CN109726857A - Cyanobacterial bloom prediction technique based on GA-Elman network - Google Patents
Cyanobacterial bloom prediction technique based on GA-Elman network Download PDFInfo
- Publication number
- CN109726857A CN109726857A CN201811553870.6A CN201811553870A CN109726857A CN 109726857 A CN109726857 A CN 109726857A CN 201811553870 A CN201811553870 A CN 201811553870A CN 109726857 A CN109726857 A CN 109726857A
- Authority
- CN
- China
- Prior art keywords
- layer
- output
- cyanobacterial bloom
- elman
- prediction
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of cyanobacterial bloom prediction techniques based on GA-Elman network, are related to cyanobacterial bloom electric powder prediction in water environment.The present invention analyzes the related causes such as the environmental factor of algal bloom, physical chemical factor and biological factor, obtaining, which influences the principal element index that cyanobacterial bloom occurs, temperature, humidity, dissolved oxygen, ammonia-nitrogen content, pH value and chlorophyll a etc., by carrying out analysis modeling to Multiple factors attributes, modified Delphi approach is combined with multifactor time series analysis, construct GA-Elman Network Prediction Model, this method can effectively improve the precision of prediction of cyanobacterial bloom, compared to traditional Elman neural network, precision of prediction improves 5%~10%.
Description
Technical field
The present invention relates to cyanobacterial bloom electric powder prediction in water environment, in particular to the generation process of cyanobacterial bloom into
A kind of method of raising cyanobacterial bloom precision of prediction of row analysis and the multifactor GA-Elman neural network prediction model established.
Background technique
Good water quality condition is primary condition for the survival of mankind, but with the development of the social economy, quality problem
Increasingly severe, algal bloom disaster caused by wherein water eutrophication is becomes the outstanding problem of water environment pollution.Water body
Eutrophication is mainly to cause a large amount of since the nutritive salt of the elements such as excessive nitrogenous, phosphorus enters among river and lake by all means
Algae or the aggravation of certain aquatic plants growths, break the ecological balance of water environment, water quality deterioration, the ecosystem are caused to be destroyed
A series of ecological environment problems.Since river water is a complicated open system, by having in wind direction, water flow and water body
The influence of the factors such as machine object is bigger, and algal bloom has sudden, time variation etc., existing cyanobacterial bloom prediction technique
It is unable to meet the requirement of precision of prediction.
The difficulty that accurately prediction has great importance with early warning and water environment protection faces is carried out to cyanobacterial bloom
Topic.At present in the research of cyanobacterial bloom prediction, according to the formation mechenism of wawter bloom, mainly there is following two categories research method.It is a kind of
It is to study the Forming Mechanism of cyanobacterial bloom using the ecological mechanism of wawter bloom growth as theoretical basis, the Ecological Changes of wawter bloom is carried out
It simulates and predicts.But specific to different types of water body, since environment is different, the key influence factor that wawter bloom is formed is different.It is early
It is that dynamics of ecosystem model is established on the basis based on algae mechanism study mostly that phase, which studies wawter bloom, according to rivers and lakes
Ecologic structure, function, the influence of spatial-temporal evolution pattern and physical and chemical process to the ecosystem studies the mutual of internal system
The differential equation is established in effect, tracks ecosystem state change.Modelling by mechanism is divided into again according to algal grown mechanism model and floats
Swim biological time space distribution model.Another kind is to establish system model using the method for artificial intelligence, is predicted.In recent years
Come, intelligent modeling method does not need to carry out extremely in-depth study, nonlinear system to mechanism due to the intelligence of its information processing
The features such as modeling ability of uniting is strong, many advanced intellectual technologies are widely applied in various Nonlinear Modeling fields, using intelligence
Technique study algal bloom formation mechenism and forecasting problem become current one of main trend.
Model parameter needed for intelligent modeling method is few, and calculating process is fast and convenient, and especially nerual network technique has strong
Big self-learning ability and Nonlinear Processing ability, in conjunction with neural network combination intelligent model in practical projects by favor,
The model can be reacted in wawter bloom forming process well in interactively and its non-linear, the uncertain spy of forming process
Sign, is the important directions of following wawter bloom formation mechenism and forecasting research.
Summary of the invention
In order to improve the precision of prediction of cyanobacterial bloom, the present invention to the environmental factor of algal bloom, physical chemical factor and
The related causes such as biological factor are analyzed, and obtaining, which influences the principal element index that cyanobacterial bloom occurs, temperature, humidity, dissolution
Oxygen, ammonia-nitrogen content, pH value and chlorophyll a etc., by carrying out analysis modeling to Multiple factors attributes, by improved Elman nerve net
Network is combined with multifactor time series analysis, constructs GA-Elman Network Prediction Model, and this method can effectively improve indigo plant
The precision of prediction of algae wawter bloom.
The present invention analyzes the pests occurrence rule of cyanobacterial bloom, establishes GA- for the index factor data of the cyanobacterial bloom of acquisition
The cyanobacterial bloom prediction and warning model of Elman network, and carry out system design.Research work is broadly divided at cyanobacterial bloom data
Reason and index analysis, cyanobacterial bloom prediction and warning technique study and build river and lake monitoring water environment and wawter bloom Early-Warning System three
A part.Specifically, the cyanobacterial bloom prediction technique based on GA-Elman network includes the following steps:
Step 1: cyanobacterial bloom data processing and index analysis for polynary time series data, determining influences cyanobacterial bloom
Index factor.
Step 2: being optimized using GA genetic algorithm to cyanobacterial bloom prediction Elman neural network model parameter.
Step 3: taking error compensating method, optimize GA-Elman neural network model, improves Elman neural network mould
Type precision of prediction.
The present invention has the advantages that
(1) present invention is analyzed using Multiple factors attributes, is made when establishing the temporal model of cyanobacterial bloom influence factor
Model adequately according to current time and historical juncture cyanobacterial bloom factor data, analyzes the concentration numbers of future time instance chlorophyll a
According to, and then analysis prediction effectively is carried out to complicated cyanobacterial bloom.
(2) present invention utilize Elman neural network cyanobacterial bloom prediction model, Elman nerve net in addition to input layer,
Except hidden layer and output layer, it is added to undertaking layer, is used to receive feedback signal from hidden layer, remembers hidden layer neuron history
The output of data, and then Elman neural network is improved to the sensibility of historical data, enhance network processes multidate information
The problem of ability improves the stability of cyanobacterial bloom model system, reduces prediction error and is more advantageous to processing time series data.
(3) present invention improves traditional Elman network, using GA genetic algorithm preliminary Optimized model parameter, and
Error compensation is carried out to the prediction result of model and advanced optimizes model output parameters, association's note of enhancing Elman network model
The characteristics of recalling function, the effective stability and precision of prediction for improving model system.
(4) because artificial neural network algorithm is simple and efficient, the present invention uses time series data as network model
Input data carries out the characterization factor of future time by the chlorophyll concentration and influence factor of current time and historical juncture
Prediction.And GA genetic algorithm and error compensating method is respectively adopted, better Optimal Parameters, compared to traditional Elman nerve
Network, precision of prediction improve 5%~10%.
Detailed description of the invention
Fig. 1 is cyanobacterial bloom prediction technique flow chart provided by the invention;
Fig. 2 is the optimization process flow chart that GA genetic algorithm is used in the present invention;
Fig. 3 is the GA-Elman neural network error fit optimization process schematic diagram in the present invention.
Specific embodiment
The present invention is described in detail with reference to the accompanying drawing.
The present invention proposes a kind of cyanobacteria based on GA-Elman network for the index factor data of the cyanobacterial bloom of acquisition
Wawter bloom prediction technique analyzes the pests occurrence rule of cyanobacterial bloom, establishes the cyanobacterial bloom prediction and warning model of GA-Elman network, and
Carry out system design.Research work is broadly divided into cyanobacterial bloom data processing and index analysis, cyanobacterial bloom prediction and warning method
Study and build river and lake monitoring water environment and three parts of wawter bloom Early-Warning System.
Step 1: determining the index factor for influencing cyanobacterial bloom.
Correlation analysis is carried out to each index factor data of cyanobacterial bloom.By analyzing the multiple of the cyanobacterial bloom acquired
Correlation between index factor, and using chlorophyll-a concentration value as the characteristic index of cyanobacterial bloom occurrence degree, it analyzes each
The linear dependence of index factor and chlorophyll a is selected the efficiency index factor that the cyanobacterial bloom in studied waters occurs, is rejected
Invalid index and the index factor small to research waters influence degree, avoid it from influencing the interference of cyanobacterial bloom prediction model
The accuracy of cyanobacterial bloom prediction model.Finally, being calculated in cyanobacterial bloom index factor, with factor analysis to chlorophyll a shadow
Loud importance size, so that selecting influences four big index factors, the effective precision for improving cyanobacterial bloom prediction model.
Step 2: being optimized using GA genetic algorithm to cyanobacterial bloom prediction Elman neural network model parameter.
As shown in Fig. 2, tradition GA genetic algorithm generates initial population by coding.By the weight of Elman neural network and
Threshold value connects in sequence forms a real number array, a chromosome as GA genetic algorithm.Evolutionary search be with
Adaptation density function is foundation, is searched for using the fitness value of chromosome each in population, executes genetic manipulation.Calculate current kind
The fitness value of each chromosome in group finds out adaptive optimal control degree individual and iterates (selection intersects and makes a variation) until meeting
Termination condition exports optimal solution;If termination condition is not achieved with maximum genetic iteration number be terminate calculation criterion.
Elman neural network is as shown in figure 3, wherein w1To accept connection weight (weight) of the layer to hidden layer, w2For input
Layer arrives the connection weight of hidden layer, w3For the connection weight of hidden layer to output layer, f () is the transmitting letter of hidden layer neuron
Number, g () are the transmission function of output layer neuron.If Elman neural network input vector is x (k), output vector y
(k), hidden layer output vector is h (k), and undertaking layer state feedback vector is z (k), and k indicates the moment.Then Elman neural network is defeated
Enter output relation are as follows:
Y (k)=g [w3h(k)] (1)
H (k)=f [w1z(k)+w2x(k-1)] (2)
Z (k)=h (k-1)=f [w1z(k-1)+w2x(k-2)] (3)
Step 3: using error compensating method, output to GA-Elman neural network model is carried out further excellent
Change.
The present invention uses for reference the advantages of neural network, the characteristics of according to multitiered network structure and error propagation, GA-Elman mind
The Elman neural network of the first layer building GA genetic algorithm optimization through network, the second layer on this basis to error information into
Row Multi-source Information Fusion and error compensation.
(1) network structure design in, by GA-Elman neural network from laterally be divided into two layers, first layer by input layer,
Hidden layer and intermediate output layer composition.Input vector is cyanobacterial bloom index factor data X=(x1,x2,…,xn)T, hidden layer
Vector is made of input vector x (k-1) and undertaking layer vector z (k), and intermediate output layer output vector is chlorophyll-a concentration
Y=(y1,y2,…,ym)T, hidden layer output vector is H=(h1,h2,…,hn)T, n is in input layer and hidden layer
The number of neuron;M is the number of neuron in intermediate output layer.Accept layer to weight between hidden layer be w1, input layer arrives
Weight is w between hidden layer2, hidden layer to weight between intermediate output layer is w3, according to formula (1) (2) (3) it follows that
For hidden layer, have:
Wherein, hi(k) output at i-th of neuron k moment of hidden layer, z are indicatedi(k) it indicates to accept i-th of neuron k of layer
The output at moment, xi(k-1) input at i-th of neuron k-1 moment of input layer is indicated.
Have for intermediate output layer:
Wherein, yj(k) output at intermediate j-th of neuron k moment of output layer, j=1,2 ..., m are indicated;
(2) in network structure design, such as Fig. 3, the second layer is final output layer O=(o1,o2,…,os)T, to intermediate defeated
Layer carries out Multi-source Information Fusion and error compensation out, and s is final output layer vector number;If desired output vector D=(d1,
d2,…,ds)T, intermediate output layer to the weight between final output layer is v.Intermediate output layer number of nodes m is equal to monitoring point
Number.It on the other hand, can setting according to training dataset and two kinds of situations of test data set in processing of the network to data set
It is fixed, the corresponding two-way propagation process for generating error.
For final output layer, have:
Wherein, oqIt (k) is the output at q-th of neuron k moment of final output layer.
(3) error back propagation learns;
When the reality output of network model is not equal to truthful data, that is, when there is deviation, output error E are as follows:
When above formula is further spread out, have:
It can be seen from the above error is the function of each layer weight, adjustment weight can change error, it is clear that adjustment weight
Principle be exactly to reduce weight error constantly, can be obtained by formula (5) (6) (7) (8):
Wherein, η refers to learning rate, is a given constant, 0 < η < 1.P=1,2,3;Δwp, Δ v be each layer weight
Error.
After finding out the new weight of each layer, training GA-Elman cyanobacterial bloom prediction model, input layer is by water temperature, PH
Value, ammonia-nitrogen content, the current time of five influence factors of dissolved oxygen and chlorophyll-a concentration and historical juncture data composition, it is defeated
Layer is made of the future time instance of chlorophyll concentration out, i.e., to one step of concentration forward prediction of chlorophyll.If defeated after error compensation
Outgoing vector C=(c1,c2,…,ct)T, t is the number of output vector, it can be deduced that:
Wherein, r=1,2 ... t.
Claims (3)
1. the cyanobacterial bloom prediction technique based on GA-Elman network, it is characterised in that: include the following steps,
Step 1: cyanobacterial bloom data processing and index analysis for polynary time series data, determine the finger for influencing cyanobacterial bloom
Mark factor;
Step 2: being optimized using GA genetic algorithm to cyanobacterial bloom prediction Elman neural network model parameter;
Step 3: taking error compensating method, optimize GA-Elman neural network model, it is pre- to improve Elman neural network model
Survey precision.
2. the cyanobacterial bloom prediction technique according to claim 1 based on GA-Elman network, it is characterised in that: third step
Described in GA-Elman neural network model, from being laterally divided into two layers, first layer is by input layer, hidden layer and intermediate output layer
Composition, input vector are cyanobacterial bloom index factor data X=(x1,x2,…,xn)T, hidden layer vector is by input vector x (k-
1) and layer vector z (k) composition is accepted, intermediate output layer output vector is chlorophyll-a concentration Y=(y1,y2,…,ym)T, hidden layer
Output vector is H=(h1,h2,…,hn)T, n is the number of neuron in input layer and hidden layer;M is mind in intermediate output layer
Number through member;Accept layer to weight between hidden layer be w1, input layer to weight between hidden layer is w2, hidden layer to centre
Weight is w between output layer3;For hidden layer, have:
Wherein, hi(k) output at i-th of neuron k moment of hidden layer, z are indicatedi(k) it indicates to accept i-th of the neuron k moment of layer
Output, xi(k-1) input at i-th of neuron k-1 moment of input layer is indicated;
Have for intermediate output layer:
Wherein, yj(k) output at intermediate j-th of neuron k moment of output layer, j=1,2 ..., m are indicated;
The second layer is final output layer O=(o1,o2,…,os)T, Multi-source Information Fusion and error compensation are carried out to intermediate output layer,
S is final output layer vector number;If desired output vector D=(d1,d2,…,ds)T, intermediate output layer to final output layer it
Between weight be v;Intermediate output layer number of nodes m is equal to monitoring points;
For final output layer, have:
Wherein, oqIt (k) is the output at q-th of neuron k moment of final output layer;
(3) error back propagation learns;
When the reality output of GA-Elman neural network model is not equal to truthful data, that is, when there is deviation, output error E are as follows:
When above formula is further spread out, have:
It is obtained by formula (5) (6) (7) (8):
Wherein, η refers to learning rate, is a given constant, 0 < η < 1;P=1,2,3;Δwp, Δ v be each layer weight error;
After finding out the new weight of each layer, training GA-Elman cyanobacterial bloom prediction model, input layer is by water temperature, pH value, ammonia
Nitrogen content, the current time of five influence factors of dissolved oxygen and chlorophyll-a concentration and historical juncture data composition, output layer by
The future time instance of chlorophyll concentration forms, i.e., to one step of concentration forward prediction of chlorophyll;If the output vector C after error compensation
=(c1,c2,…,ct)T, t is the number of output vector, it obtains:
Wherein, r=1,2 ... t.
3. the cyanobacterial bloom prediction technique according to claim 1 based on GA-Elman network, it is characterised in that: the first step
Described in index factor be water temperature, pH value, ammonia-nitrogen content, dissolved oxygen and chlorophyll-a concentration.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811553870.6A CN109726857A (en) | 2018-12-19 | 2018-12-19 | Cyanobacterial bloom prediction technique based on GA-Elman network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811553870.6A CN109726857A (en) | 2018-12-19 | 2018-12-19 | Cyanobacterial bloom prediction technique based on GA-Elman network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109726857A true CN109726857A (en) | 2019-05-07 |
Family
ID=66296234
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811553870.6A Pending CN109726857A (en) | 2018-12-19 | 2018-12-19 | Cyanobacterial bloom prediction technique based on GA-Elman network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109726857A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113011397A (en) * | 2021-04-27 | 2021-06-22 | 北京工商大学 | Multi-factor cyanobacterial bloom prediction method based on remote sensing image 4D-FractalNet |
CN114201923A (en) * | 2022-02-16 | 2022-03-18 | 中国地质调查局水文地质环境地质调查中心 | Processing method and device of landslide displacement prediction model |
CN114821285A (en) * | 2022-04-15 | 2022-07-29 | 北京工商大学 | System and method for predicting cyanobacterial bloom based on ACONV-LSTM and New-GANs combination |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103984996B (en) * | 2014-05-26 | 2017-02-15 | 北京工商大学 | Optimization prediction method for taboo searching algorithm and genetic algorithm of algal bloom mechanism time varying model |
CN107729988A (en) * | 2017-09-30 | 2018-02-23 | 北京工商大学 | Blue-green alga bloom Forecasting Methodology based on dynamic depth confidence network |
-
2018
- 2018-12-19 CN CN201811553870.6A patent/CN109726857A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103984996B (en) * | 2014-05-26 | 2017-02-15 | 北京工商大学 | Optimization prediction method for taboo searching algorithm and genetic algorithm of algal bloom mechanism time varying model |
CN107729988A (en) * | 2017-09-30 | 2018-02-23 | 北京工商大学 | Blue-green alga bloom Forecasting Methodology based on dynamic depth confidence network |
Non-Patent Citations (6)
Title |
---|
ZAIWEN LIU: "Prediction technique for water-bloom in lakes based on elman network", 《 2009 IEEE INTERNATIONAL CONFERENCE ON AUTOMATION AND LOGISTICS》 * |
吴志伟: "嵌入式电熔镁炉智能控制系统研究", 《中国博士学位论文全文数据库 工程科技Ⅰ辑》 * |
曹芙: "基于Elman神经网络及优化算法的混合模型的研究及应用", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
李梦寻: "供水水源地水质实时监测与智能化预警方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》 * |
梅贵琴: "改进的Elman神经网络和网络参数优化算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
黄浙丰: "基于时序神经网络的藻类水华预测模型研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113011397A (en) * | 2021-04-27 | 2021-06-22 | 北京工商大学 | Multi-factor cyanobacterial bloom prediction method based on remote sensing image 4D-FractalNet |
CN113011397B (en) * | 2021-04-27 | 2024-03-29 | 北京工商大学 | Multi-factor cyanobacterial bloom prediction method based on remote sensing image 4D-Fractalnet |
CN114201923A (en) * | 2022-02-16 | 2022-03-18 | 中国地质调查局水文地质环境地质调查中心 | Processing method and device of landslide displacement prediction model |
CN114821285A (en) * | 2022-04-15 | 2022-07-29 | 北京工商大学 | System and method for predicting cyanobacterial bloom based on ACONV-LSTM and New-GANs combination |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Recknagel | ANNA–Artificial Neural Network model for predicting species abundance and succession of blue-green algae | |
Muttil et al. | Machine-learning paradigms for selecting ecologically significant input variables | |
CN106529818B (en) | Water quality assessment Forecasting Methodology based on Fuzzy Wavelet Network | |
CN109726857A (en) | Cyanobacterial bloom prediction technique based on GA-Elman network | |
CN101796928B (en) | Method for predicting effect of water quality parameters of aquaculture water on growth conditions of aquaculture living beings | |
CN109241313A (en) | A kind of image search method based on the study of high-order depth Hash | |
Liu et al. | Prediction of dissolved oxygen content in aquaculture of Hyriopsis cumingii using Elman neural network | |
CN107506857A (en) | Urban lake storehouse blue-green alga bloom multi variant based on fuzzy support vector machine | |
Li et al. | Prediction and elucidation of the population dynamics of Microcystis spp. in Lake Dianchi (China) by means of artificial neural networks | |
CN109978024B (en) | Effluent BOD prediction method based on interconnected modular neural network | |
Recknagel et al. | Modelling and prediction of phyto‐and zooplankton dynamics in Lake Kasumigaura by artificial neural networks | |
CN115795367A (en) | Algal bloom outbreak prediction method based on machine learning and application | |
CN110119165A (en) | A kind of aquaculture pond dissolved oxygen detection device | |
Miao et al. | A hybrid neural network and genetic algorithm model for predicting dissolved oxygen in an aquaculture pond | |
Palani et al. | Development of a neural network model for dissolved oxygen in seawater | |
CN115828758B (en) | Seawater three-dimensional prediction method and system based on improved firework algorithm optimization network | |
CN108520348B (en) | Ecological index prediction method based on mangrove forest ecological big data | |
Deng et al. | Application of neural network based on PSO algorithm in prediction model for dissolved oxygen in fishpond | |
He et al. | Improving BP neural network for the recognition of face direction | |
Altunkaynak et al. | A comparative study of hydrodynamic model and expert system related models for prediction of total suspended solids concentrations in Apalachicola Bay | |
CN114782740A (en) | Remote sensing water quality monitoring method combining genetic optimization and extreme gradient promotion | |
CN106959360B (en) | The rare-earth mining area farmland water pH value flexible measurement method to develop using backward difference | |
CN112906285A (en) | Method, system and equipment for power reconstruction of river and lake water system and computer storage medium | |
Li et al. | Dissolved oxygen prediction in Apostichopus japonicus aquaculture ponds by BP neural network and AR model | |
Wang et al. | Prediction method of cyanobacterial blooms spatial-temporal sequence based on deep belief network and fuzzy expert system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190507 |