factor of influence is on the impact of cellular; The timing definition of cellular is equally spaced, is stiff, inadequate on the impact reaction of the time scale caused by objective condition.There is limitation equally in ANN model.Find under study for action, the precision of ANN model is very large by the impact of sample, and its training sample must summarize the various situations in reality.If sample to contain face wide, so can not simulate or dope in sample the state not have appearance.The training of ANN model is difficult to grasp, and the criterion that setting parameter neither one is set, needs continuous debugging and verification, just can decide, and its rule training is because of the difference of its algorithm selected, and training time and training effect be institute's difference also.This needs in the training process of network, strengthens the supervision to network from every side, and its reason place of serious analysis.And the error convergence of learning rate to network training is extremely important.Learning rate is too little, then restrain too slow, learning rate is too large, then may cause paralysis and the instability of network.Suitable learning rate can ensure training effectiveness, can ensure again the stable of network.Needing design adaptive learning rate, is the good way addressed this problem.The key of MAS model how to define intelligent body type in complication system and rule of conduct thereof.In model, intelligent body type is too much, and whole system is very complicated, causes computing velocity very slow, crosses the complicacy not embodying again system at least.But the factor affecting forest disease and pest generation and diffusion is intricate, except natural cause impact, also by the impact being difficult to the social factor estimated.Such as, the various natural enemies of host, the activity (felling diseased wood and infection wood, transport etc.) of the mankind, road network and village distribution etc., if consider habitat and the uncertain factor such as mankind's activity, natural enemy, MAS system will be very complicated, cause forecasting process long, even paralyse.
Summary of the invention
According to an aspect of the present invention, provide a kind of system for predicting Harmful disaster, comprise: Harmful hazard prediction model module, it comprises geography cellular automata model, artificial nerve network model and multiple agent model, wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, the state transition rules of geography cellular automata is in order to obtain the cellular state of cellular subsequent time according to the current cellular state of cellular and neighborhood thereof and cellular attribute, and state transition rules is obtained by training of human artificial neural networks model, wherein, artificial nerve network model be input as disaster factor of influence, export as disaster degree, and wherein, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, wherein, the analysis result of Mankind action intelligent body and harmful organism Plant population change intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading, parameter setting module, for arranging the parameter of geography cellular automata model, artificial nerve network model and multiple agent model, training module, for training artificial nerve network model, thus obtains state transition rules, data input module, for providing the value of current cellular state and one or more cellular attribute, and for cellular state from training of human artificial neural networks to training module that be provided for and cellular property value to geography cellular automata, data outputting module, for providing the output data of geography cellular automata model, as predicting the outcome.
According to another aspect of the present invention, provide a kind of method for predicting Harmful disaster, comprise: set up Harmful hazard prediction model, it comprises geography cellular automata model, artificial nerve network model and multiple agent model, wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, the state transition rules of geography cellular automata is in order to obtain the cellular state of cellular subsequent time according to the current cellular state of cellular and neighborhood thereof and cellular attribute, and state transition rules is obtained by training of human artificial neural networks model, wherein, artificial nerve network model be input as disaster factor of influence, export as disaster degree, and wherein, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, wherein, the analysis result of Mankind action intelligent body and harmful organism Plant population change intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading, for arranging the parameter of geography cellular automata model, artificial nerve network model and multiple agent model, artificial nerve network model is trained, thus obtains state transition rules, the value of current cellular state and one or more cellular attribute is provided to geography cellular automata, and is provided for cellular state and the cellular property value of training of human artificial neural networks, there is provided the output data of geography cellular automata model, as predicting the outcome.
Accompanying drawing explanation
Fig. 1 schematically shows Technology Roadmap of the present invention;
Fig. 2 shows the composition schematic diagram of cellular automaton;
Fig. 3 shows three kinds of stress and strain model of two dimensional cellular automaton;
Fig. 4 shows neuronic structural representation;
Fig. 5 shows the basic structure of BP neural network;
Fig. 6 shows the uml class structural drawing of geography cellular automata;
Fig. 7 shows the uml class structural drawing of BP artificial neural network;
Fig. 8 shows according to an embodiment of the invention for predicting the system of Harmful disaster.
Embodiment
The factor of influence that the present invention occurs Harmful disaster, comprehensively extract from aspects such as biological characteristics, Harmful living environment and disaster genesis mechanism, and cellular Automation Model, artificial nerve network model, multiple agent model are combined, space angle have effectively achieved the simulation that Harmful disaster occurs.
Realize the spatial prediction of Harmful disaster, can make predicts the outcome implements to plot, the hilltop, improves the practicality and operability that predict the outcome, has important practical value.
The present invention expands and has deepened the applied research of cellular Automation Model on Harmful hazard prediction.This is mainly reflected in: by the method for disaster simulation prediction and model, be expanded to the angle in space; The cellular connotation of Extended cellular automata, simulates the situation that Harmful disaster occurs more exactly; The neighborhood type of Extended cellular automata, defines different neighborhood types for different Harmful disaster, obtains the status information of periphery cellular more exactly; Using the mode that artificial neural network obtains as cellular automaton transformation rule, avoid the subjective factor in transformation rule definition procedure and uncertain factor.
The present invention is by remote sensing technology (RS), Geographic Information System (GIS), cellular automaton (CellularAutomata, CA), artificial neural network (Artificial Neural network, and multiple agent (Multiagent system ANN), MAS) technology combines, research embodies the non-linear space forecast model of disaster generation Diffusion Law, has important scientific meaning.
GIS and remote sensing technology, the fixed point that can occur disaster, timing and dynamic comprehensive information carry out integrated analysis, there is provided the information such as change in time and space scope and the extent of injury of disaster, this is the research of Harmful disaster genesis mechanism, the foundation of forecast model provides indispensable technical support; The research of Harmful disaster, analyzes from the angle of entomological population ecology, can be combined with each other with geography model, with the generation solved spatially with spread problem.Wherein CA model is applicable to simulation geographical complexity system very much, it allows atural object oneself state and information around to take into full account, and with the form opening extended mode evolution rule and other models couplings, can using the part of spatial information as disaster occurrence factor.CA model is used for the prediction of Harmful disaster space, needs to define most crucial part---state transition rules, how to act on the occurrence and harm degree of disaster to describe the disaster factor that makes a difference.But Forest Pest Species is various, and biological characteristics is different, and its disaster pests occurrence rule factor of influence is different.Need to select a kind of Distributed Parallel Computing Model, in order to solve the nonlinear problem in transformation rule definition procedure.ANN model is compared with other models, and maximum advantage is to train voluntarily according to sample data, develop create-rule rule, and easily combines with other models.This point for Harmful disaster Occurrence forecast, can perfect, improve by the pests occurrence rule extensively approved, can inquire into again, find new pests occurrence rule.MAS model is the emerging model of the one be based upon on multi-agent Technology, across the evolution process of yardstick, from bottom to up the simulating forest ecosystem, can very be suitable for analysis and the dynamic similation of disaster spreading.In addition, the biological characteristics of Harmful and pests occurrence rule difference, the factor of a model chosen is also different.Therefore, need, on the basis of its general character of research, based on the model selecting highly versatile, to build Harmful hazard prediction model.
Traditional Forecasting Methodology can not accurately, the region that occurs of fast prediction disaster, this is to Harmful diaster prevention and control, no matter be the guidance in policy, or in the enforcement of operability, all causes certain difficulty.The solution of the present invention, by spatial simulation forecast model, be combined with each other with multiple technologies means, can grasp the trend that Harmful disaster occurs as soon as possible, effectively formulate prophylactico-therapeutic measures more in detail, effectively control.
The comprehensive CA of the present invention, MAS, ANN and remote sensing, it is non-linear that Spatial Data Analysis is set up, Intellectualized space forecast model, can from biological characteristics, Harmful living environment, the impact of human activity and other uncertain factor, the aspects such as disaster genesis mechanism are carried out comprehensively, quick and precisely extract the factor of influence that Harmful disaster occurs, solve in disaster generation and the forecasting problem under spreading the indefinite situation of mechanism, space angle realizes Harmful disaster effectively occur and the prediction of diffusion and dynamic similation, set up new Intellectualized space Forecast and simulation technical system, support and auxiliary Harmful diaster prevention and control decision-making, expand and in-depth CA, ANN, the applied research of MAS in the simulation of Harmful hazard prediction.Achievement in research will have important scientific value and apply meaning.
The present invention is by cellular automaton (CA), artificial neural network (ANN), multiple agent (MAS), GIS spacial analytical method combines with disaster genesis mechanism, build the ANN-CA-MAS non-linear space forecast model based on Harmful disaster pests occurrence rule, the connotation of each component of analytical model in Harmful hazard prediction, the spatial and temporal scales of model, development prototype system, application remote sensing image, statistical data and ground investigation data, realize Harmful disaster generation area, space distribution, the spatial prediction of extent of injury and the dynamic similation of diffusion process.Particular content is as follows:
(1) model is set up.By extracting the factor of influence that Harmful disaster occurs, in conjunction with the principle of cellular automaton, artificial neural network and multiple agent, the connotation Sum fanction of cellular Automation Model and multiple agent model is expanded, set up based on cellular automaton, artificial neural network and multiple agent Harmful hazard prediction model (hereinafter referred to as: ANN-CA-MAS Forest Pest).
(2) exploitation realizes.Follow the principle of object based programming, encapsulation ANN-CA-MAS ForestPest model, and exploitation realizes prototype system on this basis.
(3) hazard prediction.The ANN-CA-MAS Forest Pest built is utilized to predict disaster.Exemplarily, select dendrolimus punctatus and fall webworms to be disaster generation research object respectively, set up and predict solution accordingly, utilize ANN-CA-MAS Forest Pest to predict its disaster.
The technology used in the present invention means comprise following several:
(1) GIS and remote sensing.GIS is as the effective means of data acquisition, process, spatial analysis and management and instrument, data in studying, comprise spatial data and attribute data, management is got up effectively, and for the realization of model provide data input, the function such as data export, data exhibiting and data analysis.Realization for model on remote sensing technology Time and place provides raster data support, uses the multiple method of Remote Sensing Image Processing Technology, fully can extract the terrestrial object information that remotely-sensed data comprises, thus enrich the data source of model.
(2) cellular automaton (Cellular Automata, CA): the feature occurred in conjunction with Harmful disaster, the expansion of connotation Sum fanction is carried out on hazard prediction to cellular Automation Model, comprises definition cellular state, cellular neighborhood rule, cellular state transformation rule.
(3) multiple agent (MAS): the feature occurred in conjunction with Harmful disaster, defines intelligent body behavioural characteristic and describe, determine intelligent body type and rule of conduct.
(4) artificial neural network (Artificial Neural Network, ANN).Utilize the ability of artificial neural network process high complexity nonlinear system, by sample training, obtain Harmful disaster pests occurrence rule to predict that it occurs.The rule that artificial neural network obtains, especially for the indefinite Harmful disaster of some pests occurrence rules, can replace complicated cellular automaton definition status transformation rule.
(5) Object-Oriented Design (Object Oriented Design, OOD).Follow the basic thought of object-oriented progreamming method: encapsulation and extensibility, be packaged into object by each several part of model or its operation, strengthen transplantability and the extensibility of model.
Fig. 1 schematically shows Technology Roadmap of the present invention.
One, geography cellular automata (Geo-CA) model
1 cellular automaton
1) key concept
Cellular automaton (Cellular Automata, CA) is defined in cellular with the cellular composition of discrete, finite state spatially, according to certain local rule, and the dynamical system that discrete time dimension develops.But it is different from general kinetic model.Cellular automaton does not conceptually have clear and definite equation form, and depicts the primitive rule of a series of Construction of A Model.Say from this angle, cellular automaton is that a class meets the overall of the model of common rule, or perhaps the method frame of a model construction.(Zhou Chenghu, 1999)
Cellular automaton can describe by various ways, describes can be expressed as a quaternary ancestral with formal language:
CA＝(L d，S，N，f) (2-1)
Wherein, L represents the mesh space of a regular partition, and its elementary cell is then cellular; D is the dimension of L, and mesh space can be the rule space of any positive integer dimension; S represents a discrete finite aggregate, is used for representing the state s of each cellular; N represents the Neighbourhood set that cellular number is n, and its any Neighbourhood set is all the subset of mesh space.Like this, N can be expressed as the set of cellular in all neighborhoods, namely comprises a space vector of n different cellular state, is designated as:
N＝(s 1，s 2，s 3，...s n)，s i∈Z，i∈(1，...，n) (2-2)
F represents a mapping function: namely according to the incompatible state value determining t+1 this cellular of moment of the state group of all neighbours of certain cellular of t, f is usually also referred to as transfer function or evolution rule.
2) basic comprising
What cellular automaton was the most basic is configured with four parts: cellular, cellular space, neighborhood rule, transformation rule and time.Fig. 2 shows the composition schematic diagram of cellular automaton.
(1) cellular (Cell).According to certain rule, being arranged in the elementary cell in the mesh space of regular partition, is then cellular.The state value value of cellular is in a finite state set, and the state in each cellular certain period can only be one in this finite state set.
(2) cellular space (Lattice).The space networks point set that cellular distributes, is called as cellular space.In cellular space, cellular is regularly arranged according to certain rule, adjacent one another are.Because of arrangement mode and the dimension that produces is different, can be called that N ties up cellular automaton.In geographical space, generally adopt two dimensional cellular automaton.
Fig. 3 shows three kinds of stress and strain model of two dimensional cellular automaton.
(3) neighborhood rule (Neighbor).In cellular Spatial distributions change procedure, the next state of cellular is determined by its original state and its neighborhood cellular state.Which cellular neighborhood rule specifies clearly belongs to center cellular, and which cellular belongs to neighborhood cellular.Neighborhood rule formulates more complicated, and usually, in geographical space, for the cellular space of the regular cubic stress and strain model of modal two dimension, neighborhood rule can have following a few class:
Feng A. Nuo Yiman (Von.Nuemann) type: adjacent four cellulars in upper and lower, left and right of a cellular are the neighbours of this cellular.Here, neighbours' radius r is 1, is equivalent to four neighborhoods in image processing, four directions.
B mole of (Moore) type: the upper and lower, left and right of a cellular, upper left, upper right, bottom right, adjacent eight cellulars in lower-left are the neighbours of this cellular.Neighbours' radius r is similarly 1, be equivalent to the eight neighborhood in image processing, from all directions to.
Mole (Moore) type of C expansion: above neighbours' radius r is expanded to 2 or larger, namely obtain mole type neighbours of so-called expansion.
The formulation of neighborhood rule, has his own strong points in various cellular Automation Model, but all first according to the arrangement mode of cellular in cellular space, and then will formulate suitable neighborhood rule according to concrete research contents and object.
From the formulation of neighborhood, a cellular depends in the state in certain moment and only depends on the state of this cellular surrounding neighbors cellular, and its state updating of cellular in cellular space is synchronous, and whole cellular space shows as and changes on discrete time dimension.
(4) transformation rule (Rule)
Transformation rule is a kinetic function, and it is described that how according to the current state of cellular and the state of this cellular of neighborhood states determination subsequent time thereof, is a state transition function.This is the core place of cellular automaton, is also the point of penetration with other dynamic evolution models couplings.
(5) time (Time)
Cellular automaton is as a dynamic system, and its change on time dimension is discrete, and namely time t is a round values, and equidistant continuously.If time interval dt=1, if t=0 is the preliminary examination moment, so t=1 is its subsequent time.In transfer function, a cellular is directly decided by this cellular and the domain neighborhood property thereof of t in the state in t+1 moment, although in the state of the cellular in t-1 moment and the state remote effect of neighborhood cellular thereof cellular in the t+1 moment.
The state participating in each cellular of dynamic evolution changed along with the time, and its state changes according to the rule of a local.That is, neighborhood answers the concrete definition of cellular automaton and different, but is all defined in the subrange of space.Such as, it is four neighborhoods or eight neighborhood that the neighborhood rule of the general two dimensional cellular automaton based on grid can be formulated.
3) essential characteristic
Cellular Automation Model has following 5 essential characteristics:
(1) cellular be distributed according to certain regular partition discrete cellular spatially;
(2) differentiation of system is carried out according to constant duration substep, and time variable gets the moment point of unique step;
(3) each cellular has clear and definite state, and the state of cellular can only get limited discrete value;
(4) state value that cellular subsequent time develops determined by the transformation rule determined.
(5) transformation rule of each cellular only determined by the cellular state in local neighborhood.
