CN109492793A - A kind of dynamic grey Fil Haast neural network landslide deformation prediction method - Google Patents

A kind of dynamic grey Fil Haast neural network landslide deformation prediction method Download PDF

Info

Publication number
CN109492793A
CN109492793A CN201811150026.9A CN201811150026A CN109492793A CN 109492793 A CN109492793 A CN 109492793A CN 201811150026 A CN201811150026 A CN 201811150026A CN 109492793 A CN109492793 A CN 109492793A
Authority
CN
China
Prior art keywords
neural network
sequence
data
grey
value
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.)
Granted
Application number
CN201811150026.9A
Other languages
Chinese (zh)
Other versions
CN109492793B (en
Inventor
邓洪高
姚鹏远
孙希延
纪元法
王守华
符强
严素清
吴孙勇
付文涛
赵松克
李有明
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.)
Guilin University of Electronic Technology
Original Assignee
Guilin University of Electronic Technology
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 Guilin University of Electronic Technology filed Critical Guilin University of Electronic Technology
Priority to CN201811150026.9A priority Critical patent/CN109492793B/en
Publication of CN109492793A publication Critical patent/CN109492793A/en
Application granted granted Critical
Publication of CN109492793B publication Critical patent/CN109492793B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Economics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Tourism & Hospitality (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a kind of dynamic grey Fil Haast neural network landslide deformation prediction methods, it is related to landslide monitoring deformation prediction technical field, the technical issues of solution, is to provide a kind of landslide deformation prediction method with high accuracy, comprising the following steps: establishes slip mass accumulating displacement initial data；To data prediction；Data are fitted by Grey Markov chain predicting model；It calculates residual error and constructs residual sequence；Training GA-BP neural network；Obtain prediction residual sequence；Calculate Combined model forecast value.The present invention is greatly improved to landslide deformation prediction precision.

Description

A kind of dynamic grey Fil Haast neural network landslide deformation prediction method
Technical field
The present invention relates to landslide monitoring deformation prediction technical field more particularly to a kind of dynamic grey Fil Haast nerves Network landslide deformation prediction method.
Background technique
Geological disaster is one of the main problem that mankind nowadays society faces, and China is by the geology calamity that causes naturally and manually Evil mainly has earthquake, slope rock and soil body's displacement, ground deformation and land deterioration etc..It is counted according to Department of Land Resources, 2015 complete All kinds of geological disasters 8224, which occur, for Kuomintang-Communist rises, wherein 5616, landslide, accounting is up to 68.3%.Landslide disaster oneself become China One of most important geological disaster, and active and effectively prevent and treat landslide etc. the important measures of geological disasters first is that landslide etc. shapes Become hidden danger point and carry out long-term observation, is sounded an alarm before landslide disaster generation, redundant labor transfer is reminded, to reduce and reduce damage It loses.If cannot monitor and effectively forecast in time, landslide disaster then may bring tremendous economic to damage once occurring to local resident It loses, or even endangers resident's life security.Therefore, carrying out forecast before landslide disaster generation is particularly important, and forecasts step Key be using the high prediction technique of accuracy.
The prediction on landslide includes time prediction and spatial prediction, and wherein landslide forecasting model mainly directly determines cunning The forecasting model and deformation prediction model of slope unstability time, the Comprehensive Evaluation forecast of Combined model forecast result and side slope shape Method etc. in terms of the judgement of change stage and prediction criterion for temporary prediction of landslide.The forecasting model for directly determining the landslide failure time is usually basis Model directly calculates the model of landslide failure time, as vegetarian rattan model, Su Aijun model, Yue Qilun model, good fortune are limited (Fukuzono) model, Voight model, golden section predicted method etc..Most of in these models is all using creep theory as base Plinth is predicted that deficiency is that precision of prediction is subject to certain restrictions by establishing creep empirical equation.Deformation prediction model is first Go out slip mass deformation quantity later for a period of time according to model prediction, is then slided in conjunction with prediction criterion for temporary prediction of landslide appropriate Slope prediction.Deformation prediction model includes the types such as Statistical Prediction Model and Nonlinear Prediction Models.Statistical Prediction Model has grey Prediction model, biological growth model, regression analysis model, time series models, exponential smoothing etc..Wherein grey forecasting model Grey GM (1,1) model and grey Fil Haast (Verhulst) model can be divided into again.Grey GM (1,1) model is suitable for Sequence with stronger exponential law can only describe dull change procedure, for nonmonotonic swing developmental sequence or have full " S " type sequence prediction error of sum is larger.Grey Markov chain predicting model is mainly used to the process that description has saturation state, i.e., " S " type process, and the displacement time curve for the dynamic evolution that comes down just meets this process.Nonlinear Prediction Models have chaos pre- Survey model, Fractal Geometry Model, Catastrophic Theory Model, neural network model etc..It is wherein more in neural network model A kind of model is BP neural network model.The model uses error backpropagation algorithm, and theory of algorithm is the most intuitive.
Above-mentioned prior art model still remains some problems, as the Grey Markov chain predicting model of the prior art uses Stale information makes prediction result error larger；BP neural network algorithm exists in terms of network initial weight and threshold value selection asks Topic, causes prediction result precision not high enough.
Summary of the invention
In view of the deficiencies of the prior art, it is high to be to provide a kind of landslide deformation prediction precision for technical problem solved by the invention Method.
In order to solve the above technical problems, the technical solution adopted by the present invention is that a kind of dynamic grey Fil Haast nerve net Network landslide deformation prediction method, comprising the following steps:
(1) slip mass accumulating displacement initial data is established, a GNSS monitoring net is established in landslide area, passes through GPRS Either three-dimensional coordinate data is back to server by 3G 4G network, and data are stored to database；
(2) to data prediction, three-dimensional coordinate data is read from database and carries out pretreatment operation；The data are pre- Processing includes Kalman filtering smoothing processing and rejects the outlier and exceptional value that Kalman filtering can not filter out using 3 σ criterion；
(3) data are fitted by Grey Markov chain predicting model, specifically as follows step by step:
(1) deformation quantity data sequence is constructed, the accumulation position of slip mass is calculated using pretreated three-dimensional coordinate data Shifting value si:
In formula, (xrf,yrf,zrf) be Landslide Monitoring point reference coordinate, siIndicate the accumulation shift value at i moment；
Calculation formula is displaced by accumulation, obtains accumulation displacement value sequence S(0):
S(0)={ s(0)(1),s(0)(2),s(0)(3),…,s(0)(n)}；
(2) dynamic updates data length L, parameter a, b is arranged, detailed process is as follows:
1) one-accumulate is done to the data sequence that length is L, the length of Grey Markov chain predicting model parameter dynamical update is set L is spent, L should be less than the length n of accumulation displacement value sequence, do one-accumulate to the data sequence that length is L；
2) building does one-accumulate to the data sequence that length is L and obtains one-accumulate sequence close to average generation sequence
In formula,Subscript j indicates cycle-index；
ByIt constructs close to average generation sequence Z(1):
In formula,
3) matrix is constructed, value sequence S is displaced according to accumulation(0)With close to average generation sequence Z(1)Construct matrix B, YN:
4) by Least Square Method parameter coefficient, by matrix B, YNGrey is estimated by least square method The parameter a of Verhulst modeljAnd bj:
In formula, parameter a is development coefficient, and size reflects accumulation displacement value sequence S(0)Development trend；Parameter b is Grey actuating quantity, meaning are the systemic effect amount covered with grey information；
5) grey Verhulst time response series are constructed, with obtained ajAnd bjValue constructs the grey Verhulst time Response sequence:
6) it obtains fitting data ordered series of numbers, a regressive operation is done to time response series, obtains fitting data sequence:
7) grey Verhulst fitting data sequence is exported, by process 2) arrive process 5) n-L+1 loop iteration is carried out, often Original series of iteration reject a stale data, and increase a new data, and according to new data sequence calculating parameter A, b, to realize parameter a, the dynamic of b updates, and grey Verhulst fitting data sequence is exported after the completion of loop iteration
(4) it calculates residual error and constructs residual sequence, residual error is calculated by original data sequence and fitting data sequence, is obtained To residual sequence
(5) training GA-BP neural network, including it is following step by step:
(1) topological structure of BP neural network and initialization are determined, BP is determined according to the number for outputting and inputting parameter The topological structure of neural network, the input layer number and output layer number of nodes of BP neural network are respectively by input/output argument Number determines that the selection of best node in hidden layer can refer to following formula:
Q < p-1
Q=log2p
In formula, p is input layer number, and q is node in hidden layer, and r is output layer number of nodes；
The approximate range that node in hidden layer is determined by above formula determines the best of hidden layer with trial and error procedure come final Number of nodes；
(2) basic parameter for determining genetic algorithm, by the basic parameter of BP neural network structure determination genetic algorithm, at random Generate the initial population W=(W that population scale is P1,W2,…,Wp)T, it includes P individuals, a data area is then selected, Generate individual W in a population in the range using linear interpolation functioniReal vector w1,w2,…,wm, and as One chromosome of genetic algorithm；Chromosome contains all weights and threshold value of entire BP neural network, with real coding Mode chromosome is encoded, to obtain high-precision weight and threshold value, the calculation method of code length D are as follows:
D=p*q+q*r+q+r
P is input layer number in formula, and q is node in hidden layer, and r is output layer number of nodes；
(3) population is initialized；
(4) fitness value is calculated, the evolution parameter of BP neural network is given, by what is obtained in previous step by linear interpolation Real vector carries out assignment to the weight and threshold value of BP neural network as the chromosome of genetic algorithm, and input training sample carries out Neural metwork training, the precision for reaching setting obtain network training output valve, with the prediction output of BP neural network and it is expected defeated The inverse of absolute error quadratic sum between out is as fitness function:
(5) selection operation, using Propertional model selection operator, the selection strategy based on fitness ratio is to every generation Chromosome in population is selected, select probability are as follows:
In formula: P is the number of population at individual, FiFor the fitness value of individual i；
(6) crossover operation, using real number interior extrapolation method for k-th of chromosome gkWith first of chromosome by j intersections Operating method are as follows:
In formula, h is the random number on section [0,1]；
(7) mutation operation, j-th of gene for choosing i-th of individual make a variation, mutation probability select one it is smaller Value, calculation method are as follows:
F (g)=r2(1-t/tmax)
In formula, r1It is the random number on section [0,1]；gmaxAnd gminRespectively gijBound, r2It is a random number, T is current iteration number, tmaxFor maximum evolutionary generation；
(8) best initial weights threshold value is obtained, the value of fitness function is calculated, judges whether to meet the preset end item of algorithm Part；If satisfied, then the best initial weights of the neural network after optimization and threshold value are exported；If not satisfied, then back to step by step (4) Continue iteration, until meeting termination condition；
(9) training GA-BP neural network, if the prediction order of built-up pattern is L, then by E(0)(k-1),E(0)(k- 2),…,E(0)(k-L) as the training input sample of GA-BP neural network, by E(0)(k) prediction as GA-BP neural network Desired value trains GA-BP neural network；
(6) prediction residual sequence is obtained, the residual sequence that GA-BP neural network training model predicts is
(7) Combined model forecast value is calculated, is in residual sequenceOn the basis of plus by Grey Markov chain predicting model Obtained match value obtains the final predicted value of built-up pattern.
Compared with prior art, the Grey Markov chain predicting model that the present invention uses is compared to original grey Verhulst Model is improved, and by dynamic undated parameter a, b, reduces influence of the stale information to precision of prediction, to a certain extent Precision of prediction is improved, while improved Grey Markov chain predicting model and GA-BP neural network are combined prediction, is obtained Prediction result compared to single prediction model, precision improves a lot.
Detailed description of the invention
Fig. 1 is overall flow figure of the present invention；
Fig. 2 is that Grey Markov chain predicting model data are fitted flow chart；
Fig. 3 is GA-BP neural network prediction flow chart.
Specific embodiment
A specific embodiment of the invention is further described with reference to the accompanying drawing, but is not to limit of the invention It is fixed.
Fig. 1 shows a kind of dynamic grey Fil Haast neural network landslide deformation prediction method, comprising the following steps:
(1) slip mass accumulating displacement initial data is established, a GNSS monitoring net is established in landslide area, passes through GPRS Either three-dimensional coordinate data is back to server by 3G 4G network, and data are stored to database；
(2) to data prediction, three-dimensional coordinate data is read from database and carries out pretreatment operation；The data are pre- Processing includes Kalman filtering smoothing processing and rejects the outlier and exceptional value that Kalman filtering can not filter out using 3 σ criterion
(3) data are fitted by Grey Markov chain predicting model, specifically step by step as shown in Figure 2:
(1) deformation quantity data sequence is constructed, the accumulation position of slip mass is calculated using pretreated three-dimensional coordinate data Shifting value si:
In formula, (xrf,yrf,zrf) be Landslide Monitoring point reference coordinate, siIndicate the accumulation shift value at i moment；
Calculation formula is displaced by accumulation, obtains accumulation displacement value sequence S(0):
S(0)={ s(0)(1),s(0)(2),s(0)(3),…,s(0)(n)}；
(2) dynamic updates data length L, parameter a, b is arranged, detailed process is as follows:
1) one-accumulate is done to the data sequence that length is L, the length of Grey Markov chain predicting model parameter dynamical update is set L is spent, L should be less than the length n of accumulation displacement value sequence, do one-accumulate to the data sequence that length is L；
2) building does one-accumulate to the data sequence that length is L and obtains one-accumulate sequence close to average generation sequence
In formula,Subscript j indicates cycle-index；
ByIt constructs close to average generation sequence Z(1):
In formula,
3) matrix is constructed, value sequence S is displaced according to accumulation(0)With close to average generation sequence Z(1)Construct matrix B, YN:
4) by Least Square Method parameter coefficient, by matrix B, YNGrey is estimated by least square method The parameter a of Verhulst modeljAnd bj:
In formula, parameter a is development coefficient, and size reflects accumulation displacement value sequence S(0)Development trend；Parameter b is Grey actuating quantity, meaning are the systemic effect amount covered with grey information；
5) grey Verhulst time response series are constructed, with obtained ajAnd bjValue constructs the grey Verhulst time Response sequence:
6) it obtains fitting data ordered series of numbers, a regressive operation is done to time response series, obtains fitting data sequence:
7) grey Verhulst fitting data sequence is exported, by process 2) arrive process 5) n-L+1 loop iteration is carried out, often Original series of iteration reject a stale data, and increase a new data, and according to new data sequence calculating parameter A, b, to realize parameter a, the dynamic of b updates, and grey Verhulst fitting data sequence is exported after the completion of loop iteration
(4) it calculates residual error and constructs residual sequence, residual error is calculated by original data sequence and fitting data sequence, is obtained To residual sequence
(5) training GA-BP neural network, including it is following step by step, as shown in Figure 3:
(1) topological structure of BP neural network and initialization are determined, BP is determined according to the number for outputting and inputting parameter The topological structure of neural network, the input layer number and output layer number of nodes of BP neural network are respectively by input/output argument Number determines that the selection of best node in hidden layer can refer to following formula:
Q < p-1
Q=log2p
In formula, p is input layer number, and q is node in hidden layer, and r is output layer number of nodes；
The approximate range that node in hidden layer is determined by above formula determines the best of hidden layer with trial and error procedure come final Number of nodes；
(2) basic parameter for determining genetic algorithm, by the basic parameter of BP neural network structure determination genetic algorithm, at random Generate the initial population W=(W that population scale is P1,W2,…,Wp)T, it includes P individuals, a data area is then selected, Generate individual W in a population in the range using linear interpolation functioniReal vector w1,w2,…,wm, and as One chromosome of genetic algorithm；Chromosome contains all weights and threshold value of entire BP neural network, with real coding Mode chromosome is encoded, to obtain high-precision weight and threshold value, the calculation method of code length D are as follows:
D=p*q+q*r+q+r
P is input layer number in formula, and q is node in hidden layer, and r is output layer number of nodes；
(3) population is initialized；
(4) fitness value is calculated, the evolution parameter of BP neural network is given, by what is obtained in previous step by linear interpolation Real vector carries out assignment to the weight and threshold value of BP neural network as the chromosome of genetic algorithm, and input training sample carries out Neural metwork training, the precision for reaching setting obtain network training output valve, with the prediction output of BP neural network and it is expected defeated The inverse of absolute error quadratic sum between out is as fitness function:
(5) selection operation, using Propertional model selection operator, the selection strategy based on fitness ratio is to every generation Chromosome in population is selected, select probability are as follows:
In formula: P is the number of population at individual, FiFor the fitness value of individual i；
(6) crossover operation, using real number interior extrapolation method for k-th of chromosome gk and first of chromosome by j intersections Operating method are as follows:
In formula, h is the random number on section [0,1]；
(7) mutation operation, j-th of gene for choosing i-th of individual make a variation, mutation probability select one it is smaller Value, calculation method are as follows:
F (g)=r2(1-t/tmax)
In formula, r1It is the random number on section [0,1]；gmaxAnd gminRespectively gijBound, r2It is a random number, T is current iteration number, tmaxFor maximum evolutionary generation；
(8) best initial weights threshold value is obtained, the value of fitness function is calculated, judges whether to meet the preset end item of algorithm Part；If satisfied, then the best initial weights of the neural network after optimization and threshold value are exported；If not satisfied, then back to step by step (4) Continue iteration, until meeting termination condition；
(9) training GA-BP neural network, if the prediction order of built-up pattern is L, then by E(0)(k-1),E(0)(k- 2),…,E(0)(k-L) as the training input sample of GA-BP neural network, by E(0)(k) prediction as GA-BP neural network Desired value trains GA-BP neural network；
(6) prediction residual sequence is obtained, the residual sequence that GA-BP neural network training model predicts is
(7) Combined model forecast value is calculated, is in residual sequenceOn the basis of plus by Grey Markov chain predicting model Obtained match value obtains the final predicted value of built-up pattern.
Compared with prior art, the Grey Markov chain predicting model that the present invention uses is compared to original grey Verhulst Model is improved, and by dynamic undated parameter a, b, reduces influence of the stale information to precision of prediction, to a certain extent Precision of prediction is improved, while improved Grey Markov chain predicting model and GA-BP neural network are combined prediction, is obtained Prediction result compared to single prediction model, precision improves a lot.
Detailed description is made that embodiments of the present invention in conjunction with attached drawing above, but the present invention be not limited to it is described Embodiment.To those skilled in the art, without departing from the principles and spirit of the present invention, to these implementations Mode carries out various change, modification, replacement and variant are still fallen in protection scope of the present invention.

Claims (4)

  1. The deformation prediction method 1. a kind of dynamic grey Fil Haast neural network comes down, which comprises the following steps:
    (1) establish slip mass accumulating displacement initial data, establish a GNSS monitoring net in landslide area, by GPRS or Three-dimensional coordinate data is back to server by 3G 4G network, and data are stored to database；
    (2) to data prediction, three-dimensional coordinate data is read from database and is pre-processed；
    (3) data are fitted by Grey Markov chain predicting model；
    (4) it calculates residual error and constructs residual sequence, residual error is calculated by original data sequence and fitting data sequence, is obtained residual Difference sequence
    (5) training GA-BP neural network；
    (6) prediction residual sequence is obtained, the residual sequence that GA-BP neural network training model predicts is
    (7) Combined model forecast value is calculated, is in residual sequenceOn the basis of plus being obtained by Grey Markov chain predicting model Match value obtain the final predicted value of built-up pattern.
  2. The deformation prediction method 2. dynamic grey according to claim 1 Fil Haast neural network comes down, feature exist In in step (2), the data prediction includes Kalman filtering smoothing processing and rejects Kalman's filter using 3 σ criterion The outlier and exceptional value that wave can not filter out.
  3. The deformation prediction method 3. dynamic grey according to claim 1 Fil Haast neural network comes down, feature exist In step (3) is specifically as follows step by step:
    (1) deformation quantity data sequence is constructed, the accumulation shift value of slip mass is calculated using pretreated three-dimensional coordinate data si:
    In formula, (xrf,yrf,zrf) be Landslide Monitoring point reference coordinate, siIndicate the accumulation shift value at i moment；
    Calculation formula is displaced by accumulation, obtains accumulation displacement value sequence S(0):
    S(0)={ s(0)(1),s(0)(2),s(0)(3),…,s(0)(n)}；
    (2) dynamic updates data length L, parameter a, b is arranged, detailed process is as follows:
    1) one-accumulate is done to the data sequence that length is L, the length L, L of Grey Markov chain predicting model parameter dynamical update is set The length n that should be less than accumulation displacement value sequence does one-accumulate to the data sequence that length is L；
    2) building does one-accumulate to the data sequence that length is L and obtains one-accumulate sequence close to average generation sequence
    In formula,Subscript j indicates cycle-index；
    ByIt constructs close to average generation sequence Z(1):
    In formula,
    3) matrix is constructed, value sequence S is displaced according to accumulation(0)With close to average generation sequence Z(1)Construct matrix B, YN:
    4) by Least Square Method parameter coefficient, by matrix B, YNGrey Verhulst mould is estimated by least square method The parameter a of typejAnd bj:
    In formula, parameter a is development coefficient, and size reflects accumulation displacement value sequence S(0)Development trend；Parameter b is grey Actuating quantity, meaning are the systemic effect amount covered with grey information；
    5) grey Verhulst time response series are constructed, with obtained ajAnd bjValue constructs grey Verhulst time response Sequence:
    6) it obtains fitting data ordered series of numbers, a regressive operation is done to time response series, obtains fitting data sequence:
    7) grey Verhulst fitting data sequence is exported, by process 2) arrive process 5) carry out n-L+1 loop iteration, every iteration Original series reject a stale data, and increase a new data, and according to new data sequence calculating parameter a, b, To realize parameter a, the dynamic of b updates, and grey Verhulst fitting data sequence is exported after the completion of loop iteration
  4. The deformation prediction method 4. dynamic grey according to claim 1 Fil Haast neural network comes down, feature exist In step (5) is specifically as follows step by step:
    (1) topological structure of BP neural network and initialization are determined, BP nerve is determined according to the number for outputting and inputting parameter The topological structure of network, the input layer number and output layer number of nodes of BP neural network are respectively by the number of input/output argument It determines, the selection of best node in hidden layer can refer to following formula:
    Q < p-1
    Q=log2p
    In formula, p is input layer number, and q is node in hidden layer, and r is output layer number of nodes；
    The approximate range that node in hidden layer is determined by above formula, with trial and error procedure come the final optimal node for determining hidden layer Number；
    (2) basic parameter for determining genetic algorithm, it is random to generate by the basic parameter of BP neural network structure determination genetic algorithm Population scale is the initial population W=(W of P1,W2,…,Wp)T, it includes P individuals, then select a data area, utilize Linear interpolation function generates individual W in a population in the rangeiReal vector w1,w2,…,wm, and as heredity One chromosome of algorithm；Chromosome contains all weights and threshold value of entire BP neural network, with the side of real coding Formula encodes chromosome, to obtain high-precision weight and threshold value, the calculation method of code length D are as follows:
    D=p*q+q*r+q+r
    P is input layer number in formula, and q is node in hidden layer, and r is output layer number of nodes；
    (3) population is initialized；
    (4) fitness value is calculated, the evolution parameter of BP neural network, the real number that will be obtained in previous step by linear interpolation are given Vector carries out assignment to the weight and threshold value of BP neural network as the chromosome of genetic algorithm, and input training sample carries out nerve Network training, the precision for reaching setting obtain network training output valve, with the prediction output of BP neural network and desired output it Between absolute error quadratic sum inverse as fitness function:
    (5) selection operation, using Propertional model selection operator, the selection strategy based on fitness ratio is to every generation population In chromosome selected, select probability are as follows:
    In formula: P is the number of population at individual, FiFor the fitness value of individual i；
    (6) crossover operation, using real number interior extrapolation method for k-th of chromosome gkWith first of chromosome by j crossover operations Method are as follows:
    In formula, h is the random number on section [0,1]；
    (7) mutation operation, j-th of gene for choosing i-th of individual make a variation, and mutation probability selects a smaller value, Calculation method is as follows:
    F (g)=r2(1-t/tmax)
    In formula, r1It is the random number on section [0,1]；gmaxAnd gminRespectively gijBound, r2It is a random number, t is to work as Preceding the number of iterations, tmaxFor maximum evolutionary generation；
    (8) best initial weights threshold value is obtained, the value of fitness function is calculated, judges whether to meet the preset termination condition of algorithm；If Meet, then exports the best initial weights of the neural network after optimization and threshold value；If not satisfied, then continuing back to (4) step by step It is iterated, until meeting termination condition；
    (9) training GA-BP neural network, if the prediction order of built-up pattern is L, then by E(0)(k-1),E(0)(k-2),…,E(0) (k-L) as the training input sample of GA-BP neural network, by E(0)(k) come as the prediction desired value of GA-BP neural network Training GA-BP neural network.
CN201811150026.9A 2018-09-29 2018-09-29 Dynamic gray Verlag neural network landslide deformation prediction method Active CN109492793B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811150026.9A CN109492793B (en) 2018-09-29 2018-09-29 Dynamic gray Verlag neural network landslide deformation prediction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811150026.9A CN109492793B (en) 2018-09-29 2018-09-29 Dynamic gray Verlag neural network landslide deformation prediction method

Publications (2)

Publication Number Publication Date
CN109492793A true CN109492793A (en) 2019-03-19
CN109492793B CN109492793B (en) 2022-06-07

Family

ID=65689328

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811150026.9A Active CN109492793B (en) 2018-09-29 2018-09-29 Dynamic gray Verlag neural network landslide deformation prediction method

Country Status (1)

Country Link
CN (1) CN109492793B (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110059392A (en) * 2019-04-11 2019-07-26 桂林电子科技大学 A kind of landslide deformation prediction method
CN110111377A (en) * 2019-06-06 2019-08-09 西南交通大学 A kind of shake rear region Landslide hazard appraisal procedure considering earthquake displacement field
CN110457655A (en) * 2019-08-12 2019-11-15 长安大学 Slope Deformation Prediction method
CN110470481A (en) * 2019-08-13 2019-11-19 南京信息工程大学 Fault Diagnosis of Engine based on BP neural network
CN110824142A (en) * 2019-11-13 2020-02-21 杭州鲁尔物联科技有限公司 Geological disaster prediction method, device and equipment
CN111027686A (en) * 2019-12-26 2020-04-17 杭州鲁尔物联科技有限公司 Landslide displacement prediction method, device and equipment
CN111199313A (en) * 2019-12-26 2020-05-26 航天信息股份有限公司 Method and system for predicting landslide accumulated displacement trend based on neural network
CN113642812A (en) * 2021-10-15 2021-11-12 西南交通大学 Beidou-based micro-deformation prediction method, device, equipment and readable storage medium
CN114386177A (en) * 2022-01-18 2022-04-22 西北工业大学 Method for estimating model drag coefficient gray of flying wing gliding wing
CN114386177B (en) * 2022-01-18 2024-11-15 西北工业大学 Gray estimation method for wing section resistance coefficient of flying wing glider

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106529185A (en) * 2016-11-24 2017-03-22 西安科技大学 Historic building displacement combined prediction method and system
CN107368928A (en) * 2017-08-03 2017-11-21 西安科技大学 A kind of combination forecasting method and system of ancient building sedimentation
CN107578093A (en) * 2017-09-14 2018-01-12 长安大学 The Elman neural network dynamic Forecasting Methodologies of Landslide Deformation

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106529185A (en) * 2016-11-24 2017-03-22 西安科技大学 Historic building displacement combined prediction method and system
CN107368928A (en) * 2017-08-03 2017-11-21 西安科技大学 A kind of combination forecasting method and system of ancient building sedimentation
CN107578093A (en) * 2017-09-14 2018-01-12 长安大学 The Elman neural network dynamic Forecasting Methodologies of Landslide Deformation

Non-Patent Citations (9)

* Cited by examiner, † Cited by third party
Title
PRATIK CHATURVEDI等: "Prediction of Landslide Deformation Using BackPropagation", 《COMPUTATIONAL INTELLIGENCE: THEORIES, APPLICATIONS AND FUTURE DIRECTIONS》, 23 June 2016 (2016-06-23) *
文海家: "基于GIS的滑坡灾变智能预测系统及研究应用", 《中国博士学位论文全文数据库 基础科学辑》 *
文海家: "基于GIS的滑坡灾变智能预测系统及研究应用", 《中国博士学位论文全文数据库 基础科学辑》, 15 March 2005 (2005-03-15), pages 40 - 41 *
曹智辉: "煤矿工业广场变形预测的改进GA-BP模型建立与分析", 《中国优秀硕士论文全文数据库 工程科技Ⅱ辑》 *
曹智辉: "煤矿工业广场变形预测的改进GA-BP模型建立与分析", 《中国优秀硕士论文全文数据库 工程科技Ⅱ辑》, 15 February 2014 (2014-02-15), pages 19 - 34 *
王晓颖: "改进BP神经网络模型的地基变形预测", 《测绘与空间地理信息》 *
王晓颖: "改进BP神经网络模型的地基变形预测", 《测绘与空间地理信息》, vol. 40, no. 3, 25 March 2017 (2017-03-25), pages 1 - 2 *
高良博等: "灰色Verhulst-BP模型在沉降分析中的应用", 《地理空间信息》 *
高良博等: "灰色Verhulst-BP模型在沉降分析中的应用", 《地理空间信息》, 26 August 2016 (2016-08-26) *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110059392A (en) * 2019-04-11 2019-07-26 桂林电子科技大学 A kind of landslide deformation prediction method
CN110111377A (en) * 2019-06-06 2019-08-09 西南交通大学 A kind of shake rear region Landslide hazard appraisal procedure considering earthquake displacement field
CN110111377B (en) * 2019-06-06 2022-10-21 西南交通大学 Evaluation method for risk of regional landslide after earthquake by considering earthquake displacement field
CN110457655A (en) * 2019-08-12 2019-11-15 长安大学 Slope Deformation Prediction method
CN110457655B (en) * 2019-08-12 2022-11-25 长安大学 Slope deformation prediction method
CN110470481B (en) * 2019-08-13 2020-11-24 南京信息工程大学 Engine fault diagnosis method based on BP neural network
CN110470481A (en) * 2019-08-13 2019-11-19 南京信息工程大学 Fault Diagnosis of Engine based on BP neural network
CN110824142A (en) * 2019-11-13 2020-02-21 杭州鲁尔物联科技有限公司 Geological disaster prediction method, device and equipment
CN110824142B (en) * 2019-11-13 2022-06-24 杭州鲁尔物联科技有限公司 Geological disaster prediction method, device and equipment
CN111199313A (en) * 2019-12-26 2020-05-26 航天信息股份有限公司 Method and system for predicting landslide accumulated displacement trend based on neural network
CN111027686A (en) * 2019-12-26 2020-04-17 杭州鲁尔物联科技有限公司 Landslide displacement prediction method, device and equipment
CN111027686B (en) * 2019-12-26 2023-06-16 杭州鲁尔物联科技有限公司 Landslide displacement prediction method, device and equipment
CN113642812A (en) * 2021-10-15 2021-11-12 西南交通大学 Beidou-based micro-deformation prediction method, device, equipment and readable storage medium
CN113642812B (en) * 2021-10-15 2022-02-08 西南交通大学 Beidou-based micro-deformation prediction method, device, equipment and readable storage medium
CN114386177A (en) * 2022-01-18 2022-04-22 西北工业大学 Method for estimating model drag coefficient gray of flying wing gliding wing
CN114386177B (en) * 2022-01-18 2024-11-15 西北工业大学 Gray estimation method for wing section resistance coefficient of flying wing glider

Also Published As

Publication number Publication date
CN109492793B (en) 2022-06-07

Similar Documents

Publication Publication Date Title
CN109492793A (en) A kind of dynamic grey Fil Haast neural network landslide deformation prediction method
Li et al. Neural-network-based cellular automata for simulating multiple land use changes using GIS
Bayram et al. Comparison of multi layer perceptron (MLP) and radial basis function (RBF) for construction cost estimation: the case of Turkey
Huang et al. An integrated neural-fuzzy-genetic-algorithm using hyper-surface membership functions to predict permeability in petroleum reservoirs
CN108009674A (en) Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks
Yeh et al. Simulation of development alternatives using neural networks, cellular automata, and GIS for urban planning
CN114239114A (en) Truss stress prediction and lightweight method based on transfer learning fusion model
Johari et al. Modelling the mechanical behaviour of unsaturated soils using a genetic algorithm-based neural network
CN103105246A (en) Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm
CN108985514A (en) Load forecasting method, device and equipment based on EEMD and LSTM
CN107423857B (en) A kind of long-term water multiple target joint probability Forecasting Methodology in region
CN114510870B (en) Method and device for predicting residual life of underground structure of urban rail transit
CN108832663A (en) The prediction technique and equipment of the generated output of micro-capacitance sensor photovoltaic generating system
CN110059392A (en) A kind of landslide deformation prediction method
CN104732067A (en) Industrial process modeling forecasting method oriented at flow object
CN113779881A (en) Method, device and equipment for predicting capacity of dense water-containing gas reservoir
Inzunza-Aragón et al. Use of artificial neural networks and response surface methodology for evaluating the reliability index of steel wind towers
Ismail Estimating moment capacity of ferrocement members using self-evolving network
CN117455551A (en) Industry electricity consumption prediction method based on industry relation complex network
US11927717B2 (en) Optimized methodology for automatic history matching of a petroleum reservoir model with Ensemble Kalman Filter (EnKF)
AU2021100592A4 (en) Quantum seeded hybrid evolutionary computational process for constrained optimization
CN115408915A (en) Sensor arrangement method and system
Mučenski et al. Estimation of recycling capacity of multistorey building structures using artificial neural networks
Kapelan et al. Robust least cost design of water distribution systems using GAs
CN108596781A (en) Data mining and prediction integration method for power system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant