CN109284910A - A kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm - Google Patents
A kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm Download PDFInfo
- Publication number
- CN109284910A CN109284910A CN201811025281.0A CN201811025281A CN109284910A CN 109284910 A CN109284910 A CN 109284910A CN 201811025281 A CN201811025281 A CN 201811025281A CN 109284910 A CN109284910 A CN 109284910A
- Authority
- CN
- China
- Prior art keywords
- price
- land
- sample
- datum
- appraisal
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 74
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 37
- 238000013135 deep learning Methods 0.000 title claims abstract description 24
- 238000012360 testing method Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 21
- 238000009313 farming Methods 0.000 claims abstract description 20
- 238000013480 data collection Methods 0.000 claims abstract description 8
- 238000011156 evaluation Methods 0.000 claims description 18
- 230000000007 visual effect Effects 0.000 claims description 15
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 9
- 238000013139 quantization Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 238000011161 development Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000012937 correction Methods 0.000 claims description 3
- 238000013016 damping Methods 0.000 claims description 3
- 238000013278 delphi method Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- 206010034719 Personality change Diseases 0.000 claims 1
- 238000013507 mapping Methods 0.000 abstract description 6
- 238000005070 sampling Methods 0.000 description 12
- 238000012706 support-vector machine Methods 0.000 description 9
- 239000002689 soil Substances 0.000 description 6
- 238000011835 investigation Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- FNMKZDDKPDBYJM-UHFFFAOYSA-N 3-(1,3-benzodioxol-5-yl)-7-(3-methylbut-2-enoxy)chromen-4-one Chemical compound C1=C2OCOC2=CC(C2=COC=3C(C2=O)=CC=C(C=3)OCC=C(C)C)=C1 FNMKZDDKPDBYJM-UHFFFAOYSA-N 0.000 description 1
- 241000208340 Araliaceae Species 0.000 description 1
- 208000032170 Congenital Abnormalities Diseases 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007698 birth defect Effects 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007596 consolidation process Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 239000003621 irrigation water Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 239000005416 organic matter Substances 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001568 sexual effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0283—Price estimation or determination
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Forestry; Mining
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Accounting & Taxation (AREA)
- Tourism & Hospitality (AREA)
- Finance (AREA)
- Marine Sciences & Fisheries (AREA)
- Mining & Mineral Resources (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Agronomy & Crop Science (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Animal Husbandry (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
Abstract
The invention proposes a kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm, steps are as follows: collects marketing data, constructs the sample data set of land price；It is normalized using characteristic of the z-score method to sample, and raw sample data collection is split as training sample set and test sample set at random；Datum Price Appraisal of Arable Land model is constructed using depth confidence network algorithm, sample training study is carried out based on deep layer network structure, saves the parameter of the highest deep layer network structure of fitting precision；The characteristic value of farming land assessment unit is input to the land price that assessment unit is calculated in trained deep layer network structure；Assessment unit rank delimited using total score frequency method, area weighting factor method is selected to carry out benchmark land price calculating.The present invention can establish the mapping relations of land price and Factors influencing land price with higher fitting precision, and striked benchmark land price and agricultural geological measuring maintain preferable consistency on space distribution rule.
Description
Technical field
The present invention relates to the technical fields of farming land soil Appraisal of Standard Land Price, and in particular to one kind is calculated based on deep learning
The Datum Price Appraisal of Arable Land method of method.
Background technique
Arable land evaluation (agricultural land appraisal, ALA) work is that China makes for promotion rural holding
The important process reformed with institutional deepening and carried out can be managed collectively agricultural land rationally, in accordance with the law for national science, cultivate
Land market provides scientific accurate farming land standard price system.Datum Price Appraisal of Arable Land is for smoothly carrying out soil
The work of the Rural land managements such as conversion of contracted managerial power, Land Requisition Compensation, consolidation, real estate properties reasonable disposition has very heavy
The meaning wanted.
China's common method in Price Assessment working practice includes arithmetic average model and regression model: being counted flat
Equal model is averaged the sampling point land price in certain homogenous area to determine the zonal basis land price, has a disadvantage in that the result is tight
Space dependent on sampling point in region and distributed number situation again；The regression calculations model such as linear model and exponential model is by building
The mathematical model between sampling point land price and Intensive land use or Grading unit function points is found to predict benchmark land price, but the mathematical modulo
Type needs artificially determine numerous influence factor weighted values in advance, with very big subjectivity, empirical with uncertain, and nothing
Complicated non-linear relation between the accurate simulation land price of method and its influence factor.For defect existing for conventional method, grind
Study carefully personnel to attempt to improve traditional regression models using the methods of fuzzy mathematics and cloud model, but the above method is more paid close attention to and determined
The project evaluation chain of sexual factor does not improve the birth defect of regression model.It is influenced due to influencing the natural, social, economic etc. of land price
Factor spatially all has randomness and structural, to eliminate influence of the spatial autocorrelation of variable to regression result itself,
Kriging Spatial Interpolation Method be used to construct land price solve equation, but this method do not eliminate the effects of the act factor weight value with
Meaning property.In recent ten years, artificial nerve network model is widely applied in Appraisal of Standard Land Price research.Artificial neural network
Model has the advantage of autonomous learning land price sample characteristics, and does not need to determine weight, overcome the determination of multifactor weight by
Human factor influences big defect, and the mapping relations between land price and influence factor are established using multilayer neural network.?
Have scholar it is further proposed that be based on support vector machines (support vector machine, SVM) Appraisal of Standard Land Price model,
The algorithm either will slightly be better than artificial nerve network model to the fit solution of sample and to the estimation precision of land price.Nothing
By being that artificial nerve network model or support vector machines belong to shallow-layer learning algorithm, limited computing unit leads to shallow-layer
The network of study is difficult to comprehensively characterize the complicated function relationship between influence factor and agricultural standard land price, and with sample
The floating of this quantity and diversity increase, and shallow Model can not also adapt to complicated sample, Price Assessment precision also therefore by
Larger impact.
The deep learning technology risen in recent years makes it have powerful answer due to possessing multilayered nonlinear mapping network layer
Miscellaneous function ability to express is applied widely in the solution of complicated the problems such as classifying, identifying, predict, and obtains good
Effect and efficiency.The development that deep learning is considered as neural network, Hinton etc. think that deep-neural-network structure can be with
Learn to the more deep more essential feature of object.
Summary of the invention
For subjectivity existing for existing model strong, shallow-layer network be unable to characterize influence factor and agricultural standard land price it
Between the technical problems such as complicated function relationship, the invention proposes a kind of agricultural standard land prices based on deep learning algorithm to comment
Estimate method, utilize the classic algorithm in deep learning method ----depth confidence network (deep belief network, DBN)
Datum Price Appraisal of Arable Land model is constructed, using the deep layer network structure of DBN algorithm more accurately characterizes influence factor
Complicated function relationship between agricultural standard land price, and then improve the reasonability of agricultural standard land price.
In order to achieve the above object, the technical scheme is that a kind of agricultural ground reference based on deep learning algorithm
Land evaluation methods, its step are as follows:
Step 1: collecting marketing data and construct Factors influencing land price feature architecture, collects the relevant space number of land price
According to the sample characteristics data set of building land price；
Step 2: being normalized using characteristic of the z-score method to sample, and by sample characteristics data
Collection is split as training sample and test sample according to setting ratio at random, gives all sample land price labels；
Step 3: Datum Price Appraisal of Arable Land model is constructed using depth confidence network algorithm, is based on deep layer network knot
Structure carries out sample training study, carries out test assessment to deep layer network, saves the structure ginseng of the highest deep layer network of fitting precision
Number；
Step 4: to farming land assessment unit, each characteristic value quantifies, and with being input to trained ground reference
In valence assessment models, the land price of each assessment unit is calculated；
Step 5: using total score frequency method to export land price as foundation using all assessment units and delimit assessment unit rank,
And area weighting factor method is selected, benchmark land price meter is carried out using the land price of the assessment unit in each rank and corresponding area weight
It calculates.
The Factors influencing land price feature architecture is the arable land price evaluation Factor system determined using Delphi method, arable land
Price evaluation Factor system include planting income, farming land hire out, contract subcontract, land development, farming land mortgage and it is agricultural
Ground requisition；The land price of sample is adapted to the price level of agricultural standard land price intension with benifit-sharing contract, using the term
Correction factor, farming land amount of cure, phase day amendment and regional conditions amendment carry out coefficient amendment.
The sample characteristics data set X of the step 1 land price is indicated are as follows:
Wherein,For d-th of characteristic value of first of sample,Indicate first of sample plot in some attributive character
Quantized value, 1≤l≤L, 1≤d≤D, L be sample data set quantity, D be each sample data Characteristic Number.
The method that z-score method carries out characteristic value normalization in the step 2 are as follows:
Wherein,For normalization after data feature values,For d-th of characteristic value of first of sample,For sample spy
Levy the average of d-th of characteristic value of data set X, σdFor the standard deviation of sample characteristics data set d-th of characteristic value of X.
The data set of all sample land price labels in the step 2 are as follows: Y=[y1 y2 ... yl ... yL]T；
Wherein, ylIndicate the corresponding actual market price in first of sample plot, actual market price ylWith first of sample number
According to xlIt is corresponding.
The construction method of Datum Price Appraisal of Arable Land model in the step 3 are as follows:
(1) sample pre-training: using the D dimensional feature vector of all training sample data collection as input, land price label data
Collecting Y is output, and the unsupervised greedy method learnt is limited in Datum Price Appraisal of Arable Land model to train layer by layer for use
Boltzmann machine；In each layer, visual layers h is calculatedjAnd hidden layer viThe state of unit:
Wherein, p (hj=1) it indicates by implying layer unit viIt is mapped to visual layers hjThe value of unit, p (vi=1) indicate visual
Layer unit hjIt is mapped to hidden layer viThe value of unit, w indicate the weight between visual layers and hidden layer, bjIndicate visual layers biasing
Amount, ciIndicate hidden layer amount of bias, parameter space (w, b, c) is updated by deep learning algorithm；
(2) it finely tunes: being trained using BP network of the supervised learning mode to the last layer: the last layer is limited glass
The output of the graceful machine of Wurz passes to output end by BP input terminal, then according to the error of the output result of propagated forward and desired value from
Output end carries out backpropagation to input terminal, and then is finely adjusted to the network parameter of entire depth confidence network, until iteration
Until number reaches setting value；
(3) it tests and assesses: the data of test sample are input to trained Datum Price Appraisal of Arable Land model, benefit
Average assessment errors rate λ is calculated with the output land price of Datum Price Appraisal of Arable Land model and sample label land price:
Wherein, ytestIndicate the practical land price of test sample, NtestIndicate the quantity of test sample set, y ' indicates model
Output land price, average assessment errors rate λ chooses and puts down as the standard for measuring Datum Price Appraisal of Arable Land model evaluation performance
The parameter of the smallest Datum Price Appraisal of Arable Land model of equal assessment errors rate λ, and saved.
The parameter of the Datum Price Appraisal of Arable Land model includes the structural parameters of limited Boltzmann machine network, connection
Weight, offset parameter (w, b, c) and the neuron number of neural network, connection weight, amount of bias.
The method that the characteristic value of the assessment unit is quantified are as follows: for three kinds of point, line, surface different geometric elements because
The comprehensive function that sublayer selects common space quantization method that different each assessment unit of factor pair are calculated divides, and according to
Composition multi-dimensional matrix is normalized in z-score method；The space quantization method be straight line damped method, exponential damping normal,
Maximin method or mean value degree method.
The land price calculation method of assessment unit is by the way that assessment unit feature vector to be input to one by one in the step 4
In the deep layer network structure of the trained Datum Price Appraisal of Arable Land model of step 3, calculates output and obtain each assessment list
The land price of member.
Benchmark land prices at different levels are calculated using area weighting factor method in the step 5 are as follows:
Wherein, BValuemIndicate the benchmark land price of m rank, AreanIndicate the face of n-th of assessment unit of m rank
Product, TotalAreamIndicate the sum of the area of all units of m rank, ZValuenExpression is commented using n-th that model is sought
Estimate the land price of unit, N indicates m level assessment unit sum.
The beneficial effects of the present invention are: 1) deep learning method and Datum Price Appraisal of Arable Land problem are mutually tied for the first time
It closes, provides new approaches for Price Assessment work；2) compared to traditional homing method, artificial neural network and support vector machines
Equal shallow-layers learning model, the deep layer network structure of DBN algorithm obviously can preferably excavate the deep layer of farming land land price sample set
Feature obtains higher Price Assessment precision；3) operational efficiency of DBN algorithm is not less than artificial neural network and supporting vector
Machine, and as the increase for calculating sample can embody certain advantage；4) benchmark land price that measuring and calculating obtains can be good at instead
Mirror the spatial distribution characteristic of agricultural geological measuring.In short, the present invention can simulate soil by depth confidence network with degree of precision
Complicated Nonlinear Mapping relationship between price and influence factor, and each rank benchmark land price of assessment unit is sought, in China's agriculture
There is important practical significance and application value in land used Appraisal of Standard Land Price.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is model flow figure of the invention.
Fig. 2 is the supervised training convergence graph of the present invention with artificial neural network algorithm.
Fig. 3 is the farmland quality grade figure of present invention specific implementation case.
Fig. 4 is some effects factor profile of present invention specific implementation case.
Fig. 5 is the benchmark land price distribution map of present invention specific implementation case.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under that premise of not paying creative labor
Embodiment shall fall within the protection scope of the present invention.
As shown in Figure 1, a kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm, in depth confidence network
Factors influencing land price and sample are constructed using its unsupervised and Training method on the deep layer network structure basis of model
Complex mapping relation between this land price can be distinguished the feature vector of each rank assessment unit defeated based on the mapping network
Enter, the land price of output assessment unit is calculated by depth confidence network model parameter, then pass through weighting or the side being averaged
Method can seek the fifty-fifty valence i.e. benchmark land price of a certain rank.Using certain city arable land Appraisal of Standard Land Price as case study on implementation, specifically
Steps are as follows:
Step 1: collecting marketing data and construct Factors influencing land price feature architecture, collect land price correlation space data,
Construct land price sample data set.
Sampling point data field investigation is carried out in the form of investigation form within the scope of the whole city, while being carried out using handhold GPS
Sample point location, it is ensured that each administrative village includes at least ten or more sampling points, amounts to 12555 paddy fields and nonirrigated farmland investment
Output sampling point.To each sampling point from planting income, farming land hire out, contract subcontract, land development, farming land mortgage and agriculture
Several aspects such as land used requisition carry out data collection and investigation work.
To the total revenue (major product yield × unit price+byproduct yield × unit price) of the sampling point of preliminary investigation, general expenses
(substance expense+labour cost+investment opportunity cost+tax) and net earnings (soil total revenue-soil total cost) are itemized
It checks, is noted abnormalities data using twice of standard deviation, and it is corrected or is rejected, then united to modified sampling point
One number and standardization processing are simultaneously put in storage spare, implement on Grading unit figure by the specific location of sampling point, effective sampling point is total
11496.
Sampling point land price is adapted to the price level of agricultural standard land price intension with benifit-sharing contract, is repaired using the term
Positive coefficient, farming land amount of cure, phase day amendment and the amendment of regional conditions correction factor.
According to the Factor system recommended in agricultural land evaluation national rule, local Bureau of Land and Resources, agricultural are being solicited
Office, water conservancy traffic and be engaged in land grading estimation professional person opinion basis on, with reference to it is existing research for farming land
The analysis of land price impact factor finally determines Puning City arable land price evaluation Factor system using Delphi method, and total 19 are commented
The valence factor, as shown in table 1 below.And E in the formula of scoring criteria in table 1cFor evaluation unit factor c actual value, EminFor evaluation unit
Factor minimum value, EmaxFor evaluation unit factor maximum value, fcIt is the function points of evaluation unit factor c；M=100；R is opposite
Distance；dcThe radius of influence for evaluation unit away from diffusion source.
The arable land of table 1 price evaluation index system
Collect the data and space of land use, weather, the hydrology, soil, landform, transportation condition, Land Economic etc.
Graph etc..Land use data from Bureau of Land and Resources provide Land Change Survey data, land use renewal survey and
Second of land investigation database；The data such as terrain slope, ensurance probability of irrigation water, pH value, the content of organic matter equally derive from territory
The Agro-land Classification And Gradation data that office provides；Road it is sensible degree, Primary Reference traffic department of bus station provide category of roads figure,
The data such as road distribution map, communication chart, each factors quantization standard can be found in table 2.
The arable land of table 2 Factors influencing land price quantizing rule
The eigenvectors matrix for constructing land price sample data set indicates sample characteristics data set X are as follows:
Wherein,For d-th of characteristic value of first of sample, indicate first of sample plot in some attributive character
Quantized value, 1≤l≤L, 1≤d≤D, L are the quantity of sample data set, and D is the Characteristic Number of each sample data.Each is gathered around
Space R can be regarded as by having the data of D featureDIn a vector, i.e. a line in sample characteristics data set X.
Sample label data set Y corresponding with sample characteristics data set X can be expressed as formula (2), ylIndicate first of sample
The corresponding actual market price in plot.
Y=[y1 y2 ... yl ... yL]T (2)
Step 2: being normalized using eigen data of the z-score method to sample, and by sample characteristics data
Collection is split as training sample and test sample according to setting ratio at random, gives the land price label of all samples.
Using formula (3) to characteristic valueIt is normalized, has both met depth confidence network algorithm to input feature vector data
The requirement of format can eliminate the influence of different dimensions again.
Wherein,For normalization after first of sample d-th of data characteristic value,For sample characteristics data set X
The average of d characteristic value, σdFor the standard deviation of sample characteristics data set d-th of characteristic value of X.
Training sample of 10000 samples as this model in randomly drawing sample characteristic data set X, remaining 1496
It is a to be used as test sample.
Step 3: Datum Price Appraisal of Arable Land model is constructed using depth confidence network algorithm: being based on its deep layer network
Structure carries out sample training study, carries out test assessment to network, saves the highest network architecture parameters of fitting precision.
3 are set by the implicit number of layers of depth confidence network, i.e., total number of plies is 5 layers, and every layer of neural unit number is successively
For 19-12-6-3-1, unsupervised training stage learning rate is 0.6, the number of iterations 200, just for the setting of dynamic regularized learning algorithm rate
Initiating quantifier parameter is 0.5, and the middle and later periods is adjusted to 0.9.
The hardware platform used is tested as 4 core of Intel (R) Core (TM) i7-4600U CPU, dominant frequency 2.1GHz, memory
16GB.Software configuration is 7 professional version of Microsoft Windows, 64 bit manipulation system, 2015 running environment of Matlab.
Sample pre-training: using the D dimensional feature vector of all training sample data collection after normalized as input, sample
This label data collection Y is output, and the unsupervised greedy method learnt carrys out the limited Boltzmann in training pattern layer by layer for use
Machine (RBM, restricted boltz-mann machine).In each layer, visual layers hjAnd hidden layer viThe state of unit is pressed
Illuminated (4) and formula (5) are calculated, and w indicates the weight between visual layers and hidden layer, bjIndicate visual layers amount of bias, ciIt indicates
Hidden layer amount of bias, parameter space (w, b, c) can be updated by deep learning algorithm.p(hj=1) it indicates by hidden layer list
First viIt is mapped to visual layers hjThe value of unit, p (vi=1) visual layer unit h is indicatedjIt is mapped to hidden layer viThe value of unit.
Fine tuning: it is trained using BP network of the supervised learning mode to the last layer, specifically by the last layer RBM
Output output end is passed to by BP input terminal, then according to the error of the output result of propagated forward and desired value from output end to
Input terminal carries out backpropagation, and then is finely adjusted to the parameter of entire depth confidence network, until the number of iterations reaches setting
Until value.
Test and assessment: test sample data set is input to trained depth confidence network model, utilizes model
Output land price and sample label land price are calculated average assessment errors rate λ and as measurement depth confidence network evaluations
The standard of energy, the depth confidence network model for selecting average assessment errors rate minimum, saves its network architecture parameters.Average assessment
Shown in error rate λ such as formula (6), in formula, ytestIndicate the practical land price of test sample, NtestIndicate the quantity of test sample set,
The output land price of y ' expression model.
The artificial neural network algorithm of depth confidence network (DBN) algorithm and two kinds of heterogeneous networks structures of the application
(BPANN1, BPANN2), algorithm of support vector machine (SVM) carry out algorithm effect comparison, using formula (6) to test of heuristics result
It being assessed, the results are shown in Table 3, wherein Mo indicates error in land price, and Max indicates worst error, and Min indicates minimal error,
Mean indicates mean error.The supervised training process of DBN algorithm and BPANN2 are as shown in Figure 2.
The test result of the different assessment models of table 3
It can be seen that relative error ratio tetra- models of BPANN1, BPANN2, SVM of DBN model are wanted respectively by the result of table 3
Low 3.61%, 8.14%, 3.12%.Compared to BPANN1, SVM model, the deep layer framework of DBN model obviously can preferably be dug
Pick up the further feature of valence sample set, and 19 original dimensional features have been abstracted into 3 dimension high-order features and have been assessed, can have been obtained
Fitting precision more better than shallow structure.
Parameter by the depth confidence network model of optimal measuring accuracy include the structural parameters of RBM network, connection weight,
Offset parameter space (w, b, c) and the neuron number of neural network, connection weight, amount of bias are stored in file
In Model.txt text.
Step 4: to farming land assessment unit, each characteristic value quantifies, and is input to trained deep layer confidence
Among network model, the land price of each assessment unit is calculated.
Assessment unit delimited: by Puning City's present landuse map, administrative division boundary, topographic map, farming land Quality Map
The progress such as layer are schemed to be stacked more, form closed assessment unit using the superposition method of Arcgis software, are less than 6mm to area2Figure
Spot is edited automatically and merger, ultimately generates 8970 assessment units.
The quantization of assessment unit characteristic value: according to " Farmland grading ", " Farmland Grading regulation " and " farming land is advised
Journey ", for the different geometric elements of three kinds of point, line, surface factor layer choosing select common space quantization method be calculated it is different because
Comprehensive function point of the son to each assessment unit, space quantization method such as straight line damped method, exponential damping normal, minimax
Value method or mean value degree method, and it is normalized according to formula (3), construct the multi-dimensional matrix as shown in formula (1).
Assessment unit land price calculates: the network parameter of deep layer confidence network is read from file Model.txt, after quantization
Assessment unit eigenvalue cluster at vector matrix inputted among deep layer confidence network model one by one, output each can be obtained
The land price of assessment unit.
Step 5: using total score frequency method to export land price as foundation using all assessment units and delimit assessment unit rank,
And area weighting factor method is selected, benchmark land price meter is carried out using the land price of the assessment unit in each rank and corresponding area weight
It calculates.
Combining local reality to calculate gained assessment unit land price according to depth confidence network model using total score frequency method will
Arable land assessment unit is divided into 5 grades.Then benchmark land prices at different levels are calculated in formula (7), wherein BValuemTable
Show the benchmark land price of m rank, AreanIndicate the area of n-th of assessment unit in m rank, TotalAreamIndicate m grades
The sum of the area of other all assessment units, ZValuenIndicate n-th of the assessment unit sought using deep layer confidence network model
Land price, N indicate m level assessment unit sum.The maximum value of rank m is 5, and specific calculated result is shown in Table 4.
The arable land rank of table 4 and benchmark land price
Fig. 3 is based on city's farmland quality spatial data drawing gained, and Fig. 4 is from territory, the obtained space in traffic part
Data drawing gained, Fig. 5 are based on the acquired benchmark land price distribution of results figure of the method for the present invention.It can be with by Fig. 3-Fig. 5 comparison
Find out that the present invention calculates gained arable land benchmark land price and farmland quality shows more consistent heterogeneity in spatial distribution and advises
Rule, and depth confidence network model can comprehensively consider the factors such as natural quality, Traffic area location in training early period, so that two
There is also certain othernesses for the spatial distribution of person.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm, which is characterized in that its step are as follows:
Step 1: collecting marketing data and construct Factors influencing land price feature architecture, collects the relevant spatial data structure of land price
Build the sample characteristics data set of land price；
Step 2: it is normalized using characteristic of the z-score method to sample, and sample characteristics data set is pressed
It is split as training sample and test sample at random according to setting ratio, gives all sample land price labels；
Step 3: using depth confidence network algorithm construct Datum Price Appraisal of Arable Land model, based on deep layer network structure into
The study of row sample training carries out test assessment to deep layer network using test sample, saves the highest deep layer network of fitting precision
Structural parameters；
Step 4: to farming land assessment unit, each characteristic value quantifies, and is input to trained standard land price and comments
Estimate in model, the land price of each assessment unit is calculated；
Step 5: it uses total score frequency method to export land price as foundation using all assessment units and delimit assessment unit rank, and select
With area weighting factor method, benchmark land price calculating is carried out using the land price of the assessment unit in each rank and corresponding area weight.
2. the Datum Price Appraisal of Arable Land method according to claim 1 based on deep learning algorithm, which is characterized in that
The Factors influencing land price feature architecture is the arable land price evaluation Factor system determined using Delphi method, price evaluation of ploughing
Factor system include planting income, farming land hire out, contract subcontract, land development, farming land mortgage and farming land requisition；Fortune
The land price of sample is adapted to the price level of agricultural standard land price intension with benifit-sharing contract, using annual correction coefficient,
Farming land amount of cure, phase day amendment and regional conditions amendment carry out coefficient amendment.
3. the Datum Price Appraisal of Arable Land method according to claim 1 based on deep learning algorithm, which is characterized in that
The sample characteristics data set X of the step 1 land price is indicated are as follows:
Wherein,For d-th of characteristic value of first of sample,Indicate amount of first of sample plot in some attributive character
Change value, 1≤l≤L, 1≤d≤D, L are the quantity of sample data set, and D is the Characteristic Number of each sample data.
4. the Datum Price Appraisal of Arable Land method according to claim 3 based on deep learning algorithm, which is characterized in that
The method that z-score method carries out characteristic value normalization in the step 2 are as follows:
Wherein,For normalization after data feature values,For d-th of characteristic value of first of sample,For sample characteristics data
Collect the average of d-th of characteristic value of X, σdFor the standard deviation of sample characteristics data set d-th of characteristic value of X.
5. the Datum Price Appraisal of Arable Land method according to claim 3 based on deep learning algorithm, which is characterized in that
The data set of all sample land price labels in the step 2 are as follows: Y=[y1 y2 ... yl ... yL]T；
Wherein, ylIndicate the corresponding actual market price in first of sample plot, actual market price ylWith first of sample data xl
It is corresponding.
6. the Datum Price Appraisal of Arable Land method according to claim 5 based on deep learning algorithm, which is characterized in that
The construction method of Datum Price Appraisal of Arable Land model in the step 3 are as follows:
(1) sample pre-training: using the D dimensional feature vector of the data set of all training samples as input, land price label data collection Y
For output, the method for use unsupervised greedy study layer by layer trains limited Bohr in Datum Price Appraisal of Arable Land model
Hereby graceful machine；In each layer, visual layers h is calculatedjAnd hidden layer viThe state of unit:
Wherein, p (hj=1) it indicates by implying layer unit viIt is mapped to visual layers hjThe value of unit, p (vi=1) visual layers list is indicated
First hjIt is mapped to hidden layer viThe value of unit, w indicate the weight between visual layers and hidden layer, bjIndicate visual layers amount of bias, ci
Indicate hidden layer amount of bias, parameter space (w, b, c) is updated by deep learning algorithm；
(2) it finely tunes: being trained using BP network of the supervised learning mode to the last layer: the last layer is limited Bohr hereby
The output of graceful machine passes to output end by BP input terminal, then according to the error of the output result of propagated forward and desired value from output
It holds input terminal to carry out backpropagation, and then the network parameter of entire depth confidence network is finely adjusted, until the number of iterations
Until reaching setting value；
(3) it tests and assesses: the data of test sample being input to trained Datum Price Appraisal of Arable Land model, utilize agriculture
Average assessment errors rate λ is calculated in the output land price and sample label land price of land used Appraisal of Standard Land Price model:
Wherein, ytestIndicate the practical land price of test sample, NtestIndicate the quantity of test sample set, y ' indicates the defeated of model
Land price out, average assessment errors rate λ are averagely commented as the standard for measuring Datum Price Appraisal of Arable Land model evaluation performance, selection
Estimate the parameter of the smallest Datum Price Appraisal of Arable Land model of error rate λ, and is saved.
7. the Datum Price Appraisal of Arable Land method according to claim 6 based on deep learning algorithm, which is characterized in that
The parameter of the Datum Price Appraisal of Arable Land model includes the structural parameters of limited Boltzmann machine network, connection weight, partially
Set parameter (w, b, c) and the neuron number of neural network, connection weight, amount of bias.
8. the Datum Price Appraisal of Arable Land method according to claim 6 based on deep learning algorithm, which is characterized in that
The method that the characteristic value of the assessment unit is quantified are as follows: selected for the factor layer choosing of three kinds of point, line, surface different geometric elements
The comprehensive function point of different each assessment unit of factor pair is calculated in common space quantization method, and according to z-score
Composition multi-dimensional matrix is normalized in method；The space quantization method is straight line damped method, exponential damping normal, minimax
Value method or mean value degree method.
9. the Datum Price Appraisal of Arable Land method according to claim 6 based on deep learning algorithm, which is characterized in that
The land price calculation method of assessment unit is by the way that assessment unit feature vector is input to step 3 instruction one by one in the step 4
In the deep layer network structure for the Datum Price Appraisal of Arable Land model perfected, calculates output and obtain the ground of each assessment unit
Valence.
10. the Datum Price Appraisal of Arable Land method according to claim 1 based on deep learning algorithm, feature exist
In calculating benchmark land prices at different levels using area weighting factor method in the step 5 are as follows:
Wherein, BValuemIndicate the benchmark land price of m rank, AreanIndicate the area of n-th of assessment unit of m rank,
TotalAreamIndicate the sum of the area of all units of m rank, ZValuenIndicate that n-th of the assessment sought using model is single
The land price of member, N indicate m level assessment unit sum.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811025281.0A CN109284910A (en) | 2018-09-04 | 2018-09-04 | A kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811025281.0A CN109284910A (en) | 2018-09-04 | 2018-09-04 | A kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109284910A true CN109284910A (en) | 2019-01-29 |
Family
ID=65183903
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811025281.0A Pending CN109284910A (en) | 2018-09-04 | 2018-09-04 | A kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109284910A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110955663A (en) * | 2019-12-02 | 2020-04-03 | 重庆市勘测院 | Large-scale regional land resource asset liability statement compiling method |
| CN113435707A (en) * | 2021-06-03 | 2021-09-24 | 大连钜智信息科技有限公司 | Soil testing and formulated fertilization method based on deep learning and weighted multi-factor evaluation |
-
2018
- 2018-09-04 CN CN201811025281.0A patent/CN109284910A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110955663A (en) * | 2019-12-02 | 2020-04-03 | 重庆市勘测院 | Large-scale regional land resource asset liability statement compiling method |
| CN113435707A (en) * | 2021-06-03 | 2021-09-24 | 大连钜智信息科技有限公司 | Soil testing and formulated fertilization method based on deep learning and weighted multi-factor evaluation |
| CN113435707B (en) * | 2021-06-03 | 2023-11-10 | 大连钜智信息科技有限公司 | Soil testing formula fertilization method based on deep learning and weighting multi-factor evaluation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zabihi et al. | A GIS-based fuzzy-analytic hierarchy process (F-AHP) for ecotourism suitability decision making: A case study of Babol in Iran | |
| CN109858647B (en) | Regional flood disaster risk evaluation and estimation method coupled with GIS and GBDT algorithm | |
| Tarmizi et al. | Impact of the economic growth and acquisition of land to the construction cost index in North Sumatra | |
| Banker et al. | Some models for estimating technical and scale inefficiencies in data envelopment analysis | |
| CN108536908B (en) | Method for evaluating watershed water environment safety based on non-point source nitrogen and phosphorus loss risk | |
| CN106355334A (en) | Farmland construction area determining method | |
| Lu et al. | Evaluation the effect of cultivated land protection policies based on the cloud model: A case study of Xingning, China | |
| CN105931116A (en) | Automated credit scoring system and method based on depth learning mechanism | |
| CN109360018A (en) | A kind of fuzzy zone land price estimation method based on artificial neural network | |
| BACHEV et al. | Sustainability of Agricultural Industries in Bulgaria. | |
| Ahmadi Dehrashid et al. | Food security assessment in rural areas: evidence from Iran | |
| CN104867074A (en) | Student comprehensive quality evaluation method based on genetic algorithm optimization BP neural network | |
| CN106952170A (en) | A kind of county domain Natural quality of cultivated land evaluation method based on random forest | |
| Ghajar et al. | An adaptive network-based fuzzy inference system for rock share estimation in forest road construction | |
| Bagherzadeh et al. | Parametric-based neural networks and TOPSIS modeling in land suitability evaluation for alfalfa production using GIS | |
| Li et al. | The functional assessment of agricultural ecosystems in Hubei Province, China | |
| CN109284910A (en) | A kind of Datum Price Appraisal of Arable Land method based on deep learning algorithm | |
| Sohrabi et al. | Risk-supported case-based reasoning approach for cost overrun estimation of water-related projects using machine learning | |
| Bąk et al. | Fuzzy cognitive maps and their application in the economic sciences | |
| Siddiqui et al. | Interpreting the nature of rainfall with ai and big data models | |
| Antonopoulos et al. | Evaluation of different methods to estimate monthly reference evapotranspiration in a Mediterranean area | |
| Khan et al. | Irrigation water requirement prediction through various data mining techniques applied on a carefully pre-processed dataset | |
| Ertunç | A new approach with AHP for land grading in land consolidation projects | |
| Lynam | Scientific measurements and villagers' knowledge: an integrative multi-agent model from the semi-arid areas of Zimbabwe | |
| BACHEV | Assessment of agrarian sustainability at various levels: The case of Bulgaria |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190129 |