It can be seen, the constraint condition of cellular automaton is non-constant width pine, and this makes cellular automaton can formulate oneself specific rules and content easily neatly according to different field, and combines with other models.
2 geography cellular automata
Geography cellular automata (Geo Cellular Automata is called for short GeoCA) is the thematic model of cellular automaton in geographical complicated phenomenon research application, and it is substantially the cellular Automation Model of an expansion.Geography cellular automata is under the framework of cellular automaton, to its comprehensive expansion in geographical connotation, and integrated multiple Theories and methods, builds the model framework that of Simulation and analysis geographical space complication system is conceptual and deemed-to-satisfy4.This model framework will provide the support in Theories and methods meaning from now in the application in field of science particularly for cellular automaton.
In GeoCA, geographical cellular state is a polytomy variable normally, comprises corresponding geographical connotation.Adopt Object--oriented method to encapsulate cellular and state thereof, the geographical entity becoming and there is certain " intelligent " can be further expanded.General, cellular space corresponds to the geographical space under Cartesian coordinates system, and neighborhood space corresponds to spatial neighbors relation.Transformation rule is the core place of cellular automaton, and it has embodied a concentrated reflection of the interaction of self entity in cellular space, neighborhood entity and various envirment factor.This interaction has been endowed different connotations according to different application, thus makes geography cellular automata expand to various specific geography cellular automata.
Two, BP artificial neural network
1 key concept
BP (Back Propagation) algorithm is to utilize the error of output layer to estimate the error of the directly front conducting shell of output layer, recycles the error of the more front one deck of this estimation of error.So just can obtain the estimation of error of every other each layer, so just define the error shown by output terminal to transmit contrary direction step by step to the process that the input end of network transmits along with input signal, backpropagation algorithm of thus gaining the name.The multistage acyclic network using BP algorithm to carry out learning also just is called BP network.
2 basic comprisings
(1) neuron (Neuron)
Neuron is the most elementary cell of neural network, it is by simulation biological neuron, accept one group and carry out other neuronic input signals in automatic network, all signals are weighted to activation (Activation) state of rear decision self, and provide suitable output.Fig. 4 shows neuronic structural representation.Wherein, x 1~ x nfor each component of input vector, ω 1~ ω nfor the weights of each cynapse of neuron, f is transport function, is generally nonlinear function, and t is that neuron exports.With the accumulative effect representing the input signal that this neuron obtains, have:
net = Σ x i ω i - - - ( 2 - 3 )
It can be seen, neuronic function is after the inner product of trying to achieve input vector and weight vector, obtains a scalar result through a nonlinear transfer function.
In BP neural network, the activation function that the BP neuron as its elementary cell uses must can be led everywhere.
(2) topology of networks
The basic structure of BP neural network generally comprises three layers: input layer, a hidden layer (also claiming middle layer) and an output layer, i represents the neuron number of input layer, and j represents the neuron number of hidden layer, and k represents the neuron number of output layer.Neuron on every one deck all realizes entirely being connected by weights and each neuron on adjacent layer, and its basic structure as shown in Figure 5.
3 training process general introductions
In neural network, its training process is in fact the connection between neuron is weighed to the process adjusted according to sample set.And BP neural network performs is have tutor to train, the vector as such in (input vector, desirable output vector) of its sample set shape is to forming.These vectors are right, can be to gather in actual motion system to get, and also can be the results of neural network actual motion.
The training of BP neural network is divided into two stages: forward direction stage and back-propagation stage.The forward direction stage is divided into two steps:
(1) from sample set, get a sample (X p, Y p), by X pinput network;
(2) corresponding actual output O is calculated p.
In the process, sample inputs from input layer as information vector, by transmission conversion step by step, is finally delivered to output layer, so, have
O p＝F n(...F 2(F 1(X pW (1))W (2))...)W (n)) (2-4)
The back-propagation stage is also divided into two steps:
(1) actual output O is calculated py is exported with corresponding ideal perror;
(2) weight matrix is adjusted by the mode of minimization error.
Here, the error metric of single P sample can be expressed as:
E p = 1 2 Σ j = 1 m ( y pj - o pj ) 2 - - - ( 2 - 5 )
And the error metric of whole sample set is in network:
E = Σ E p - - - ( 2 - 6 )
4 rudimentary algorithms
The learning process of network can be divided into two parts: forward-propagating and backpropagation.In forward-propagating process, the neuronic state of every one deck only affects lower one deck neuroid.If output layer can not obtain desired output, the error namely between real output value and desired output not within error tolerance range, then needs to proceed to back-propagation process.In the process, error signal returns along original interface channel, by revising the neuronic weights of each layer, successively propagating to input layer and calculating in the past.Then, forward-propagating process is again entered.Forward-propagating and these two processes of backpropagation are repeatedly used in whole training process, by interative computation, error signal are reduced to minimum.
As can be seen from BP algorithmic procedure, BP network training needs to carry out under tutor's vector instructs, and its study is based upon on the basis of gradient descent method.
According to structure and the training process of BP neural network, can basic BP algorithm be designed as follows.
For sample set
S＝{(X 1，Y 1)，(X 2，Y 2)，..，(X s，Y s)} (2-7)
Network is according to (X 1, Y 1) calculate actual output O 1with error metric E 1, to W (1), W (2)..., W (n)respectively do and once adjust; Then according to (X 2, Y 2) calculate actual output O 2with error metric E 2, to W (1), W (2)..., W (n)second time adjustment is done in distribution ... circulation like this is gone down, until according to (X s, Y s) calculate actual output O swith error metric E s, to W (1), W (2)..., W (n)do the s time adjustment.So just, to a circular treatment of each sample in sample set.After such circulation terminates, need the summation of the error metric calculating whole sample set, and compare with the requirement of system, if the requirement of discontented pedal system, then need to continue, until meet, namely
&Sigma; E p < &epsiv; - - - ( 2 - 8 )
Three, multiple agent
Multi-agent system (Multi-Agent-System) is the developing direction that in field of artificial intelligence research is brand-new. to the research of multiple agent from the nineties in last century, its theoretical foundation is traditional artificial intelligence (AI, Artificial Intelligence), distributed AC servo system (DC, Distributed Control) and Distributed Calculation (DC, Distributed Computer).This term of multiple agent had both represented a structural system, was also the thought of a kind of programming newly after object based programming (OOP).
Agent has abundant intension, and its Chinese name word has " main body ", " intelligent body ", " procurator " or " node " etc., in different discipline background, have different implications.Therefore Agent does not have the clear and definite definition of a unification, and different researchists gives the different structure of Agent, content and ability in the system of oneself, to facilitate the further investigation of oneself specific direction.Table 3-1 lists the general property of intelligent body, for a concrete multi-agent system, might not have all properties listed in Table.It has been generally acknowledged that, an attribute that Agent is the most basic should comprise front four attribute shown in following table: autonomy, reactivity, social, initiative, and then has other attributes according to the actual conditions of application.And the Agent that definition only has front four attribute is weak Agent, the Agent that can also have other attributes is strong Agent.
Agent not only possesses self problem solving ability and performance-based objective, and can mutually cooperate, and reaches common overall goals.Like this, multi-agent system be just defined as by multiple can the system that forms of mutually mutual Agent computing unit.
Multi-agent system adopts from bottom idea about modeling from bottom to top, is not identical with traditional modeling approach from top to down.Its core feed back by the circulation between the local detail model of reaction individual configurations function and the overall situation shows and corrected, and studies the global behavior how variations in detail locally gives prominence to complexity.
Multi-agent system just can construct the system model with labyrinth and function according to the response rule of the system local detail studying a question required, Agent and the behavior of various local.Although microscopic individual behavior wherein may be fairly simple, the global behavior caused by reciprocation between microscopic individual may be very complicated.In multi-agent system, the global behavior that the behavior of microscopic individual and reciprocation show emerges in a non-linear fashion.The combination of individual behavior decides global behavior, says it on the contrary, and global behavior determines again the environment that individuality carries out decision-making.
The attribute of table 1 intelligent body
Many intelligence system plays important role in modern computer science and application thereof.Modern computing platform and computing environment are not only open and heterogeneous, and be large-scale distributed, computing machine is no longer the system of an independent operating, and the close ties between computing machine, between computing machine and user make computing machine and information handling system become increasingly complex.Traditional centralized computation schema can not adapt to the requirement of large-scale distributed information handling system, and multi-agent system is that Distributed Calculation provides a very convenient and effective platform.
The foundation of four ANN-CA Forest-Pest models of the present invention
1 Harmful disaster makes a difference the factor
Only exemplarily, Harmful disaster is selected to occur in China frequent, endanger serious region, select dendrolimus punctatus (Dendrolimus punctatus), pine wood nematode (Bursaphelenchusxylophilus), anoplophora glabripennis (Anop10phora glabripennis), the Main Forest harmful organisms such as fall webworms (Hlyphantriacunea) are model test object, collect interrelated data, specify the biology of research object, the ecological mechanism that ecological characteristics and disaster occur, comprise harmful bio distribution, growth and breeding, perch, land occupation condition, stand structure, meteorologic factor, artificial interference etc., specify space-time development rule and factor of influence that China is mainly harmful to biological epidemics generation and diffusion, for the structure of model framework provides foundation.
1) biological characteristics
Biological characteristics refers to biological exterior representations, the character such as comprise form, distribution, growth and breeding, perch.General, the biological characteristics of Harmful comprises its history of life, life habit, morphological feature, distribution and Characteristics of Damage, route of transmission and natural enemy etc.The biological characteristics of harmful organism determines its time of origin, duration, generation area, harm object, annidation endangered, spreads mode, prevention and controls etc., having extremely important meaning to the monitoring of its disaster, prediction and prevention, is also the original starting point that forecast model is set up.
2) land occupation condition
Land occupation condition comprises: height above sea level, the gradient, slope aspect, position, slope etc.Land occupation condition to a certain extent, determines the envirment factors such as the temperature of the trees that are injured, daylighting, wind direction, humidity.Different Harmfuls, because of the difference of its biological characteristics, the provincial characteristics of its adaptation distribution, growth, procreation is also not quite similar.Such as, dendrolimus sibiricus is mainly distributed in height above sea level less than 700 meters, the leeward In The Low Mountain-hill Region faced south.
3) stand structure
Stand structure comprises: border etc. in standing forest composition, standing forest level, the age of stand, canopy density, woods.The woods are on the one hand for harmful organism provides food source, and another aspect is again for which providing habitat.The composition, level etc. of standing forest have impact on the food chain structure of forest ecological environment and ecological autogenous control ability.General, the standing forest of single structure, ground cover rareness, parasitic sufficient, thus easily become disaster generation base.Border in canopy density or woods, can affect the daylighting in harmful organism habitat, temperature, humidity etc., thus affect the Growth and reproduction of harmful organism.
4) meteorologic factor
Meteorologic factor comprises: temperature, humidity, quantity of precipitation, sunshine number etc., and some compound meteorological factors.The history of life of harmful organism, the differentiation of form are closely related with meteorology.Because of the regional disparity of weather, the annidation of harmful organism limits its distribution, and its growth, development and fecundity all must carry out under the temperature adapted to, damp condition.Temperature, humidity too high or too low, even can make harmful organism lifetime or population quantity decline.Such as typhoon, this seasonal climate characteristic of heavy rain also can affect the extent of injury of harmful organism.
5) time
According to the division of time length, for the prediction of harmful organism disaster, short-term forecasting, medium-and long-term forecasting and long-term forecasting can be divided into.The hazard prediction of different harmful organism, has its distinctive time limit.In addition, research in the past shows, the outburst of harmful organism disaster has certain periodicity.To the assurance of the time scale of prediction, be conducive to harmful organism diaster prevention and control of preparing for deployment targetedly.
2 basic comprisings
ANN-CA-MAS Forest Pest Harmful hazard prediction model is made up of three parts: the geography cellular automata of expansion, multiple agent and BP artificial neural network.
1) geography cellular automata expanded
According to ultimate principle and the formation of cellular Automation Model, and its primitive rule expanded in geographical connotation, on Harmful hazard prediction, connotation expansion is carried out to geography cellular automata.
(1) cellular.Cellular corresponds to the forest resourceies entity in survey region.Cellular state, according to the generation state of Harmful disaster and plague grade, is divided into: do not occur, occur, health, negligible risk, moderate endanger, severe endangers 6 kinds of states.Wherein:
Do not occur: refer to the forest resourceies entity corresponding to cellular position, it does not comprise current Harmful and takes food object, or does not comprise the atural object with vegetation information, as: water body, exposed soil, rock etc.
Occur: refer to the forest resourceies corresponding to cellular position, its Harmful population density (insect density, infestation index, capture rate) reaches certain hazard rating.
Healthy: refer to the forest resourceies corresponding to cellular position, its Harmful population density (insect density, infestation index, capture rate) does not reach certain hazard rating.
Negligible risk: refer to the forest resourceies corresponding to cellular position, the hazard rating that its Harmful population density (insect density, infestation index, capture rate) reaches is slight.
Moderate endangers: refer to the forest resourceies corresponding to cellular position, the hazard rating that its Harmful population density (insect density, infestation index, capture rate) reaches is moderate.
Severe endangers: refer to the forest resourceies corresponding to cellular position, the hazard rating that its Harmful population density (insect density, infestation index, capture rate) reaches is severe.
(2) cellular space.For keeping general with the raster data of GIS and remotely-sensed data in data, the general form adopting square net.The definition space of cellular, needs the content of consideration three aspects: the space scale of Harmful hazard prediction, the spatial accuracy of Harmful hazard prediction and the travelling speed of model.
(3) neighborhood rule.The selection of neighborhood type, different because of the difference of harmful organism disaster pests occurrence rule, need establish according to harmful organism disaster pests occurrence rule.The radius of neighbourhood will according to harmful organism disaster generation state, grade and the interaction closing on standing forest patch, and its rate of propagation, scope and direction are determined.
(4) transformation rule.Expressed by transformation rule, be affect the various factor pair disaster generation state of harmful organism disaster generation and the common effectiveness of grade.For the harmful organism disaster that pests occurrence rule is clear and definite, (Multi Criteria Decision can be judged based on such as multiple criteria, MCD), the method for gray scale prediction (Gray Decision, GD) formulates its corresponding transformation rule.In an embodiment of the present invention, the formulation of transformation rule completes by the training of BP artificial nerve network model.
(5) time rule.Jointly determined by the time scale of the generation data of Harmful disaster occurrence frequency and input.
Cellular Automation Model is in the embody rule of harmful organism hazard prediction, and each ingredient all must be closely connected its biological characteristics and specializing, and could obtain the specifying information of various factor of influence exactly, improve precision of prediction.
2) BP artificial neural network
Artificial nerve network model has good fault-tolerance, is very applicable to rule indefinite harmful organism disaster Occurrence forecast.BP artificial neural network needs according to concrete prediction object and requires concrete structure.BP artificial neural network generally adopts three-tier architecture: input layer, hidden layer and output layer.
(1) input layer: the corresponding training sample of neuron of input, each neuronic input, as x 1, x 2..., x n, to make a difference the factor corresponding to each disaster.
(2) hidden layer: be generally 1 ~ 2 layer.Neuron number is identical with input layer data.
(3) output layer: training stage, the expectation value of the corresponding training sample of neuron of output.The simulation and forecast stage, the analog computation result of the corresponding cellular state of neuron of output.Output neuron can have multiple expectation value to export.But do not advise the computational burden therefore increasing network.
(4) excitation function: provide two kinds of S type functions: BipolarSigmoid function and Sigmoid function.
Wherein, Sigmoid function formula is:
f ( net ) = 1 1 + e - net - - - ( 2 - 9 )
BipolarSigmoid function formula is:
f ( net ) = 2 1 + e - net - 1 - - - ( 2 - 10 )
3) multiple agent
(1) type of intelligent body
Intelligent body quantity in model is too much unsuitable, and computing velocity too much will be caused very slow, and practicality is not strong.The intelligent body of two types is devised altogether in this model.One is artificial interference intelligent body, is an abstract intelligence body; Two is harmful organism Plant population change intelligent bodies.
(2) intelligent body rule of conduct
A Mankind action intelligent body
Very crucial effect is played in the generation development of human activity to disease and pest, therefore Mankind action factor is encapsulated in an intelligent body.Artifical influence factor is very numerous and diverse, such as: policy guidance, foster the conservation and utilization etc. of felling, trapping communication media, vegetation transport, biological control, reinforcement quarantine, natural enemy.Consider these influence factors, existing effect of being reacted man's activity by two large indexs.One is the existing technical merit in operation side, comprises the academic level of operation side personnel and existing prevention and elimination of disease and pests equipment amount level.Two is operation side's inputs in the prevention and control of plant diseases, pest control, comprises personnel and drops into, learns input, goods and materials input.Personnel drop into can grow prevention and control of plant diseases, pest control team; Study drops into and can improve thought level, the technical merit of personnel, thus more efficiently, scientifically carry out the preventing and controlling of disease and pest; Goods and materials drop into the input of the medicine that refers to anti-worm of curing the disease and equipment etc., and biological pesticide, high-new control equipment prevent and treat dynamics by what greatly strengthen disease and pest.
The computing method of index:
The side of operation personnel educational background level=undergraduate education and above academic number/manage area
The existing prevention and elimination of disease and pests equipment amount level=existing equipment amount in operation side is worth/manages area
The side of operation personnel input=period personnel drop into/(managing the area * time)
Operation side learns the study of input=period and drops into/(managing the area * time)
Manage square object money input=period goods and materials to drop into/(managing the area * time)
Give a mark to each index obtained, each index has five grades, corresponding score 1 ~ 5 point, and grade interval rule of thumb divides, and desired value is higher, and higher grade (first-class in second-class), score is lower.Finally, each index can obtain a mark, is added the mark that obtains divided by 5, if the mark obtained is 1, then thinks to manage in region and disease and pest does not occur, be in health status; The mark obtained is between 1 ~ 2, then think to manage in region disease and pest occurs; The mark obtained is between 2 ~ 3, then think that it is slight for managing disease and pest hazard rating in region; The mark obtained is between 3 ~ 4, then think that managing disease and pest hazard rating in region is moderate; The mark obtained is between 4 ~ 5, then think that managing disease and pest hazard rating in region is severe.To each cell assignment of raster data, in each operation region, all cell assignment are identical, and it is 1 that assignment does not occur, and it is 2 that assignment occurs, and slight assignment is 3, and moderate assignment is 4, and severe assignment is 5.
B harmful organism Plant population change intelligent body
Nature and society exist the phenomenon of a large amount of s type changes, logistic Logistic model is almost the unique mathematical model describing the growth of s type, this is a continuous print, monotonically increasing, be upper asymptotic s type curve with parameter k, its pace of change increases slower at the beginning, interlude growth rate is accelerated, later growth recession and tending towards stability, utilize it can characterize the Number dynamics of population, describe the propagation process of a certain research object, also can be used as the theoretical foundation of other complex models.
Logistic equation is encapsulated as an intelligent body in this model, and the space as very important factor stimulation disease and pest develops.
The logistic differential equation of population growth is:
dN/dt＝rN[(K-N)/K]
Wherein N is population at individual quantity, i.e. Population Size; T is time variable; R is the instantaneous rate of increase of population; K is carrying capacity of the environment, also claims Carrying capacity.By above formula distortion and integration, the integration type that can obtain logistic equation is:
N t＝k/(1+e a-rt)
Finally, judge the disease and pest incidence of certain time point according to the population quantity obtained, be respectively: do not occur, occur, slight, moderate, severe, according to incidence to the grid assignment in survey region, be followed successively by 1,2,3,4,5.
3 basic processes
ANN-CA-MAS Forest Pest model comprises three elements: the geography cellular automata of expansion, multiple agent and BP artificial neural network, wherein, geography cellular automata is organized data, obtain, comprise the neighborhood information etc. of the state of cellular and attribute, cellular, artificial interference and harmful organism species information process through processing by multiple agent, thus obtain reformed cellular state.BP artificial neural network calculates to data, comprises the training of rule, the execution of acquisition and rule, results in and transfer to the state of geography cellular automata to cellular to upgrade again.
According to the geography cellular automata expanded, multiple agent and the basic process of BP artificial neural network and the feature of triplicity, the basic process of ANN-CA-MAS Forest Pest model can be divided into two parts: the sampling of (1) cellular is merged with intelligent body sample information and carries out the training of cellular transformation rule, thus obtains cellular transformation rule; (2) state simulation of the next time in cellular space.
In cellular sampling with cellular transformation rule training process, what first will carry out is cellular sampling, in cellular space, namely chooses the representative cellular of some, and extract corresponding sample information.Sample information comprises input information and output information two parts.Input information comprises state, cellular attribute, the neighborhood information of cellular t time, and output information then comprises the state of cellular t+1 time.Intelligent body information acquisition obtains.What then will carry out is networking rule training.First the cellular state in input information and intelligent body analysis result are merged, obtain new cellular state, then input amendment (comprising new cellular state, attribute, neighborhood information) is input in the input layer of BP artificial neural network, output sample is input in output layer, can network training be carried out after corresponding training parameter is set, train successfully and can obtain cellular transformation rule.
The concrete grammar that cellular state in described input information and intelligent body analysis result merge can be averaging reclassification after both being added according to certain weight.Such as, if weight is 1: 1, the state of the cellular of the same area is negligible risk, and the state that intelligent body draws is severe harm, then the result merged will endanger for moderate.The different extents of injury can represent with different numerical value, can conveniently calculate like this.Weight can be arranged according to the actual conditions of different regions by user, also a weight first arbitrarily can be set, then carry out network training, obtain a training rules by historical data, then judge that whether this weight is suitable, if improper, correspondingly to adjust.Such as can obtain a training rules by the data of 02 year and 04 year, then judge that whether weight is arranged suitable, if improper, uses other weights instead by the data of 04 year and 06 year.
Cellular transformation rule obtains successfully, travels through the cellular in whole cellular space, one by one cellular new state, cellular attribute and neighborhood information after fusion is input to computing in BP artificial neural network, and the result of calculation returned is the cellular state in the future of simulation.
Various geographical spatial datas needed for model running, need to carry out necessary pretreatment operation, ensure the consistent of coordinate system, projection information, coordinate figure and spatial resolution.In model running process, need the data-interface of GIS module and the support of graphical user interface.
4 class formation designs
In one particular embodiment of the present invention, follow the principle of Object-Oriented Design (OOD), according to ultimate principle and the method for cellular Automation Model, multiple agent model and BP artificial nerve network model, with C#.NET 2.0 language, three is encapsulated.Be to be noted that this packaged type is only exemplary, instead of limitation of the present invention.
1) geography cellular automata
: the first, all cellulars of cellular space unified management comprise its location, acquisition, setting and the acquisition to neighborhood to the main thought that geographical cellular Automation Model encapsulates; The second, for cellular self, by harmful organism disaster generation state encapsulation to status attribute, the factor that harmful organism disaster made a difference is packaged into factor of the habitat attribute; 3rd, store the attribute information such as state, factor of the habitat of all cellulars in cellular space with data Layer, read operation can be carried out to data Layer in cellular and cellular space; Finally, individual packages operation that cellular neighborhood information is obtained.
According to above thinking, as follows to geographical cellular Automation Model encapsulation: cellular class (Cell), cellular spatial class (Lattice), neighborhood class (Neighbor), data Layer class (DataLayer), data Layer container class (DataLayerContainer) etc.
The uml class structural drawing of geography cellular automata as shown in Figure 6.Following table 2 lists the class declaration of geography cellular automata, and table 3 lists generic attribute and the method for geography cellular automata.
Table 2 geography cellular automata class declaration
Table 3 geography cellular automata generic attribute and method
2) BP artificial neural network
To the main thought of BP artificial nerve network model encapsulation be: the first, according to the basic comprising of artificial neural network, its component part encapsulated; The second, the data-interface of network and inner computing management are packaged into BP neural network, reduce the complicacy of operation; 3rd, utilize network layer to manage neuron; 4th, neuron calculates; Finally, activation function individual packages, is inherited by unified function interface, is convenient to the definition of multiple activation function.
According to above thinking, as follows to the encapsulation of BP artificial nerve network model: neuron base class (Neuron), to activate neural metaclass (ActivationNeuron), network base class (Network), activation network class (ActivationNetwork), BP learning process class (BackPropagationLearning), activation network network layers class (ActivationLayer), S type function class (SigmoidFunction, BipolarSigmoidFunction) and activation function interface (IActivationFunction).
The uml class structural drawing of BP artificial neural network as shown in Figure 7.Following table 4 lists the class declaration of BP neural network, and table 5 lists BP neural network generic attribute and method.
Table 4BP neural network class declaration
Table 2-4BP neural network generic attribute and method
3) intelligent body
A Mankind action intelligent body
Agent is defined as follows:
B harmful organism Plant population change intelligent body
Agent is defined as follows:
The result of intelligent body feedback is transferred to cellular space layer, and the status attribute in cellular is carried out alternately.
Five according to the system for predicting Harmful disaster of the embodiment of the present invention
As those skilled in the art illustrates known according to above, the present invention in fact provides a kind of system for predicting Harmful disaster and corresponding method.Therefore, see above, the more detailed understanding obtained system for predicting Harmful disaster of the present invention and corresponding method can be described.
In one aspect of the invention, provide a kind of according to an embodiment of the invention for predicting the system of Harmful disaster.As shown in Figure 8, this is used for predicting that the system 800 of Harmful disaster comprises: Harmful hazard prediction model module (ANN-CA-MAS Forest Pest) 801, it comprises geography cellular automata model, artificial nerve network model and multiple agent model, wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, the state transition rules of geography cellular automata is in order to obtain the cellular state of cellular subsequent time according to the current cellular state of cellular and neighborhood thereof and cellular attribute, and state transition rules is obtained by training of human artificial neural networks model, wherein, artificial nerve network model be input as disaster factor of influence, export as disaster degree, and wherein, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, wherein, the analysis result of Mankind action intelligent body and harmful organism Plant population change intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading, parameter setting module 802, for arranging the parameter of geography cellular automata model, artificial nerve network model and multiple agent model, training module 803, for training artificial nerve network model, thus obtains state transition rules, data input module 804, for providing the value of current cellular state and one or more cellular attribute, and for cellular state from training of human artificial neural networks to training module that be provided for and cellular property value to geography cellular automata, data outputting module 805, for providing the output data of geography cellular automata model, as predicting the outcome.
According to a further embodiment of the invention, what described cellular attribute was selected from following classification is one or more: biological characteristics; Land occupation condition; Stand structure; Meteorologic factor.
According to a further embodiment of the invention, what described biological characteristics comprised the form of harmful organism, distribution, growth and breeding and perched in characteristic is one or more; It is one or more that described land occupation condition comprises in height above sea level, the gradient, slope aspect and position, slope; It is one or more that described stand structure comprises in standing forest composition, standing forest level, the age of stand, canopy density, woods in border; It is one or more that described meteorologic factor comprises in temperature, temperature, quantity of precipitation and sunshine number.
According to a further embodiment of the invention, described cellular state comprises at least two in the following: do not occur, and occurs, healthy, negligible risk, and moderate endangers; Severe endangers.
According to a further embodiment of the invention, the cellular space in described geography cellular automata model is square mesh, and cellular corresponds to one or more remote sensing grid.
According to a further embodiment of the invention, described artificial nerve network model is back propagation artificial neural network model.
According to a further embodiment of the invention, the influence factor of described Mankind action intelligent body comprises the existing technical merit in operation side and the operation side input in the prevention and control of plant diseases, pest control.
According to a further embodiment of the invention, described harmful organism Plant population change intelligent body uses the dynamic evolution process of logistic model representation harmful organism population.
According to a further embodiment of the invention, described parameter setting module is for arranging at least one in the following parameter: the time step in geography cellular automata model, cellular size, neighborhood, cellular attribute; The training parameter of artificial nerve network model; Operation side's technical merit parameter of Mankind action intelligent body and prevention and control of plant diseases, pest control input parameter, and the population growth parameter of harmful organism Plant population change intelligent body.
According to a further embodiment of the invention, the data that described data input module provides come from remotely-sensed data.
The six roots of sensation is according to the method for predicting Harmful disaster of the embodiment of the present invention
In another aspect of the present invention, provide a kind of method for predicting Harmful disaster, the method comprises: set up Harmful hazard prediction model, it comprises geography cellular automata model, artificial nerve network model and multiple agent model, wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, the state transition rules of geography cellular automata is in order to obtain the cellular state of cellular subsequent time according to the current cellular state of cellular and neighborhood thereof and cellular attribute, and state transition rules is obtained by training of human artificial neural networks model, wherein, artificial nerve network model be input as disaster factor of influence, export as disaster degree, and wherein, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, wherein, the analysis result of Mankind action intelligent body and harmful organism Plant population change intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading, the parameter of geography cellular automata model, artificial nerve network model and multiple agent model is set, artificial nerve network model is trained, thus obtains state transition rules, the value of current cellular state and one or more cellular attribute is provided to geography cellular automata, and is provided for cellular state and the cellular property value of training of human artificial neural networks, there is provided the output data of geography cellular automata model, as predicting the outcome.
According to a further embodiment of the invention, what described cellular attribute was selected from following classification is one or more: biological characteristics; Land occupation condition; Stand structure; Meteorologic factor.
According to a further embodiment of the invention, what described biological characteristics comprised the form of harmful organism, distribution, growth and breeding and perched in characteristic is one or more; It is one or more that described land occupation condition comprises in height above sea level, the gradient, slope aspect and position, slope; It is one or more that described stand structure comprises in standing forest composition, standing forest level, the age of stand, canopy density, woods in border; It is one or more that described meteorologic factor comprises in temperature, temperature, quantity of precipitation and sunshine number.
According to a further embodiment of the invention, described cellular state comprises at least two in the following: do not occur, and occurs, healthy, negligible risk, and moderate endangers; Severe endangers.
According to a further embodiment of the invention, wherein, the cellular space in described geography cellular automata model is square mesh, and cellular corresponds to one or more remote sensing grid.
According to a further embodiment of the invention, described artificial nerve network model is back propagation artificial neural network model.
According to a further embodiment of the invention, the influence factor of described Mankind action intelligent body comprises the existing technical merit in operation side and the operation side input in the prevention and control of plant diseases, pest control.
According to a further embodiment of the invention, described harmful organism Plant population change intelligent body uses the dynamic evolution process of logistic model representation harmful organism population.
According to a further embodiment of the invention, described parameters comprises at least one that arrange in the following parameter: the time step in geography cellular automata model, cellular size, neighborhood, cellular attribute; The training parameter of artificial nerve network model; Operation side's technical merit parameter of Mankind action intelligent body and prevention and control of plant diseases, pest control input parameter, and the population growth parameter of harmful organism Plant population change intelligent body.
According to a further embodiment of the invention, the cellular state provided described in and cellular property value come from remotely-sensed data.
The present invention has following characteristic and innovation:
(1) in theory, establish ANN-CA-MAS model, extend CA, MAS connotation in Harmful hazard prediction.Than forecast model in the past, there is intellectuality, nonlinear characteristic.
(2) in method, make full use of CA, ANN, MAS, RS and GIS advantage separately, in conjunction with biology and the Ecological Characteristics of harmful organism, the multiple factors such as comprehensive topography and geomorphology, stand structure, meteorological factor, artificial interference and harmful organism life cycle, realize the prediction under different condition, make it to occur with Harmful disaster and the rule of development adapts.By cellular space, traditional numerical prediction is risen to spatial prediction, predicts the outcome and have more decision value.Obtain CA cellular state transformation rule by ANN, the impact that the life cycle that application activity intelligent body simulates artificial interference and harmful organism spreads disaster, the limitation of a certain model of simple application can be overcome, static prediction is developed into performance prediction and simulation.
(3) technical, application Object--oriented method, carries out class formation design, makes it both to reflect that Harmful disaster occurs, with the complex mechanism of diffusion, to have again clear and definite ecological connotation.Encapsulation CA model, ANN and MAS model, and realize under GIS platform, enhance reusability and the extensibility of model.

Claims (20)

1., for predicting a system for Harmful disaster, comprising:
Harmful hazard prediction model module, it comprises geography cellular automata model, artificial nerve network model and multiple agent model,
Wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, the state transition rules of geography cellular automata is in order to obtain the cellular state of cellular subsequent time according to the current cellular state of cellular and neighborhood thereof and cellular attribute, and state transition rules is obtained by training of human artificial neural networks model
Wherein, artificial nerve network model be input as disaster factor of influence, export as disaster degree,
And wherein, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, wherein, the analysis result of Mankind action intelligent body and harmful organism Plant population change intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading;
Parameter setting module, for arranging the parameter of geography cellular automata model, artificial nerve network model and multiple agent model;
Training module, for training artificial nerve network model, thus obtains state transition rules;
Data input module, for providing current cellular state and cellular attribute data to geography cellular automata, and for be provided for training module artificial neural network training cellular state and cellular attribute data;
Data outputting module, for providing the output data of geography cellular automata model, as predicting the outcome.
2. system according to claim 1, it is one or more that wherein said cellular attribute is selected from following classification:
Biological characteristics;
Land occupation condition;
Stand structure;
Meteorologic factor.
3. system according to claim 2, wherein,
It is one or more that described biological characteristics comprises the form of harmful organism, distribution, growth and breeding and perches in characteristic;
It is one or more that described land occupation condition comprises in height above sea level, the gradient, slope aspect and position, slope;
It is one or more that described stand structure comprises in standing forest composition, standing forest level, the age of stand, canopy density, woods in border;
It is one or more that described meteorologic factor comprises in temperature, temperature, quantity of precipitation and sunshine number.
4. system according to claim 1, wherein, described cellular state comprises at least two in the following: do not occur, and occurs, healthy, negligible risk, and moderate endangers; Severe endangers.
5. system according to claim 1, wherein, the cellular space in described geography cellular automata model is square mesh, and cellular corresponds to one or more remote sensing raster data.
6. system according to claim 1, wherein, described artificial nerve network model is back propagation artificial neural network model.
7. system according to claim 1, wherein, the influence factor of described Mankind action intelligent body comprises the existing technical merit in operation side and the operation side input in the prevention and control of plant diseases, pest control.
8. system according to claim 1, wherein, described harmful organism Plant population change intelligent body uses the dynamic evolution process of logistic model representation harmful organism population.
9. system according to claim 1, wherein, described parameter setting module is for arranging at least one in the following parameter:
Time step in geography cellular automata model, cellular size, neighborhood, cellular attribute;
The training parameter of artificial nerve network model;
Operation side's technical merit parameter of Mankind action intelligent body and prevention and control of plant diseases, pest control input parameter, and the population growth parameter of harmful organism Plant population change intelligent body.
10. system according to claim 1, the data that wherein said data input module provides are remotely-sensed data.
11. 1 kinds, for predicting the method for Harmful disaster, comprising:
Set up Harmful hazard prediction model, it comprises geography cellular automata model, artificial nerve network model and multiple agent model,
Wherein, cellular in geography cellular automata model represents geographic area, and the cellular state had for representing disaster degree and the one or more cellular attributes for representing disaster factor of influence, the state transition rules of geography cellular automata is in order to obtain the cellular state of cellular subsequent time according to the current cellular state of cellular and neighborhood thereof and cellular attribute, and state transition rules is obtained by training of human artificial neural networks model
Wherein, artificial nerve network model be input as disaster factor of influence, export as disaster degree,
And wherein, multiple agent model comprises Mankind action intelligent body, for representing the impact of mankind's activity on cellular state, and harmful organism Plant population change intelligent body, for representing the dynamic evolution process of harmful organism population, wherein, the analysis result of Mankind action intelligent body and harmful organism Plant population change intelligent body will merge with current cellular state, thus obtain the current cellular state after upgrading;
The parameter of geography cellular automata model, artificial nerve network model and multiple agent model is set;
Artificial nerve network model is trained, thus obtains state transition rules;
There is provided current cellular state and cellular attribute to geography cellular automata, and be provided for cellular state and the cellular property value of training of human artificial neural networks;
There is provided the output data of geography cellular automata model, as predicting the outcome.
12. methods according to claim 11, it is one or more that wherein said cellular attribute is selected from following classification:
Biological characteristics;
Land occupation condition;
Stand structure;
Meteorologic factor.
13. methods according to claim 12, wherein,
It is one or more that described biological characteristics comprises the form of harmful organism, distribution, growth and breeding and perches in characteristic;
It is one or more that described land occupation condition comprises in height above sea level, the gradient, slope aspect and position, slope;
It is one or more that described stand structure comprises in standing forest composition, standing forest level, the age of stand, canopy density, woods in border;
It is one or more that described meteorologic factor comprises in temperature, temperature, quantity of precipitation and sunshine number.
14. methods according to claim 11, wherein, described cellular state comprises at least two in the following: do not occur, and occurs, healthy, negligible risk, and moderate endangers; Severe endangers.
15. methods according to claim 11, wherein, the cellular space in described geography cellular automata model is square mesh, and cellular corresponds to one or more remote sensing grid.
16. methods according to claim 11, wherein, described artificial nerve network model is back propagation artificial neural network model.
17. methods according to claim 11, wherein, the influence factor of described Mankind action intelligent body comprises the existing technical merit in operation side and the operation side input in the prevention and control of plant diseases, pest control.
18. methods according to claim 11, wherein, described harmful organism Plant population change intelligent body uses the dynamic evolution process of logistic model representation harmful organism population.
19. methods according to claim 11, wherein, described parameters comprises at least one that arrange in the following parameter:
Time step in geography cellular automata model, cellular size, neighborhood, cellular attribute;
The training parameter of artificial nerve network model;
Operation side's technical merit parameter of Mankind action intelligent body and prevention and control of plant diseases, pest control input parameter, and the population growth parameter of harmful organism Plant population change intelligent body.
20. methods according to claim 11, the wherein said cellular state that provides and cellular property value derive from remotely-sensed data.
CN201110400419.2A 2011-12-06 2011-12-06 Harmful disaster prediction system and method Active CN102496077B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201110400419.2A CN102496077B (en) 2011-12-06 2011-12-06 Harmful disaster prediction system and method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201110400419.2A CN102496077B (en) 2011-12-06 2011-12-06 Harmful disaster prediction system and method

