CN107844870A - Heavy metal content in soil Forecasting Methodology based on Elman neural network models - Google Patents
Heavy metal content in soil Forecasting Methodology based on Elman neural network models Download PDFInfo
- Publication number
- CN107844870A CN107844870A CN201711281809.6A CN201711281809A CN107844870A CN 107844870 A CN107844870 A CN 107844870A CN 201711281809 A CN201711281809 A CN 201711281809A CN 107844870 A CN107844870 A CN 107844870A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- layer
- msubsup
- soil
- 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
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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/02—Agriculture; Fishing; Mining
Abstract
The present invention relates to the heavy metal content in soil Forecasting Methodology based on Elman neural network models, solves do not consider the defects of transference of the heavy metal in soil causes to be difficult to Accurate Prediction heavy metal content in soil compared with prior art.The present invention comprises the following steps:The acquisition and pretreatment of sample data;Construct the Elman Neural Network Prediction models based on storehouse autoencoder network;Unsupervised training is carried out to the Elman Neural Network Predictions model based on storehouse autoencoder network;Elman Neural Network Predictions model based on storehouse autoencoder network carries out Training;The forecast analysis of heavy metal content in soil.The present invention ensure that the precision that heavy metals in soil is predicted using heavy metal content in soil migration characteristic is taken into full account and modified Delphi approach model carries out analysis prediction.
Description
Technical field
The present invention relates to soil data analysis technical field, the soil specifically based on Elman neural network models
Content of beary metal Forecasting Methodology.
Background technology
Heavy metal pollution of soil problem has turned into the great agricultural ecological environment of China's extensive concern, to modern agriculture
Sustainable development, Agriculture Ecological Environment Security and agricultural product quality and safety with social economy constitute serious threat.More than ten years
Scientific research and substantial amounts of it was verified that due to the particularity of China's agroecological environment, indiscriminately imitate foreign technology and theory
Can not solve great environmental and problem in science that China's agriculture field is faced conscientiously, it is difficult to effectively contain agricultural environment pollution
Increasingly the developing state aggravated.
Domestic existing heavy metal-polluted soil detection method is mainly spot sampling, lab analysis, although the detection method energy
Accurate content of beary metal is obtained, but sample pre-treatments are relatively complicated, detection cycle is long, need to expend substantial amounts of manpower and materials,
It is likely to result in secondary pollution.Because China's area is big, made a concrete analysis of to be acquired great amount of samples, it is difficult to
Realize the quick detection of the automating of heavy metal in soil composition, intelligent, synthesization and precision.
LIBS method (LIBS) have almost without sample pretreatment, can simultaneously multiple element is divided
Analysis, the features such as detection time is short, suitable for the quick detection of heavy metal in soil.And found in actual analysis, it is substantial amounts of high
There is migration problem in the dimension heavy metal such as soil sample and arsenic in soil, cadmium, lead, if heavy metal can not be taken into full account in soil in time
Transference in earth, then it can not accurately realize the Accurate Prediction of heavy metal content in soil.
Therefore, how to realize that the preparation of heavy metal content in soil is predicted using LIBS technologies and have become the skill for being badly in need of solution
Art problem.
The content of the invention
The invention aims to solve not considering that transference of the heavy metal in soil causes difficulty in the prior art
A kind of the defects of with Accurate Prediction heavy metal content in soil, there is provided heavy metal content in soil based on Elman neural network models
Forecasting Methodology solves the above problems.
To achieve these goals, technical scheme is as follows:
Heavy metal content in soil Forecasting Methodology based on Elman neural network models, comprises the following steps:
The acquisition and pretreatment of sample data, the soil sample of collection is divided into training sample and test sample, utilized
The spectroscopic data of soil forms training data in LIBS technical limit spacing training samples, and training data is divided into label data and without mark
Sign data;
Construct the Elman Neural Network Prediction models based on storehouse autoencoder network;
Unsupervised training is carried out to the Elman Neural Network Predictions model based on storehouse autoencoder network, utilizes nothing
Label data carries out nothing to the storehouse autoencoder network of the Elman Neural Network Prediction models based on storehouse autoencoder network
Supervised training;
Elman Neural Network Predictions model based on storehouse autoencoder network carries out Training, will train sample
This label data is input in the storehouse autoencoder network after training and carries out feature extraction, will extract what is pre-processed after feature
Elman neural network prediction point of the historical data of the label data and heavy metal content in soil input based on storehouse autoencoder network
Model is analysed, Training is carried out to Elman Neural Network Predictions model;
The forecast analysis of heavy metal content in soil, by the LIBS light of the historical data of heavy metal content in soil and test sample
Input in the Elman Neural Network Prediction models based on storehouse autoencoder network, completed to test specimens after Spectrum data processing
The dynamic analysis prediction of this heavy metal content in soil.
The construction includes following step based on storehouse autoencoder network modified Delphi approach forecast analysis model
Suddenly:
Structure includes the storehouse autoencoder network of encoder and decoder, and storehouse autoencoder network includes multilayer neural network
Model, it is followed successively by one layer of input layer, multilayer hidden layer, one layer of output layer;
Elman neutral nets are built, Elman neutral nets include input layer, hidden layer successively, node layer is anti-as exporting
The undertaking layer of feedback and the output layer for possessing Dynamic Recurrent Neural Network model;
Using the input without label data as storehouse autoencoder network first layer input layer of training sample, storehouse own coding
Input of the output of network first tier input layer as second layer hidden layer, the output of last layer hidden layer are hidden as next layer
The input of layer, the input of the output of last layer of hidden layer as output layer, storehouse autoencoder network first layer input layer it is hidden
Layer unit is n1, second layer hidden layer Hidden unit be n2, third layer hidden layer Hidden unit be n3, X layer hidden layer
Hidden unit is nx, last layer of output layer Hidden unit be nt;
Using the sign output data of last layer of output layer of storehouse autoencoder network as Elman neural network input layers
Input;
Storehouse autoencoder network and Elman neutral nets are stacked to form the Elman neutral nets based on storehouse own coding
Model.
The described pair of Elman Neural Network Predictions model based on storehouse autoencoder network, which carries out unsupervised training, to be included
Following steps:
Obtain without label data X={ x1,x2,...,xr, by xi∈Rn×1I-th of input as storehouse autoencoder network
Vector, it is input to the encoder with d neuron;Encoder obtains encoding z by nonlinear activation function fi∈Rd×1, its
Calculation expression is as follows:
zi=sf(Wxi+b1),
Wherein, W ∈ Rd×nIt is weight matrix, b1∈Rd×1It is coding layer bias vector;
By ziDecoder is input to, obtains decoded resultIts calculation expression is as follows:
Wherein, sgFor the activation primitive of decoder,WT∈Rn×dIt is weight matrix W device, b2∈Rn×1It is
Decoding layer bias vector;
Reconstructed by minimizingWith original input data xi, between error cost carry out feature extraction, using under gradient
The weights of drop method renewal training between layersObtain extracted feature so thatBetween
Error reaches minimum value, and its cost function is as follows:
The Elman Neural Network Predictions model based on storehouse autoencoder network carry out Training include with
Lower step:
The label data of training sample is input in the storehouse autoencoder network after training and carries out feature extraction;
Obtain the historical data of heavy metal content in soil;
Elman Neural Network Prediction model definitions is as follows:
Wherein, k is the frequency of training of Elman neutral nets, and y is output vector, and x is hidden neuron output vector, and u is
Input vector, ycFor the previous moment output valve fed back of memory output layer unit, 0≤α≤1 be from connection feedback oscillator because
Son, W3、W2、W1Respectively hidden layer to output layer, accept layer to output layer, the connection weight matrix of input layer to hidden layer, f
() and g () are respectively to hide the Nonlinear Vector function that the excitation function of layer unit and output layer unit is formed;
The parameter of Elman Neural Network Prediction models is adjusted,
If the actual output of kth time is
Error extension function is
E (k) is calculated respectively to weights W3、W2、W1Partial derivative and be 0, it is pre- to obtain modified Delphi approach
Survey analysis model training method;
Connection weight W by E (k) to hidden layer to output layer3Local derviation is sought, it is calculated as follows:
Wherein,
Order
Connection weight W by E (k) to hidden layer to output layer2Local derviation is sought, it is calculated as follows:
Do not consider yc,i(k) it is rightDependence,
Connection weight W by E (k) to hidden layer to output layer1Local derviation is sought, it is calculated as follows:
Label data after extraction feature and the historical data of heavy metal content in soil are inputted into Elman neural network predictions
Analysis model carries out parameter adjustment, and Elman Neural Network Prediction models are trained.
The forecast analysis of the heavy metal content in soil comprises the following steps:
The LIBS spectroscopic datas of test sample are input to storehouse autoencoder network and carry out feature extraction;
Preprocessed data after extraction feature and the historical data of content of beary metal are input to and improve Elman nerve nets
In network model, the analysis prediction to test sample heavy metal content in soil is carried out.
Beneficial effect
The heavy metal content in soil Forecasting Methodology based on Elman neural network models of the present invention, compared with prior art
Using heavy metal content in soil migration characteristic is taken into full account and modified Delphi approach model carries out analysis prediction, ensure
The precision of heavy metals in soil prediction.
The present invention carries out feature extraction using storehouse autoencoder network to the soil spectrum data obtained by LIBS technologies;
Using the historical data of characteristic, heavy metal content in soil data and content of beary metal as modified Delphi approach
The input of model, the forecast model of heavy metal content in soil is constructed, it is special to take full advantage of the dynamic that Elman neutral nets have
Property, effectively solve the prediction challenge that heavy metal in soil content is migrated and changed over time.
When having too many characteristic variable input in order to avoid Elman neutral nets, the convergence of network is hindered, influences prediction essence
Degree, spy extract the feature of key using storehouse autoencoder network from substantial amounts of characteristic variable.In view of heavy metal in soil
Content is migrated and changed over time, to improve the accuracy of forecast model, utilizes modified Delphi approach dynamic model
Instead of traditional BP forecast models, Elman neural network dynamics model has been carried out based on history heavy metal in soil content
The element of data incorporates.The present invention has precision of prediction height, the fireballing feature of predicated response.
Brief description of the drawings
Fig. 1 is the method precedence diagram of the present invention.
Embodiment
The effect of to make to architectural feature of the invention and being reached, has a better understanding and awareness, to preferable
Embodiment and accompanying drawing coordinate detailed description, are described as follows:
As shown in figure 1, the heavy metal content in soil Forecasting Methodology of the present invention based on Elman neural network models,
Comprise the following steps:
The first step, the acquisition and pretreatment of sample data.The soil sample of collection is divided into training sample and test specimens
This, training data is formed using the spectroscopic data of soil in LIBS technical limit spacing training samples.Examined using traditional chemical method
The content of beary metal for determining training data is surveyed, label data is classified as, training data is divided into label data and without number of tags
According to.
After impurity elimination, air-drying and handle, grind, sieve, weigh, the process such as number, it is close to pack progress for the soil of collection
Envelope preserves;Cylindrical pedotheque in uniform thickness is fabricated to using tablet press machine;Pedotheque surface is acted on using laser,
Pedotheque on surface forms one layer of heating region after being excited, and with spectrometer collection plasma signal, obtains soil
The characteristic spectral line of various elements in sample, i.e. spectroscopic data are collected.
Second step, construct the Elman Neural Network Prediction models based on storehouse autoencoder network.
Here, considering dynamic characteristic specific to Elman neutral nets, preferably can contain applied to heavy metal-polluted soil
Amount migration, it is special that traditional BP neural network is replaced with Elman neutral nets.But in Elman neutral nets, if input feature vector
Variable is larger, then can influence convergence rate, or even influence prediction result.Therefore, it is self-editing that storehouse is added in Elman neutral nets
Code network so that the dimension-reduction treatment of storehouse autoencoder network is all passed through in the prediction of Elman neutral nets, to make full use of
While Elman neural network dynamic characteristics, it is pre- to it also avoid the excessive influence on the contrary of operand caused by Elman neutral nets
The problem of surveying result.It is comprised the following steps that:
(1) structure includes the storehouse autoencoder network of encoder and decoder, storehouse autoencoder network bag in a traditional way
Multilayer neural network model is included, is followed successively by one layer of input layer, multilayer hidden layer, one layer of output layer, as input layer, first layer is hidden
Hide layer, second layer hidden layer, third layer hidden layer, n-th layer hidden layer, the structure of output layer.
(2) Elman neutral nets are built, Elman neutral nets include input layer, hidden layer, as output layer section successively
The undertaking layer of point feedback and the output layer for possessing Dynamic Recurrent Neural Network model.
(3) using the input without label data as storehouse autoencoder network first layer input layer of training sample, the second layer
Input of the output of hidden layer as third layer hidden layer, the input exported as the 4th layer of hidden layer of third layer hidden layer,
By that analogy, input of the output of last layer of hidden layer as last layer of output layer.Storehouse autoencoder network first layer is defeated
The Hidden unit for entering layer is n1, second layer hidden layer Hidden unit be n2, third layer hidden layer Hidden unit be n3, X layer
The Hidden unit of hidden layer is nx, last layer of output layer Hidden unit be nt。
(4) inputted the sign output data of last layer of output layer of storehouse autoencoder network as Elman neutral nets
The input of layer, i.e. input of the feature after the extraction of storehouse autoencoder network as Elman neutral nets.
(5) storehouse autoencoder network and Elman neutral nets are stacked to form the Elman nerve nets based on storehouse own coding
Network model.
3rd step, unsupervised training is carried out to the Elman Neural Network Predictions model based on storehouse autoencoder network.
Utilize the storehouse autoencoder network without label data to the Elman Neural Network Prediction models based on storehouse autoencoder network
Carry out unsupervised training.By to high dimensional data carry out feature extraction to realize dimensionality reduction, so as to using feature new after dimensionality reduction come
Characterize initial data so that carry out data prediction before structure Elman models, reduce the complexity in time and space.It has
Body step is as follows:
(1) obtain without label data X={ x1,x2,...,xr, by xi∈Rn×1I-th as storehouse autoencoder network
Input vector, it is input to the encoder with d neuron.
Encoder obtains encoding z by nonlinear activation function fi∈Rd×1, its calculation expression is as follows:
zi=sf(Wxi+b1),
Wherein, W ∈ Rd×nIt is weight matrix, b1∈Rd×1It is coding layer bias vector.
(2) by ziDecoder is input to, obtains decoded resultIts calculation expression is as follows:
Wherein, sgFor the activation primitive of decoder,WT∈Rn×dIt is weight matrix W device, b2∈Rn×1It is
Decoding layer bias vector.
(3) reconstructed by minimizingWith original input data xi, between error cost carry out feature extraction, using ladder
Spend the weights of descent method renewal training between layersObtain extracted feature so thatIt
Between error reach minimum value, its cost function is as follows:
4th step, the Elman Neural Network Predictions model based on storehouse autoencoder network carry out Training.
The label data of training sample is input in the storehouse autoencoder network after training and carries out feature extraction, will have been carried
The input of the historical data of the label data pre-processed after feature and heavy metal content in soil is taken based on storehouse autoencoder network
Elman Neural Network Prediction models, Training is carried out to Elman Neural Network Predictions model.It is specifically walked
It is rapid as follows:
(1) label data of training sample is input in the storehouse autoencoder network after training and carries out feature extraction;
(2) historical data of heavy metal content in soil is obtained, here, the historical data of heavy metal content in soil is used to be directed to
The migration characteristic of heavy metal-polluted soil carries out analysis use, and the data have splendid approximation capability to the accuracy of data prediction.
(3) it is Elman Neural Network Prediction model definitions is as follows:
Wherein, k is the frequency of training of Elman neutral nets, and y is output vector, and x is hidden neuron output vector, and u is
Input vector, ycFor the previous moment output valve fed back of memory output layer unit, 0≤α≤1 be from connection feedback oscillator because
Son, W3、W2、W1Respectively hidden layer to output layer, accept layer to output layer, the connection weight matrix of input layer to hidden layer, f
() and g () are respectively to hide the Nonlinear Vector function that the excitation function of layer unit and output layer unit is formed.
For the migration characteristic of heavy metal-polluted soil, following traditional Elman Neural Network Predictions model is carried out
It is correspondingly improved,
The migration characteristic problem according to specific to heavy metal content in soil, by the x in above formulac(hide layer unit to be fed back
Previous moment output valve) replace with yc(the previous moment output valve that memory output layer unit is fed back), therefore, after improvement
The output vector y (k) of Elman neural network models is anti-dependent on hidden neuron output vector x and memory output layer unit institute
The previous moment output vector y of feedbackc。
(4) parameter of Elman Neural Network Prediction models is adjusted.Calculated and joined using traditional gradient descent method
Several methods, but because Elman Neural Network Prediction models are correspondingly improved, then it is obtained a result just different,
The derivation for new Elman Neural Network Predictions model parameter adjustment is provided herein.
If the actual output of kth time is
Error extension function is
E (k) is calculated respectively to weights W3、W2、W1Partial derivative and be 0, it is pre- to obtain modified Delphi approach
Survey analysis model training method.
A, the connection weight W by E (k) to hidden layer to output layer3Local derviation is sought, it is calculated as follows:
Wherein,
Order
B, the connection weight W by E (k) to hidden layer to output layer2Local derviation is sought, it is calculated as follows:
Do not consider yc,i(k) it is rightDependence,
C, the connection weight W by E (k) to hidden layer to output layer1Local derviation is sought, it is calculated as follows:
(5) label data after extraction feature and the historical data of heavy metal content in soil are inputted into Elman neutral nets
Forecast analysis model carries out parameter adjustment, Elman Neural Network Prediction models is trained so that Elman nerve nets
Network forecast analysis model has analysis predictive ability.
5th step, the forecast analysis of heavy metal content in soil.By the historical data of content of beary metal and test sample
The Elman Neural Network Prediction models based on storehouse autoencoder network are inputted after the processing of LIBS spectroscopic datas, are completed to surveying
The dynamic analysis prediction of this heavy metal content in soil of sample, so as to analyze the difference between actual result and prediction result, evaluation
The quality of improved model.It is comprised the following steps that:
(1) the LIBS spectroscopic datas of test sample are input to storehouse autoencoder network and carry out feature extraction.
(2) preprocessed data after extraction feature and the historical data of content of beary metal are input to and improve Elman god
Analysis through in network model, carrying out to test sample heavy metal content in soil is predicted.
This method is built in using Elman neural network models with reference to historical data to forecast analysis model, from
And existing soil sample is precisely predicted.Because the migration characteristic of heavy metal content in soil is to the dependence of historical data,
Using historical data as the part in model in Elman neural network models, the theoretical method of science and rigorous has been used
Logical thinking.
General principle, principal character and the advantages of the present invention of the present invention has been shown and described above.The technology of the industry
For personnel it should be appreciated that the present invention is not limited to the above embodiments, that described in above-described embodiment and specification is the present invention
Principle, various changes and modifications of the present invention are possible without departing from the spirit and scope of the present invention, these change and
Improvement is both fallen within the range of claimed invention.The protection domain of application claims by appended claims and its
Equivalent defines.
Claims (5)
1. a kind of heavy metal content in soil Forecasting Methodology based on Elman neural network models, it is characterised in that including following step
Suddenly:
11) acquisition and pretreatment of sample data, the soil sample of collection is divided into training sample and test sample, utilized
The spectroscopic data of soil forms training data in LIBS technical limit spacing training samples, and training data is divided into label data and without mark
Sign data;
12) the Elman Neural Network Prediction models based on storehouse autoencoder network are constructed;
13) unsupervised training is carried out to the Elman Neural Network Predictions model based on storehouse autoencoder network, using without mark
Data are signed to carry out without prison the storehouse autoencoder network of the Elman Neural Network Prediction models based on storehouse autoencoder network
Supervise and instruct white silk;
14) the Elman Neural Network Predictions model based on storehouse autoencoder network carries out Training, by training sample
Label data be input in the storehouse autoencoder network after training and carry out feature extraction, the mark that is pre-processed after feature will have been extracted
Sign Elman Neural Network Prediction of the historical data input based on storehouse autoencoder network of data and heavy metal content in soil
Model, Training is carried out to Elman Neural Network Predictions model;
15) forecast analysis of heavy metal content in soil, by the LIBS light of the historical data of heavy metal content in soil and test sample
Input in the Elman Neural Network Prediction models based on storehouse autoencoder network, completed to test specimens after Spectrum data processing
The dynamic analysis prediction of this heavy metal content in soil.
2. the heavy metal content in soil Forecasting Methodology according to claim 1 based on Elman neural network models, its feature
It is, the construction is comprised the following steps based on storehouse autoencoder network modified Delphi approach forecast analysis model:
21) structure includes the storehouse autoencoder network of encoder and decoder, and storehouse autoencoder network includes multilayer neural network
Model, it is followed successively by one layer of input layer, multilayer hidden layer, one layer of output layer;
22) Elman neutral nets are built, Elman neutral nets include input layer, hidden layer successively, node layer is anti-as exporting
The undertaking layer of feedback and the output layer for possessing Dynamic Recurrent Neural Network model;
23) using the input without label data as storehouse autoencoder network first layer input layer of training sample, storehouse own coding
Input of the output of network first tier input layer as second layer hidden layer, the output of last layer hidden layer are hidden as next layer
The input of layer, the input of the output of last layer of hidden layer as output layer, storehouse autoencoder network first layer input layer it is hidden
Layer unit is n1, second layer hidden layer Hidden unit be n2, third layer hidden layer Hidden unit be n3, X layer hidden layer
Hidden unit is nx, last layer of output layer Hidden unit be nt;
24) using the sign output data of last layer of output layer of storehouse autoencoder network as Elman neural network input layers
Input;
25) storehouse autoencoder network and Elman neutral nets are stacked to the Elman neutral net moulds to be formed based on storehouse own coding
Type.
3. the heavy metal content in soil Forecasting Methodology according to claim 1 based on Elman neural network models, its feature
Be, described pair of Elman Neural Network Predictions model based on storehouse autoencoder network carry out unsupervised training include with
Lower step:
31) obtain without label data X={ x1,x2,...,xr, by xi∈Rn×1I-th of input as storehouse autoencoder network
Vector, it is input to the encoder with d neuron;Encoder obtains encoding z by nonlinear activation function fi∈Rd×1, its
Calculation expression is as follows:
zi=sf(Wxi+b1),
Wherein, W ∈ Rd×nIt is weight matrix, b1∈Rd×1It is coding layer bias vector;
32) by ziDecoder is input to, obtains decoded resultIts calculation expression is as follows:
<mrow>
<msub>
<mover>
<mi>x</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mo>=</mo>
<msub>
<mi>s</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<mover>
<mi>W</mi>
<mo>^</mo>
</mover>
<msub>
<mi>z</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>b</mi>
<mn>2</mn>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Wherein, sgFor the activation primitive of decoder,WT∈Rn×dIt is weight matrix W device, b2∈Rn×1It is decoding
Layer bias vector;
33) reconstructed by minimizingWith original input data xi, between error cost carry out feature extraction, using under gradient
The weights of drop method renewal training between layersObtain extracted feature so thatBetween mistake
Difference reaches minimum value, and its cost function is as follows:
<mrow>
<mi>J</mi>
<mrow>
<mo>(</mo>
<mi>W</mi>
<mo>,</mo>
<mover>
<mi>W</mi>
<mo>^</mo>
</mover>
<mo>,</mo>
<msub>
<mi>b</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mn>2</mn>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msubsup>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</msubsup>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>s</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<mover>
<mi>W</mi>
<mo>^</mo>
</mover>
<msub>
<mi>s</mi>
<mi>f</mi>
</msub>
<mo>(</mo>
<mrow>
<msub>
<mi>Wx</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>b</mi>
<mn>1</mn>
</msub>
</mrow>
<mo>)</mo>
<mo>+</mo>
<msub>
<mi>b</mi>
<mn>2</mn>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>.</mo>
</mrow>
4. the heavy metal content in soil Forecasting Methodology according to claim 1 based on Elman neural network models, its feature
It is, the Elman Neural Network Predictions model based on storehouse autoencoder network carries out Training including following
Step:
41) label data of training sample is input in the storehouse autoencoder network after training and carries out feature extraction;
42) historical data of heavy metal content in soil is obtained;
43) it is Elman Neural Network Prediction model definitions is as follows:
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>y</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>g</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>W</mi>
<mn>3</mn>
</msup>
<mi>x</mi>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>+</mo>
<msup>
<mi>W</mi>
<mn>2</mn>
</msup>
<msub>
<mi>y</mi>
<mi>c</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>x</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>f</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>W</mi>
<mn>1</mn>
</msup>
<mi>u</mi>
<mo>(</mo>
<mrow>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>y</mi>
<mi>c</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>&alpha;y</mi>
<mi>c</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>y</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>,</mo>
</mrow>
Wherein, k is the frequency of training of Elman neutral nets, and y is output vector, and x is hidden neuron output vector, and u is input
Vector, ycThe previous moment output valve fed back for memory output layer unit, 0≤α≤1 are from the connection feedback oscillator factor, W3、
W2、W1Respectively hidden layer to output layer, accept layer to output layer, the connection weight matrix of input layer to hidden layer, f () and g
() is respectively to hide the Nonlinear Vector function that the excitation function of layer unit and output layer unit is formed;
44) parameter of Elman Neural Network Prediction models is adjusted,
If the actual output of kth time is
Error extension function is
E (k) is calculated respectively to weights W3、W2、W1Partial derivative and be 0, obtain modified Delphi approach prediction point
Analyse model training method;
441) the connection weight W by E (k) to hidden layer to output layer3Local derviation is sought, it is calculated as follows:
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>E</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mn>3</mn>
</msubsup>
</mrow>
</mfrac>
<mo>=</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>y</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mn>3</mn>
</msubsup>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>y</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<msubsup>
<mi>g</mi>
<mi>i</mi>
<mo>&prime;</mo>
</msubsup>
<mrow>
<mo>(</mo>
<mo>&CenterDot;</mo>
<mo>)</mo>
</mrow>
<mo>&lsqb;</mo>
<msub>
<mi>x</mi>
<mi>j</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>q</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>r</mi>
</munderover>
<msubsup>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mi>q</mi>
</mrow>
<mn>2</mn>
</msubsup>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>c</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mn>3</mn>
</msubsup>
</mrow>
</mfrac>
<mo>&rsqb;</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<msubsup>
<mi>&delta;</mi>
<mi>i</mi>
<mn>0</mn>
</msubsup>
<mo>&lsqb;</mo>
<msub>
<mi>x</mi>
<mi>j</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>q</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>r</mi>
</munderover>
<msubsup>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mi>q</mi>
</mrow>
<mn>2</mn>
</msubsup>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mrow>
<mi>c</mi>
<mo>,</mo>
<mi>i</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mn>3</mn>
</msubsup>
</mrow>
</mfrac>
<mo>&rsqb;</mo>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>m</mi>
<mo>,</mo>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>n</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
Wherein,
Order
442) the connection weight W by E (k) to hidden layer to output layer2Local derviation is sought, it is calculated as follows:
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>E</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>l</mi>
</mrow>
<mn>2</mn>
</msubsup>
</mrow>
</mfrac>
<mo>=</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>y</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>l</mi>
</mrow>
<mn>2</mn>
</msubsup>
</mrow>
</mfrac>
<mo>,</mo>
</mrow>
Do not consider yc,i(k) it is rightDependence,
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>l</mi>
</mrow>
<mn>2</mn>
</msubsup>
</mrow>
</mfrac>
<mo>=</mo>
<mi>&alpha;</mi>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>l</mi>
</mrow>
<mn>2</mn>
</msubsup>
</mrow>
</mfrac>
<mo>+</mo>
<msubsup>
<mi>g</mi>
<mi>i</mi>
<mo>&prime;</mo>
</msubsup>
<mrow>
<mo>(</mo>
<mo>&CenterDot;</mo>
<mo>)</mo>
</mrow>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>m</mi>
<mo>,</mo>
<mi>l</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>n</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
443) the connection weight W by E (k) to hidden layer to output layer1Local derviation is sought, it is calculated as follows:
<mrow>
<mtable>
<mtr>
<mtd>
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>E</mi>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mi>q</mi>
</mrow>
<mn>1</mn>
</msubsup>
</mrow>
</mfrac>
<mo>=</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>y</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<mfrac>
<mrow>
<mo>&part;</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mo>&part;</mo>
<msubsup>
<mi>W</mi>
<mrow>
<mi>j</mi>
<mi>q</mi>
</mrow>
<mn>1</mn>
</msubsup>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<mo>-</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mover>
<mi>y</mi>
<mo>^</mo>
</mover>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>(</mo>
<mi>k</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
<msubsup>
<mi>g</mi>
<mi>i</mi>
<mo>&prime;</mo>
</msubsup>
<mrow>
<mo>(</mo>
<mo>&CenterDot;</mo>
<mo>)</mo>
</mrow>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mn>3</mn>
</msubsup>
<msubsup>
<mi>f</mi>
<mi>j</mi>
<mo>&prime;</mo>
</msubsup>
<mrow>
<mo>(</mo>
<mo>&CenterDot;</mo>
<mo>)</mo>
</mrow>
<msub>
<mi>u</mi>
<mi>q</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<msubsup>
<mi>&delta;</mi>
<mi>i</mi>
<mn>0</mn>
</msubsup>
<msubsup>
<mi>W</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
<mn>3</mn>
</msubsup>
<msubsup>
<mi>f</mi>
<mi>j</mi>
<mo>&prime;</mo>
</msubsup>
<mrow>
<mo>(</mo>
<mo>&CenterDot;</mo>
<mo>)</mo>
</mrow>
<msub>
<mi>u</mi>
<mi>q</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>n</mi>
<mo>,</mo>
<mi>q</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mi>r</mi>
</mrow>
</mtd>
</mtr>
</mtable>
<mo>;</mo>
</mrow>
45) label data after extraction feature and the historical data of heavy metal content in soil are inputted into Elman neural network predictions
Analysis model carries out parameter adjustment, and Elman Neural Network Prediction models are trained.
5. the heavy metal content in soil Forecasting Methodology according to claim 1 based on Elman neural network models, its feature
It is, the forecast analysis of the heavy metal content in soil comprises the following steps:
51) the LIBS spectroscopic datas of test sample are input to storehouse autoencoder network and carry out feature extraction;
52) preprocessed data after extraction feature and the historical data of content of beary metal are input to and improve Elman nerve nets
In network model, the analysis prediction to test sample heavy metal content in soil is carried out.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711281809.6A CN107844870B (en) | 2017-12-07 | 2017-12-07 | Soil heavy metal content prediction method based on Elman neural network model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711281809.6A CN107844870B (en) | 2017-12-07 | 2017-12-07 | Soil heavy metal content prediction method based on Elman neural network model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107844870A true CN107844870A (en) | 2018-03-27 |
CN107844870B CN107844870B (en) | 2020-06-02 |
Family
ID=61663706
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711281809.6A Active CN107844870B (en) | 2017-12-07 | 2017-12-07 | Soil heavy metal content prediction method based on Elman neural network model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107844870B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108426812A (en) * | 2018-04-08 | 2018-08-21 | 浙江工业大学 | A kind of PM2.5 concentration value prediction techniques based on Memory Neural Networks |
CN108573105A (en) * | 2018-04-23 | 2018-09-25 | 浙江科技学院 | The method for building up of soil heavy metal content detection model based on depth confidence network |
CN109839825A (en) * | 2019-01-28 | 2019-06-04 | 华东交通大学 | A kind of forecast Control Algorithm and system of Rare-Earth Extraction Process constituent content |
CN112614552A (en) * | 2021-01-04 | 2021-04-06 | 武汉轻工大学 | BP neural network-based soil heavy metal content prediction method and system |
CN116679033A (en) * | 2023-06-07 | 2023-09-01 | 生态环境部华南环境科学研究所(生态环境部生态环境应急研究所) | Method and system for judging arsenic environmental risk of soil of industrial contaminated site |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060020563A1 (en) * | 2004-07-26 | 2006-01-26 | Coleman Christopher R | Supervised neural network for encoding continuous curves |
US20070239640A1 (en) * | 2001-10-22 | 2007-10-11 | Coppola Emery J Jr | Neural Network Based Predication and Optimization for Groundwater / Surface Water System |
CN106650212A (en) * | 2016-10-10 | 2017-05-10 | 重庆科技学院 | Intelligent plant breeding method and system based on data analysis |
CN107044976A (en) * | 2017-05-10 | 2017-08-15 | 中国科学院合肥物质科学研究院 | Heavy metal content in soil analyzing and predicting method based on LIBS Yu stack RBM depth learning technologies |
CN107179291A (en) * | 2017-05-10 | 2017-09-19 | 中国科学院合肥物质科学研究院 | Soil Heavy Metal Elements Content Forecasting Methodology based on tera-hertz spectra Yu depth autocoder |
-
2017
- 2017-12-07 CN CN201711281809.6A patent/CN107844870B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070239640A1 (en) * | 2001-10-22 | 2007-10-11 | Coppola Emery J Jr | Neural Network Based Predication and Optimization for Groundwater / Surface Water System |
US20060020563A1 (en) * | 2004-07-26 | 2006-01-26 | Coleman Christopher R | Supervised neural network for encoding continuous curves |
CN106650212A (en) * | 2016-10-10 | 2017-05-10 | 重庆科技学院 | Intelligent plant breeding method and system based on data analysis |
CN107044976A (en) * | 2017-05-10 | 2017-08-15 | 中国科学院合肥物质科学研究院 | Heavy metal content in soil analyzing and predicting method based on LIBS Yu stack RBM depth learning technologies |
CN107179291A (en) * | 2017-05-10 | 2017-09-19 | 中国科学院合肥物质科学研究院 | Soil Heavy Metal Elements Content Forecasting Methodology based on tera-hertz spectra Yu depth autocoder |
Non-Patent Citations (3)
Title |
---|
夏林: "基于全噪声自动编码器的深度神经网络优化算法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
王儒敬 等: "基于深度稀疏学习的土壤近红外光谱分析预测模型", 《发光学报》 * |
邵月红 等: "基于Elman动态神经网络的土壤墒情预测研究", 《水土保持通报》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108426812A (en) * | 2018-04-08 | 2018-08-21 | 浙江工业大学 | A kind of PM2.5 concentration value prediction techniques based on Memory Neural Networks |
CN108426812B (en) * | 2018-04-08 | 2020-07-31 | 浙江工业大学 | PM2.5 concentration value prediction method based on memory neural network |
CN108573105A (en) * | 2018-04-23 | 2018-09-25 | 浙江科技学院 | The method for building up of soil heavy metal content detection model based on depth confidence network |
CN109839825A (en) * | 2019-01-28 | 2019-06-04 | 华东交通大学 | A kind of forecast Control Algorithm and system of Rare-Earth Extraction Process constituent content |
CN109839825B (en) * | 2019-01-28 | 2022-04-12 | 华东交通大学 | Method and system for predictive control of component content in rare earth extraction process |
CN112614552A (en) * | 2021-01-04 | 2021-04-06 | 武汉轻工大学 | BP neural network-based soil heavy metal content prediction method and system |
CN112614552B (en) * | 2021-01-04 | 2022-06-07 | 武汉轻工大学 | BP neural network-based soil heavy metal content prediction method and system |
CN116679033A (en) * | 2023-06-07 | 2023-09-01 | 生态环境部华南环境科学研究所(生态环境部生态环境应急研究所) | Method and system for judging arsenic environmental risk of soil of industrial contaminated site |
CN116679033B (en) * | 2023-06-07 | 2024-01-23 | 生态环境部华南环境科学研究所(生态环境部生态环境应急研究所) | Method and system for judging arsenic environmental risk of soil of industrial contaminated site |
Also Published As
Publication number | Publication date |
---|---|
CN107844870B (en) | 2020-06-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107844870A (en) | Heavy metal content in soil Forecasting Methodology based on Elman neural network models | |
CN104408478B (en) | A kind of hyperspectral image classification method based on the sparse differentiation feature learning of layering | |
Manouchehrian et al. | Application of artificial neural networks and multivariate statistics to estimate UCS using textural characteristics | |
CN107179291B (en) | Method for predicting content of heavy metal elements in soil based on terahertz spectrum and depth automatic encoder | |
US20180032663A1 (en) | Method and apparatus for searching new material | |
CN108169745A (en) | A kind of borehole radar target identification method based on convolutional neural networks | |
CN107273490A (en) | A kind of combination mistake topic recommendation method of knowledge based collection of illustrative plates | |
CN107044976A (en) | Heavy metal content in soil analyzing and predicting method based on LIBS Yu stack RBM depth learning technologies | |
CN110309867B (en) | Mixed gas identification method based on convolutional neural network | |
CN108776831A (en) | A kind of complex industrial process Data Modeling Method based on dynamic convolutional neural networks | |
CN109975273A (en) | A kind of petrographic classification method based on laser induced breakdown spectroscopy | |
CN108710974A (en) | A kind of water body ammonia nitrogen prediction technique and device based on depth confidence network | |
CN108596274A (en) | Image classification method based on convolutional neural networks | |
CN106127242A (en) | Year of based on integrated study Extreme Precipitation prognoses system and Forecasting Methodology thereof | |
CN108875118B (en) | Method and device for evaluating accuracy of prediction model of silicon content of blast furnace molten iron | |
CN102158486A (en) | Method for rapidly detecting network invasion | |
CN102902981A (en) | Violent video detection method based on slow characteristic analysis | |
CN103967478A (en) | Method for identifying vertical well flow patterns based on conducting probe | |
CN108229062A (en) | Method based on sensibility micro-parameter prediction cemented fill mechanical response characteristic | |
CN105715253A (en) | Prediction method for flowing bottomhole pressure of gas well | |
CN107528824A (en) | A kind of depth belief network intrusion detection method based on two-dimensionses rarefaction | |
CN109145685A (en) | Fruits and vegetables EO-1 hyperion quality detecting method based on integrated study | |
CN107133879A (en) | A kind of method for screening similar oil field | |
CN103186696A (en) | Auxiliary variable simplification method for high-dimensional nonlinear soft sensor model | |
CN106295667B (en) | A kind of method and its application selecting optimal spectrum based on genetic algorithm |
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 |