Publications (2)

Publication Number Publication Date
CN102496077A CN102496077A (en) 2012-06-13
CN102496077B true CN102496077B (en) 2015-08-19

Family

ID=46187902

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201110400419.2A Active CN102496077B (en) 2011-12-06 2011-12-06 Harmful disaster prediction system and method

Country Status (1)

Country Link
CN (1) CN102496077B (en)

Families Citing this family (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102982392B (en) * 2012-11-07 2016-06-08 中国科学院亚热带农业生态研究所 A kind of agriculture plague of rats based on GIS-Geographic Information System breaks out risk index evaluation method
CN104915535B (en) * 2015-02-04 2019-04-30 湖南农业大学 Biotic population dynamic prediction analyzes global general-use key factor preset group platform
CN105930769A (en) * 2016-04-13 2016-09-07 北京林业大学 Bio-ecological characteristic-based pest and disease damage occurrence grade precision verification method
CN108595812B (en) * 2018-04-17 2022-02-15 重庆邮电大学 Rapid generation modeling simulation method based on agent vegetation distribution
CN109063319B (en) * 2018-07-27 2023-04-07 天津大学 Simulation method of biological ecosystem based on neural network
CN109584953B (en) * 2018-11-09 2022-12-23 韶关学院 Cellular automaton-based biological evolution method
CN111401590A (en) * 2018-12-14 2020-07-10 中国航天系统工程有限公司 Prediction method based on multi-Agent forest ecosystem
CN110427685A (en) * 2019-07-29 2019-11-08 中国水利水电科学研究院 A kind of animal habitat ground model building method
US11631475B2 (en) 2020-05-26 2023-04-18 Ecoation Innovative Solutions Inc. Real-time projections and estimated distributions of agricultural pests, diseases, and biocontrol agents
CN112116143B (en) * 2020-09-14 2023-06-13 贵州大学 Forest pest occurrence probability calculation processing method based on neural network
CN112465204A (en) * 2020-11-20 2021-03-09 北京林业大学 County scale forest pest and disease damage prediction method based on spatial heterogeneous characteristics
CN112801419B (en) * 2021-03-17 2021-08-27 广东技术师范大学 Rat damage degree grade prediction method and device based on adaptive receptive field SSD
CN113051314A (en) * 2021-03-25 2021-06-29 东北林业大学 Pine wood nematode propagation factor determination device and method based on data acquisition and analysis
CN113515896B (en) * 2021-08-06 2022-08-09 红云红河烟草(集团)有限责任公司 Data missing value filling method for real-time cigarette acquisition
CN114676568B (en) * 2022-01-17 2022-09-23 中国地质大学（北京） Regional geological structure evolution method and device based on cellular automaton
CN116955335A (en) * 2023-07-21 2023-10-27 北京国信达数据技术有限公司 Address data management method and system based on big data model algorithm

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
基于多智能体的土地利用模拟与规划模型;刘小平 等;《地理学报》;20061031;第61卷(第10期);第1101-1106页 *
基于多智能体的森林病虫害蔓延模拟;裴小节;《中国硕士论文全文数据库》;20100806;全文 *
林业有害生物灾害预测模型设计与实现;谢芳毅;《中国硕士论文全文数据库》;20090911;第10-33页 *

Also Published As

Publication number Publication date
CN102496077A (en) 2012-06-13

Similar Documents

Publication Publication Date Title
CN102496077B (en) Harmful disaster prediction system and method
Tang et al. Agent‐based modeling of animal movement: a review
Hare et al. Further towards a taxonomy of agent-based simulation models in environmental management
Jones et al. Simulation of biological processes
Engelen et al. Integrating constrained cellular automata models, GIS and decision support tools for urban planning and policy-making
McLane et al. The role of agent-based models in wildlife ecology and management
Lopez et al. Integrating simulation of architectural development and source–sink behaviour of peach trees by incorporating Markov chains and physiological organ function submodels into L-PEACH
Taillandier et al. A new BDI agent architecture based on the belief theory. Application to the modelling of cropping plan decision-making
Jayashree et al. Application of fuzzy cognitive maps in precision agriculture: a case study on coconut yield management of southern India’s Malabar region
Marilleau et al. Coupling agent-based with equation-based models to study spatially explicit megapopulation dynamics
Cockburn et al. Simulating social and economic specialization in small-scale agricultural societies
CN108762084A (en) Irrigation system of rice field based on fuzzy control decision and method
Perez et al. Landscape-level simulation of forest insect disturbance: Coupling swarm intelligent agents with GIS-based cellular automata model
Cuevas et al. An optimization algorithm for multimodal functions inspired by collective animal behavior
Reynolds et al. Mining the social fabric of archaic urban centers with cultural algorithms
King et al. Bioeconomic models of crop production systems: design, development, and use
David et al. Digital twins for cyber-biophysical systems: challenges and lessons learned
Beiranvand et al. A systematic review of optimization of dams reservoir operation using the meta-heuristic algorithms
CN110222133A (en) A kind of multiclass land use simulating and predicting method and device considering landscape index
Qu et al. Simulation of carbon-based model for virtual plants as complex adaptive system
Stoyanov et al. A personal assistant supporting agriculture operators
Muetzelfeldt et al. Ecological modelling of agroforestry systems.
Khan et al. Eagle perching optimizer for the online solution of constrained optimization
Skobelev et al. Further advances in models and methods for digital twins of plants
Demetriou et al. LACONISS: a planning support system for land consolidation

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant