CN106407691B - A kind of red tide plankton amount prediction technique and device - Google Patents
A kind of red tide plankton amount prediction technique and device Download PDFInfo
- Publication number
- CN106407691B CN106407691B CN201610859375.2A CN201610859375A CN106407691B CN 106407691 B CN106407691 B CN 106407691B CN 201610859375 A CN201610859375 A CN 201610859375A CN 106407691 B CN106407691 B CN 106407691B
- Authority
- CN
- China
- Prior art keywords
- group
- factor
- red tide
- marine environment
- tide plankton
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
- G16Z99/00—Subject matter not provided for in other main groups of this subclass
Abstract
The invention discloses a kind of red tide plankton amount prediction technique and devices, applied to environmental monitoring field, it include: to obtain the effective marine environment factor of M group, one group of effective marine environment factor includes same kind of effective marine environment factor that multiframe acquires in different time points；Pretreatment is normalized to the effective marine environment factor of M group, to obtain the marine environment factor after corresponding M group normalization；The marine environment factor after the m group normalization in the marine environment factor after the normalization of M group is input in corresponding ARIMA model respectively and is predicted, to predict M group marine environment factor predicted value；M group marine environment factor predicted value is input to BP neural network prediction model simultaneously to predict, to predict the red tide plankton amount predicted value in preset period.The present invention realizes the short-term Accurate Prediction of red tide plankton amount, whether red tide occurs within several days with forecast future, to be guarded against.
Description
Technical field
The present invention relates to environmental monitoring field more particularly to a kind of red tide plankton amount prediction technique and device.
Background technique
Red tide be it is a kind of sharply bred since the planktonic organism in local sea area is sudden, and flock together and cause Hai Ping
There is abnormal and smelly phenomenon in face color.Red tide is the knot that the combined factors such as biology, chemistry, the hydrology, meteorology influence
Fruit.There is complicated non-linear relation between the growth of red tide plankton, breeding, descending process and environmental factor, and these rings
Heterogeneity of the border factor with temporal continuity and spatially.Immediate offshore area frequent occurrence in recent years, the outburst of red tide
It is unbalance to not only result in Water-environment Ecological System, also will cause aquaculture disaster.
Traditional red tide prediction method is current red tide situation to be obtained by each environmental factor, but work as the quick-fried of red tide
It has resulted in that Water-environment Ecological System is unbalance and aquaculture disaster when hair, therefore red tide can not be predicted in advance.
Summary of the invention
For the embodiment of the present invention by providing a kind of red tide plankton amount prediction technique and device, solving can not be in advance to red tide
The technical issues of being predicted.
In a first aspect, the embodiment of the invention provides a kind of red tide plankton amount prediction techniques, comprising:
Obtain the effective marine environment factor of M group, wherein M is the species number of effective marine environment factor multiplied by sub- sea
The number in domain, effective marine environment factor described in same group include same kind of effective sea that multiframe acquires in different time points
Foreign environmental factor；
Pretreatment is normalized to the effective marine environment factor of the M group, to obtain ocean after corresponding M group normalization
Environmental factor；
The marine environment factor after the m group normalization in the marine environment factor after M group normalization is input to respectively
It is predicted in corresponding ARIMA model, to predict M group marine environment factor predicted value, m is followed successively by 1 to M, wherein one
The marine environment factor after the corresponding one group of normalization of ARIMA model；
The M group marine environment factor predicted value is input to BP neural network prediction model simultaneously to predict, with pre-
Measure the red tide plankton amount predicted value in preset period.
Preferably, the effective marine environment of the acquisition M group because, comprising:
Establish the relational database of primitive ocean environmental factor, wherein multiple go through is preserved in the relational database
The primitive ocean environmental factor of history monitoring, and preserve the monitoring location information and prison of each primitive ocean environmental factor
Survey temporal information；
According to the monitoring time information of each primitive ocean environmental factor and monitoring location information, from the relationship number
According to extraction M group primitive ocean environmental factor in library, wherein same group of primitive ocean in the M group primitive ocean environmental factor
Environmental factor monitors in same sub- sea area, and apart from current time less than prefixed time interval；
It is filtered out from every group of primitive ocean environmental factor in the M group primitive ocean environmental factor and meets preset condition
Factor set become the effective marine environment factor of the M group.
Preferably, it is described respectively by M group normalization after m group in the marine environment factor normalize after ocean ring
The border factor is input to predicted in corresponding ARIMA model before, the method also includes: as follows establish described in
ARIMA model:
Believed according to the monitoring location information of the primitive ocean environmental factor each in the relational database, monitoring time
Breath determines the ARIMA model for every group of primitive ocean environmental factor in the M group primitive ocean environmental factor；
The ARIMA model for being directed to every group of primitive ocean environmental factor respectively based on AIC criterion carries out parameter Estimation；
Parameter combination when obtaining minimum AIC value based on AIC criterion progress parameter Estimation is chosen, as described
The model parameter of the ARIMA model of every group of primitive ocean environmental factor.
Preferably, the M group marine environment factor predicted value is input to BP neural network prediction model simultaneously described
Before being predicted, the method also includes: the BP neural network prediction model is established as follows:
Obtain red tide plankton measurement sample；
The red tide plankton measurement sample is normalized to obtain corresponding return based on following logarithmic formula
Red tide plankton amount sample after one change:
Wherein, y ' is the value of red tide plankton amount after normalization, and y is red tide plankton measurement, ymaxFor the red tide plankton
The maximum value of measurement sample, yminFor the minimum value of the red tide plankton measurement sample；
The implicit number of plies of BP neural network is determined based on red tide plankton amount sample after the normalization and each is implicit
The number of nodes of layer；
The number of nodes of the implicit number of plies and each hidden layer based on determining BP neural network is established for the M group sea
The optimum network structure of the BP neural network of foreign environmental factor predicted value；
The optimal value that the optimum network structure is obtained by learning process is as the BP neural network prediction model.
Preferably, after the red tide plankton amount predicted value predicted in preset period, the method also includes:
The red tide plankton amount predicted value is compared with the red tide plankton measurement that the corresponding time measures；
It exports equal between the red tide plankton amount predicted value and the red tide plankton measurement of the corresponding time measurement
At least one of square error, mean absolute error, average absolute percentage error, degree of fitting comparing result.
Preferably, the type of effective marine environment factor includes: pH value, dissolved oxygen, water temperature, dissolution oxygen saturation
Degree, chlorophyll-a, phosphate, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, salinity, chemical oxygen consumption (COC) and silicate.
Second aspect, the embodiment of the invention provides a kind of red tide plankton amount prediction meanss, comprising:
Obtaining unit, for obtaining the effective marine environment factor of M group, wherein M is the kind of effective marine environment factor
Class number multiplied by sub- sea area number, effective marine environment factor described in same group include multiframe acquire in different time points it is same
Effective marine environment factor of type；
Pretreatment unit, it is corresponding to obtain for pretreatment to be normalized to the effective marine environment factor of the M group
The marine environment factor after the normalization of M group；
Environmental factor predicting unit, for respectively normalizing the m group in the marine environment factor after M group normalization
The marine environment factor is input in corresponding ARIMA model and is predicted afterwards, to predict M group marine environment factor predicted value, m
1 is followed successively by M, wherein the marine environment factor after the corresponding one group of normalization of an ARIMA model；
Red tide plankton amount predicting unit, for the M group marine environment factor predicted value to be input to BP nerve net simultaneously
Network prediction model is predicted, to predict the red tide plankton amount predicted value in preset period.
Preferably, the obtaining unit includes:
Database subelement, for establishing the relational database of primitive ocean environmental factor, wherein in the relationship
The primitive ocean environmental factor of multiple Historical Monitorings is preserved in database, and preserve each primitive ocean environment because
The monitoring location information and monitoring time information of son；
Subelement is extracted, for the monitoring time information and monitoring position letter according to each primitive ocean environmental factor
Breath extracts M group primitive ocean environmental factor, wherein in the M group primitive ocean environmental factor from the relational database
Same group of primitive ocean environmental factor monitors in same sub- sea area, and apart from current time less than prefixed time interval；
Subelement is screened, for sieving from every group of primitive ocean environmental factor in the M group primitive ocean environmental factor
The factor set for meeting preset condition is selected as the effective marine environment factor of the M group.
Preferably, the red tide plankton amount prediction meanss further include: the first modeling unit, for establishing as follows
The ARIMA model:
When for according to the monitoring location information of the primitive ocean environmental factor each in the relational database, monitoring
Between information determine the ARIMA model for every group of primitive ocean environmental factor in the M group primitive ocean environmental factor；
The ARIMA model for being directed to every group of primitive ocean environmental factor respectively based on AIC criterion carries out parameter Estimation；
Parameter combination when obtaining minimum AIC value based on AIC criterion progress parameter Estimation is chosen, as described
The model parameter of the ARIMA model of every group of primitive ocean environmental factor.
Preferably, the red tide plankton amount prediction meanss further include: the second modeling unit, for establishing as follows
The BP neural network prediction model:
Obtain red tide plankton measurement sample；
The red tide plankton measurement sample is normalized to obtain corresponding return based on following logarithmic formula
Red tide plankton amount sample after one change:
Wherein, y ' is the value of red tide plankton amount after normalization, and y is red tide plankton measurement, ymaxFor the red tide plankton
The maximum value of measurement sample, yminFor the minimum value of the red tide plankton measurement sample；
The implicit number of plies of BP neural network is determined based on red tide plankton amount sample after the normalization and each is implicit
The number of nodes of layer；
The number of nodes of the implicit number of plies and each hidden layer based on determining BP neural network is established for the M group sea
The optimum network structure of the BP neural network of foreign environmental factor predicted value；
The optimal value that the optimum network structure is obtained by learning process is as the BP neural network prediction model.
Preferably, the red tide plankton amount prediction meanss further include: comparison unit, for the red tide plankton amount to be predicted
Value is compared with the red tide plankton measurement that the corresponding time measures；As a result output unit, for exporting the red tide plankton
It measures the root-mean-square error between predicted value and the red tide plankton measurement of the corresponding time measurement, mean absolute error, put down
At least one of absolute percent error, degree of fitting comparing result.
Preferably, the type of effective marine environment factor includes: pH value, dissolved oxygen, water temperature, dissolution oxygen saturation
Degree, chlorophyll-a, phosphate, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, salinity, chemical oxygen consumption (COC) and silicate.
By the embodiments of the present invention provide one or more technical solutions, at least realize following technical effect or
Advantage:
By obtaining the effective marine environment factor of M group；Pretreatment is normalized to the effective marine environment factor of M group, with
Obtain the marine environment factor after corresponding M group normalizes；Respectively by the m group normalizing in the marine environment factor after the normalization of M group
The marine environment factor is input in ARIMA model and is predicted after change, to predict M group marine environment factor predicted value；By M group
Marine environment factor predicted value is input to BP neural network prediction model simultaneously and is predicted, to predict in preset period
Red tide plankton amount predicted value.Wherein, M is the species number of effective marine environment factor multiplied by the number in sub- sea area, and one group has
The effect marine environment factor includes same kind of effective marine environment factor that multiframe acquires in different time points；To sufficiently examine
Lateral timing dependence and the longitudinal space for having considered the marine environment factor for influencing red tide are heterogeneous, can be directed to different sub- sea areas
The different factors establish different ARIMA models to describe its temporal continuity and special heterogeneity, to predict not
Each marine environment factor come, then answering between each marine environment factor and red tide is expressed on the basis of this using BP neural network
Miscellaneous relationship, and then realize the short-term Accurate Prediction of red tide plankton amount, whether red tide will be occurred in several days with forecast future, to be prevented
It is standby.
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
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is the flow diagram of red tide plankton amount prediction technique in the embodiment of the present invention；
Fig. 2 is the function unit figure of red tide plankton amount prediction technique in the embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Refering to what is shown in Fig. 1, including the following steps: the embodiment of the invention provides a kind of red tide plankton amount prediction technique
S101, the effective marine environment factor of M group is obtained, wherein M is the species number of effective marine environment factor multiplied by sub- sea
The number in domain, same group of effective marine environment factor include same kind of effective ocean ring that multiframe acquires in different time points
The border factor.
In one embodiment, the type of effective marine environment factor includes: pH value, dissolved oxygen, water temperature, dissolved oxygen
Saturation degree, chlorophyll-a, phosphate, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, salinity, chemical oxygen consumption (COC) and silicate this 12
Class.The sea area for needing to carry out red tide plankton amount prediction is divided into multiple sub- sea areas, and every sub- sea area corresponds to a longitude and latitude range,
For example, being divided into 6 sub- sea areas, then the effective marine environment factor of 12*6 group is obtained.Every sub- sea area be corresponding with 12 groups it is effectively extra large
Foreign environmental factor: one group of pH value factor, one group of dissolved oxygen factor, one group of dissolved oxygen factor, one group of water temperature factor, one group of dissolution
The oxygen saturation factor, one group of chlorophyll-a factor, one group of phosphate factor, one group of ammonia nitrogen factor, one group of nitrite nitrogen factor,
One group of salinity factor, one group of chemical oxygen consumption (COC) factor, one group of silicate factor.
Further include following steps then before executing S101 to obtain the effective marine environment factor of M group:
Firstly, establishing the relational database of primitive ocean environmental factor, wherein preserve multiple go through in relational database
History monitoring primitive ocean environmental factor, and preserve each primitive ocean environmental factor monitoring location information and monitoring when
Between information.
Specifically, the primitive ocean environmental factor saved in relational database is that the monitoring in different sub- sea areas is arranged in set
It is standby to be collected in different time.So as to form the big data of primitive ocean environmental factor.Specifically, monitoring location information is
Position where the monitoring device indicated with latitude and longitude coordinates, monitoring time information are to acquire the time of primitive ocean environmental factor
Point.
Then, according to the monitoring time information of each primitive ocean environmental factor and monitoring location information, from relation data
M group primitive ocean environmental factor is extracted in library, wherein same group of primitive ocean environment in M group primitive ocean environmental factor because
Son detects in same sub- sea area, and apart from current time less than prefixed time interval.For example, effective for needing to obtain 36 groups
The marine environment factor, every group of effective marine environment factor need 20 frames, monitor a frame daily with various primitive ocean environmental factors
For, from the various marine environment factors in relational database in each sub- sea area all extract the same day~retroversion 19 days this 20
Frame.
Followed by it is default to filter out satisfaction from every group of primitive ocean environmental factor in M group primitive ocean environmental factor
The factor set of condition becomes the effective marine environment factor of M group.
Specifically, being sieved by carrying out digital coding to the factor for meeting preset condition in M group primitive ocean environmental factor
Except monitoring mistake or the primitive ocean environmental factor of monitoring blank, the effective marine environment factor of M group is obtained to retain.
For example, due to seawater be it is alkaline, for pH value preset condition be pH value be greater than or equal to 7, to pH value
Carry out digital coding greater than 7, to pH value less than 7 without digital coding, thus screen out mistake pH value.Similarly
, the preset condition of other class primitive ocean environmental factors can be set in those skilled in the art, in order to illustrate the succinct of book, originally
Text repeats no more.
S102, pretreatment is normalized to the effective marine environment factor of M group, to obtain sea after corresponding M group normalization
Foreign environmental factor.
Specifically, by linear method to the every of every group of effective marine environment factor in the effective marine environment factor of M group
Pretreatment is all normalized in a effective marine environment factor, specifically, each effective marine environment factor is normalized
Pretreated formula is as follows:
Wherein, x ' is the marine environment factor after normalization, and x is effective marine environment factor, xminFor corresponding a kind of ocean
The minimum value of environmental factor, xmaxFor the maximum value of corresponding a kind of marine environment factor.For example, x is effective biochemical oxygen demand
Amount, then at this point, xminFor the minimum value for monitoring chemical oxygen consumption (COC), xmaxFor the maximum value of the chemical oxygen consumption (COC) of monitoring.
S103, the m group after the normalization of M group in the marine environment factor is normalized respectively after the marine environment factor be input to
(full name is difference ARMA model (Autoregressive Integrated Moving to corresponding ARIMA model
Average Model, is abbreviated as ARIMA) in predicted, to predict M group marine environment factor predicted value, m be followed successively by 1 to
M, wherein the marine environment factor after the corresponding one group of normalization of an ARIMA model.
In the specific implementation process, it needs before executing S103, establishes normalized respectively for M group as follows
The ARIMA model of the marine environment factor afterwards, to establish M ARIMA model:
Firstly, according to the monitoring location information of primitive ocean environmental factor each in relational database, monitoring time information
Determine the ARIMA model for every group of primitive ocean environmental factor in M group primitive ocean environmental factor.So that it is determined that M ARIMA
Model.
Specifically, determine ARIMA model be ARMA (ARMA model) model, AR (autoregression) model,
One of MA (rolling average) model.It specifically, will be each according to the monitoring location information of each primitive ocean environmental factor
The corresponding sub- sea area to place of a primitive ocean environmental factor, believes further according to the monitoring time of each primitive ocean environmental factor
The monitoring of the corresponding 12 class primitive ocean environmental factors to the sub- sea area in place is surveyed temporal information and forms 12 time serieses by breath.
To form " sub- sea area number * 12 " a time series.Unit root test is carried out to each time series of formation, it is each to judge
Whether time series is stationary sequence.Stationary sequence are as follows: mean value and variance are constant in time course, and at any two
The covariance value of phase rely only on this two when during distance or lag, be otherwise non-stationary series.If carry out unit root test
There are unit roots, then the time series is non-stationary series, and non-stationary series need to carry out difference and form stationary sequence, difference time
Number scale is d.
Specifically, every class primitive ocean ring is determined by correlation analysis to stationary sequence or differentiated stationary sequence
The applicable ARIMA model of the border factor: being to carry out correlation analysis to stationary sequence or differentiated stationary sequence specifically,
Autocorrelogram and partial correlation figure are drawn, deviation―related function is determined based on autocorrelogram and partial correlation figure and auto-correlation function is truncation
Or hangover, if it is hangover that the deviation―related function of stationary sequence, which is truncation, autocorrelation, which is applicable in
AR model；If the deviation―related function of stationary sequence trails, auto-correlation function truncation, it is applicable in MR model；If stationary sequence is inclined
Correlation function, auto-correlation function are hangovers, then are applicable in arma modeling.Wherein, truncation refers to the auto-correlation letter of time series
Number (ACF) or partial autocorrelation function (PACF) are 0 property after certain rank；Hangover is ACF or PACF equal not after certain rank
For 0 property.
Then, the ARIMA model for being directed to every group of primitive ocean environmental factor respectively based on AIC criterion carries out parameter Estimation,
Parameter combination when obtaining minimum AIC value based on AIC criterion progress parameter Estimation is chosen, as every group of primitive ocean environment
The model parameter ARIMA (p, d, q) of the ARIMA model of the factor.
Specifically, the autoregression item number p and rolling average item number q of arma modeling are determined according to AIC criterion.According to AIC
Criterion determines the autoregression item p of AR model；The average mobile item number q of MA model is determined according to AIC criterion.The formula of AIC criterion
For AIC=2K-2ln (F), wherein K is item number, and F is likelihood function.
Finally, setting t, the time sequential value at t-1 ... t-p moment is respectively yt,yt-1,…,yt-p, in t, t-1 ... t-q
The error at quarter is denoted as nt,nt-1,…nt-q, nt,nt-1,…nt-qIt is white noise sequence that is mutually indepedent and meeting Gaussian Profile.a1,
a2,…,apIt is autoregression (AR) coefficient, b1,b2,…,bqIt is sliding average (MA) coefficient, then the following institute of the expression formula of arma modeling
Show:
yt=a1yt-1+a2yt-2+…+apyt-p+nt+b1nt-1+b2nt-2+bqnt-q。
S104, M group marine environment factor predicted value is input to BP (Back Propagation backpropagation) mind simultaneously
It is predicted through Network Prediction Model, to predict the red tide plankton amount predicted value in preset period.
In the specific implementation process, preset period is set smaller than 7 days, then predicts the red tide plankton amount prediction in 7 days
Value, to avoid the value inaccuracy of long-term forecast.
Further, it before executing S104, establishes carried out in advance for marine environment factor predicted value as follows
The BP neural network prediction model of survey:
Step 1: obtaining red tide plankton measurement sample；
Specifically, red tide plankton measurement sample can be that the monitoring device being arranged in each sub- sea area monitors to obtain
's.
Step 2: being normalized to obtain based on red tide plankton measurement sample of the following logarithmic formula to acquisition
Red tide plankton amount sample after corresponding normalization:Wherein, y ' is red tide after normalization
The value of biomass, y are red tide plankton measurement, ymaxFor the maximum value of red tide plankton measurement sample, yminIt is raw for red tide
The minimum value of object amount measurement samples.
Specifically, each red tide plankton measurement of red tide plankton measurement sample is obtained by logarithmic formula respectively
The value of red tide plankton amount after to each normalization, all red tide plankton amount is constituted to be formed for training BP neural after normalization
Red tide plankton amount sample after the normalization of network structure.
Step 3: based on after normalization red tide plankton amount sample determine need BP neural network structure the implicit number of plies with
And the number of nodes of each hidden layer.
Specifically, by each normalize after the BP neural network structure brought into respectively of red tide plankton amount test, base
The number of nodes of the implicit number of plies required for experimental result quality determines and each hidden layer.
Step 4: the number of nodes of the implicit number of plies and each hidden layer based on determining BP neural network establishes BP nerve
The optimum network structure of network；
Step 5: optimum network structure is obtained optimal value as BP neural network prediction model by learning process.
Specifically, learning process is as follows:
The first step, random initializtion weight θ nonzero value, and calculated loss function J (θ) by propagated forward algorithm, and
The local derviation D of initialization weight θ is calculated, calculation formula is as follows:
Second step, the maximum study number of given learning process, setting error function, default accuracy value, give input sea
Foreign environmental factor sample set: given red tide plankton amount exports sample set:
Third step, at random from input marine environment factor sample set choose an input sample, it is corresponding, from red tide plankton
An output sample is chosen in amount output sample set, carries out that global miss is calculated using the gradient descent algorithm of backpropagation
Difference, judges whether global error reaches default accuracy value, or whether study number is greater than maximum study number, when global error reaches
To default accuracy value or it is greater than maximum study number, then terminates gradient descent algorithm, to obtain making loss function J (θ) the smallest
Weighted value θ is optimal value as BP neural network prediction model, otherwise chooses next input sample and corresponding output sample
It comes back for learning next time.What is obtained makes the smallest weighted value θ optimal value of loss function J (θ).
After predicting the red tide plankton amount predicted value in preset period, by red tide plankton amount predicted value with to it is corresponding when
Between the red tide plankton measurement that measures compare；The red tide plankton that output red tide plankton amount predicted value is measured with the corresponding time
Root-mean-square error between measurement, the comparison of at least one of mean absolute error, average absolute percentage error, degree of fitting
As a result.
Specifically, the calculation formula of root-mean-square error is as follows:Wherein, RMSE is yi
For red tide plankton measurement, y "iFor red tide plankton amount predicted value, N is the number of red tide plankton measurement, according to root mean square
Error RMSE can be well reflected out the precision of red tide plankton amount predicted value.
Specifically, the calculation formula of mean absolute error are as follows:Wherein, MAE is average exhausted
To error, yiFor red tide plankton measurement, y "iFor red tide plankton amount predicted value, N is the number of red tide plankton measurement, root
The actual error situation of red tide plankton amount predicted value can be well reflected out according to mean absolute error MAE.
Specifically, the calculation formula of average absolute percentage error are as follows:
Wherein, MAPE is average absolute percentage error, yiFor red tide plankton measurement, y "iFor red tide plankton amount predicted value, N is red tide
The number of biometric measurement magnitude can be well reflected out red tide plankton amount predicted value according to average absolute percentage error MAPE
Accuracy.
Specifically, the calculation formula of degree of fitting are as follows:Wherein, R is average absolute percentage
Error, yiFor red tide plankton measurement, y "iFor red tide plankton amount predicted value, N is the number of red tide plankton measurement.When quasi-
Right numerical value then shows that the goodness of fit between red tide plankton measurement and red tide plankton amount predicted value is stronger closer to 1,
Prediction effect is better；Conversely, then illustrating that the fitting effect between red tide plankton measurement and red tide plankton amount predicted value is poor, in advance
It is poor to survey accuracy.
Based on the same inventive concept, the embodiment of the invention provides a kind of red tide plankton amount prediction meanss, with reference to Fig. 2 institute
Show, comprising: obtaining unit 201, for obtaining the effective marine environment factor of M group, wherein M is effective marine environment factor
Species number multiplied by the number in sub- sea area, effective marine environment factor includes that multiframe acquires in different time points described in same group
Same kind of effective marine environment factor；Pretreatment unit 202, for returning to the effective marine environment factor of the M group
One changes pretreatment, to obtain the marine environment factor after corresponding M group normalization；Environmental factor predicting unit 203 respectively will be used for
The marine environment factor is input to corresponding ARIMA model after m group normalization after the M group normalization in the marine environment factor
In predicted, to predict M group marine environment factor predicted value, m is followed successively by 1 to M, wherein ARIMA model corresponding one
The marine environment factor after group normalization；Red tide plankton amount predicting unit 204 is used for the M group marine environment factor predicted value
It is input to BP neural network prediction model simultaneously to be predicted, to predict the red tide plankton amount predicted value in preset period.
Preferably, the obtaining unit 201 includes:
Database subelement, for establishing the relational database of primitive ocean environmental factor, wherein in the relationship
The primitive ocean environmental factor of multiple Historical Monitorings is preserved in database, and preserve each primitive ocean environment because
The monitoring location information and monitoring time information of son；
Subelement is extracted, for the monitoring time information and monitoring position letter according to each primitive ocean environmental factor
Breath extracts M group primitive ocean environmental factor, wherein in the M group primitive ocean environmental factor from the relational database
Same group of primitive ocean environmental factor monitors in same sub- sea area, and apart from current time less than prefixed time interval；
Subelement is screened, for sieving from every group of primitive ocean environmental factor in the M group primitive ocean environmental factor
The factor set for meeting preset condition is selected as the effective marine environment factor of the M group.
Preferably, the red tide plankton amount prediction meanss further include: the first modeling unit, for establishing as follows
The ARIMA model:
When for according to the monitoring location information of the primitive ocean environmental factor each in the relational database, monitoring
Between information determine the ARIMA model for every group of primitive ocean environmental factor in the M group primitive ocean environmental factor；
The ARIMA model for being directed to every group of primitive ocean environmental factor respectively based on AIC criterion carries out parameter Estimation；
Parameter combination when obtaining minimum AIC value based on AIC criterion progress parameter Estimation is chosen, as described
The model parameter of the ARIMA model of every group of primitive ocean environmental factor.
Preferably, the red tide plankton amount prediction meanss further include: the second modeling unit, for establishing as follows
The BP neural network prediction model:
Obtain red tide plankton measurement sample；
The red tide plankton measurement sample is normalized to obtain corresponding return based on following logarithmic formula
Red tide plankton amount sample after one change:
Wherein, y ' is the value of red tide plankton amount after normalization, and y is red tide plankton measurement, ymaxFor the red tide plankton
The maximum value of measurement sample, yminFor the minimum value of the red tide plankton measurement sample；
The implicit number of plies of BP neural network is determined based on red tide plankton amount sample after the normalization and each is implicit
The number of nodes of layer；
The number of nodes of the implicit number of plies and each hidden layer based on determining BP neural network is established for the M group sea
The optimum network structure of the BP neural network of foreign environmental factor predicted value；
The optimal value that the optimum network structure is obtained by learning process is as the BP neural network prediction model.
Preferably, the red tide plankton amount prediction meanss further include: comparison unit, for the red tide plankton amount to be predicted
Value is compared with the red tide plankton measurement that the corresponding time measures；As a result output unit, for exporting the red tide plankton
It measures the root-mean-square error between predicted value and the red tide plankton measurement of the corresponding time measurement, mean absolute error, put down
At least one of absolute percent error, degree of fitting comparing result.
Preferably, the type of effective marine environment factor includes: pH value, dissolved oxygen, water temperature, dissolution oxygen saturation
Degree, chlorophyll-a, phosphate, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, salinity, chemical oxygen consumption (COC) and silicate.
The one or more embodiments provided by aforementioned present invention, at least realize following technical effect or advantage:
By obtaining the effective marine environment factor of M group；Pretreatment is normalized to the effective marine environment factor of M group, with
Obtain the marine environment factor after corresponding M group normalizes；Respectively by the m group normalizing in the marine environment factor after the normalization of M group
The marine environment factor is input in ARIMA model and is predicted after change, to predict M group marine environment factor predicted value；By M group
Marine environment factor predicted value is input to BP neural network prediction model simultaneously and is predicted, to predict in preset period
Red tide plankton amount predicted value.Wherein, M is the species number of effective marine environment factor multiplied by the number in sub- sea area, and one group effectively extra large
Foreign environmental factor includes same kind of effective marine environment factor that multiframe acquires in different time points；To fully consider
Lateral timing dependence and the longitudinal space for influencing the marine environment factor of red tide are heterogeneous, can be for different sub- sea areas not
Different ARIMA models is established to describe its temporal continuity and special heterogeneity with the factor, to predict future
Each marine environment factor, then the complicated pass between each marine environment factor and red tide is expressed on the basis of this using BP neural network
System, and then realize the short-term Accurate Prediction of red tide plankton amount, whether red tide will be occurred in several days with forecast future, to be guarded against.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (7)
1. a kind of red tide plankton amount prediction technique based on the marine environment factor characterized by comprising
Obtain the effective marine environment factor of M group, wherein M is the species number of effective marine environment factor multiplied by sub- sea area
Number, effective marine environment factor described in same group include same kind of effective ocean ring that multiframe acquires in different time points
The type of the border factor, effective marine environment factor includes: pH value, dissolved oxygen, water temperature, saturation dissolved oxygen, chlorophyll-
A, phosphate, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, salinity, chemical oxygen consumption (COC) and silicate；
Pretreatment is normalized to the effective marine environment factor of the M group, to obtain marine environment after corresponding M group normalization
The factor；
The marine environment factor after the m group normalization in the marine environment factor after M group normalization is input to correspondence respectively
ARIMA model in predicted, to predict M group marine environment factor predicted value, m is followed successively by 1 to M, wherein one
The marine environment factor after the corresponding one group of normalization of ARIMA model；
The M group marine environment factor predicted value is input to BP neural network prediction model simultaneously to predict, to predict
Red tide plankton amount predicted value in preset period, wherein the preset period was less than 7 days；
Wherein, the M group marine environment factor predicted value BP neural network prediction model is input to simultaneously described to carry out in advance
Before survey, the method also includes: the BP neural network prediction model is established as follows:
Obtain red tide plankton measurement sample；
The red tide plankton measurement sample is normalized to obtain corresponding normalization based on following logarithmic formula
Red tide plankton amount sample afterwards:
Wherein, y ' is the value of red tide plankton amount after normalization, and y is red tide plankton measurement, ymaxFor red tide plankton measurement
The maximum value of magnitude sample, yminFor the minimum value of the red tide plankton measurement sample；
The implicit number of plies and each hidden layer of BP neural network are determined based on red tide plankton amount sample after the normalization
Number of nodes；
The number of nodes of the implicit number of plies and each hidden layer based on determining BP neural network, which is established, is directed to M group ocean ring
The optimum network structure of the BP neural network of border factor predicted value；
The optimal value that the optimum network structure is obtained by learning process is as the BP neural network prediction model.
2. red tide plankton amount prediction technique as described in claim 1, which is characterized in that the effective marine environment of the acquisition M group
The factor, comprising:
Establish the relational database of primitive ocean environmental factor, wherein multiple history prisons are preserved in the relational database
The primitive ocean environmental factor of survey, and preserve each primitive ocean environmental factor monitoring location information and monitoring when
Between information；
According to the monitoring time information of each primitive ocean environmental factor and monitoring location information, from the relational database
Middle extraction M group primitive ocean environmental factor, wherein same group of primitive ocean environment in the M group primitive ocean environmental factor
The factor monitors in same sub- sea area, and apart from current time less than prefixed time interval；
Filtered out from every group of primitive ocean environmental factor in the M group primitive ocean environmental factor meet preset condition because
Subgroup becomes the effective marine environment factor of the M group.
3. red tide plankton amount prediction technique as claimed in claim 2, which is characterized in that described respectively by the M group normalizing
The marine environment factor, which is input in corresponding ARIMA model, after m group normalization after change in the marine environment factor is predicted
Before, the method also includes: establish the ARIMA model as follows:
It is true according to the monitoring location information of the primitive ocean environmental factor each in the relational database, monitoring time information
Surely for the ARIMA model of every group of primitive ocean environmental factor in the M group primitive ocean environmental factor；
The ARIMA model for being directed to every group of primitive ocean environmental factor respectively based on AIC criterion carries out parameter Estimation；
Parameter combination when obtaining minimum AIC value based on AIC criterion progress parameter Estimation is chosen, as described every group
The model parameter of the ARIMA model of primitive ocean environmental factor.
4. red tide plankton amount prediction technique as described in claim 1, which is characterized in that predicted in preset period described
Red tide plankton amount predicted value after, the method also includes:
The red tide plankton amount predicted value is compared with the red tide plankton measurement that the corresponding time measures；
Export the root mean square between the red tide plankton amount predicted value and the red tide plankton measurement of the corresponding time measurement
At least one of error, mean absolute error, average absolute percentage error, degree of fitting comparing result.
5. a kind of red tide plankton amount prediction meanss based on the marine environment factor characterized by comprising
Obtaining unit, for obtaining the effective marine environment factor of M group, wherein M is the species number of effective marine environment factor
Multiplied by the number in sub- sea area, effective marine environment factor described in same group includes the one species that multiframe acquires in different time points
Effective marine environment factor, the type of effective marine environment factor includes: that pH value, dissolved oxygen, water temperature, dissolved oxygen are full
With degree, chlorophyll-a, phosphate, ammonia nitrogen, nitrite nitrogen, nitrate nitrogen, salinity, chemical oxygen consumption (COC) and silicate；
Pretreatment unit, for pretreatment to be normalized to the effective marine environment factor of the M group, to obtain corresponding M group
The marine environment factor after normalization；
Environmental factor predicting unit, for respectively by sea after the m group normalization in the marine environment factor after M group normalization
Foreign environmental factor is input in corresponding ARIMA model and is predicted, to predict M group marine environment factor predicted value, m is successively
It is 1 to M, wherein the marine environment factor after the corresponding one group of normalization of an ARIMA model；
Red tide plankton amount predicting unit, it is pre- for the M group marine environment factor predicted value to be input to BP neural network simultaneously
It surveys model to be predicted, to predict the red tide plankton amount predicted value in preset period, wherein the preset period is less than 7
It；
Wherein, the red tide plankton amount prediction meanss further include: the second modeling unit, for establishing the BP as follows
Neural network prediction model:
Obtain red tide plankton measurement sample；
The red tide plankton measurement sample is normalized to obtain corresponding normalization based on following logarithmic formula
Red tide plankton amount sample afterwards:
Wherein, y ' is the value of red tide plankton amount after normalization, and y is red tide plankton measurement, ymaxFor red tide plankton measurement
The maximum value of magnitude sample, yminFor the minimum value of the red tide plankton measurement sample；
The implicit number of plies and each hidden layer of BP neural network are determined based on red tide plankton amount sample after the normalization
Number of nodes；
The number of nodes of the implicit number of plies and each hidden layer based on determining BP neural network, which is established, is directed to M group ocean ring
The optimum network structure of the BP neural network of border factor predicted value；
The optimal value that the optimum network structure is obtained by learning process is as the BP neural network prediction model.
6. red tide plankton amount prediction meanss as claimed in claim 5, which is characterized in that the obtaining unit includes:
Database subelement, for establishing the relational database of primitive ocean environmental factor, wherein in the relation data
The primitive ocean environmental factor of multiple Historical Monitorings is preserved in library, and preserves each primitive ocean environmental factor
Monitor location information and monitoring time information；
Subelement is extracted, for according to the monitoring time information of each primitive ocean environmental factor and monitoring location information,
M group primitive ocean environmental factor is extracted from the relational database, wherein same in the M group primitive ocean environmental factor
One group of primitive ocean environmental factor monitors in same sub- sea area, and apart from current time less than prefixed time interval；
Subelement is screened, for filtering out from every group of primitive ocean environmental factor in the M group primitive ocean environmental factor
The factor set for meeting preset condition becomes the effective marine environment factor of the M group.
7. red tide plankton amount prediction meanss as claimed in claim 6, which is characterized in that the red tide plankton amount prediction meanss are also
It include: the first modeling unit, for establishing the ARIMA model as follows:
For being believed according to the monitoring location information of the primitive ocean environmental factor each in the relational database, monitoring time
Breath determines the ARIMA model for every group of primitive ocean environmental factor in the M group primitive ocean environmental factor；
The ARIMA model for being directed to every group of primitive ocean environmental factor respectively based on AIC criterion carries out parameter Estimation；
Parameter combination when obtaining minimum AIC value based on AIC criterion progress parameter Estimation is chosen, as described every group
The model parameter of the ARIMA model of primitive ocean environmental factor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610859375.2A CN106407691B (en) | 2016-09-28 | 2016-09-28 | A kind of red tide plankton amount prediction technique and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610859375.2A CN106407691B (en) | 2016-09-28 | 2016-09-28 | A kind of red tide plankton amount prediction technique and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106407691A CN106407691A (en) | 2017-02-15 |
| CN106407691B true CN106407691B (en) | 2019-11-22 |
Family
ID=58015733
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610859375.2A Active CN106407691B (en) | 2016-09-28 | 2016-09-28 | A kind of red tide plankton amount prediction technique and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106407691B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109002888B (en) * | 2018-06-27 | 2022-01-11 | 厦门市海洋与渔业研究所 | Red tide early warning method |
| CN109934419A (en) * | 2019-04-11 | 2019-06-25 | 苏州热工研究院有限公司 | A kind of nuclear power plant's intake marine organisms amount variation prediction technique |
| CN110926433A (en) * | 2019-11-28 | 2020-03-27 | 中国船舶重工集团公司第七一九研究所 | Marine disaster early warning system for coastal nuclear power station |
| CN111860328B (en) * | 2020-07-21 | 2021-04-06 | 杭州时光坐标影视传媒股份有限公司 | Biomass estimation method based on bidirectional reflection function and forest scene illumination effect modeling |
| CN114971005A (en) * | 2022-05-20 | 2022-08-30 | 厦门大学 | Bay water temperature combination prediction method based on LSTM and differential regression model dynamic weighting |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102495937A (en) * | 2011-10-18 | 2012-06-13 | 南京信息工程大学 | Prediction method based on time sequence |
| CN102737288A (en) * | 2012-06-20 | 2012-10-17 | 浙江大学 | Radial basis function (RBF) neural network parameter self-optimizing-based multi-step prediction method for water quality |
| CN103049669A (en) * | 2013-01-07 | 2013-04-17 | 北京工商大学 | Method for forecasting water bloom and analyzing factors on basis of multivariate cyclostationary time sequence analysis and grey theory |
| CN104899653A (en) * | 2015-06-02 | 2015-09-09 | 北京工商大学 | Lake and reservoir cyanobacterial bloom prediction method based on expert system and cyanobacterial growth mechanism timing model |
-
2016
- 2016-09-28 CN CN201610859375.2A patent/CN106407691B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102495937A (en) * | 2011-10-18 | 2012-06-13 | 南京信息工程大学 | Prediction method based on time sequence |
| CN102737288A (en) * | 2012-06-20 | 2012-10-17 | 浙江大学 | Radial basis function (RBF) neural network parameter self-optimizing-based multi-step prediction method for water quality |
| CN103049669A (en) * | 2013-01-07 | 2013-04-17 | 北京工商大学 | Method for forecasting water bloom and analyzing factors on basis of multivariate cyclostationary time sequence analysis and grey theory |
| CN104899653A (en) * | 2015-06-02 | 2015-09-09 | 北京工商大学 | Lake and reservoir cyanobacterial bloom prediction method based on expert system and cyanobacterial growth mechanism timing model |
Non-Patent Citations (5)
| Title |
|---|
| CBR based system for forecasting red tides;Florentino Fdez-Riverola et al;《Knowledge-Based Systems》;20030731(第5-6期);全文 * |
| FSfRT: Forecasting System for Red Tides;Florentino Fdez-Riverola et al;《Applied Intelligence》;20041130;第21卷(第3期);全文 * |
| WATER QUALITY PREDICTION BASED ONA NOVEL HYBRID MODELOFARIMAAND RBF NEURAL NETWORK;Weihui Deng et al;《2014 IEEE 3rd International Conference on Cloud Computing and Intelligence Systems》;20141129;正文第2-3节 * |
| 基于 ARIMA-ANN 的时间序列组合预测模型;张吉刚等;《三峡大学学报（自然科学版）》;20080831;第30卷(第4期);全文 * |
| 基于ARIMA-BP模型的包头市环境可持续发展预测;宋宇辰等;《资源开发与市场》;20160228;第32卷(第2期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106407691A (en) | 2017-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106407691B (en) | A kind of red tide plankton amount prediction technique and device | |
| Harris et al. | Ecoacoustic indices as proxies for biodiversity on temperate reefs | |
| Altunkaynak | Forecasting surface water level fluctuations of Lake Van by artificial neural networks | |
| Mazzocchi et al. | Stability and resilience in coastal copepod assemblages: The case of the Mediterranean long-term ecological research at Station MC (LTER-MC) | |
| CN110070144A (en) | A kind of lake water quality prediction technique and system | |
| Rabi et al. | Modelling river temperature from air temperature: case of the River Drava (Croatia) | |
| Moreno-Gómez et al. | Evaluating acoustic indices in the Valdivian rainforest, a biodiversity hotspot in South America | |
| Deshmukh et al. | Neural-network-based data assimilation to improve numerical ocean wave forecast | |
| Pool et al. | Streamflow characteristics from modeled runoff time series–importance of calibration criteria selection | |
| CN106371155B (en) | Method of meteorological forecast and system based on big data and analysis field | |
| O’Boyle et al. | Observations on the implementation of ecosystem-based management: experiences on Canada's east and west coasts | |
| Tobin et al. | Environmental drivers of paralytic shellfish toxin producing Alexandrium catenella blooms in a fjord system of northern Southeast Alaska | |
| Barange et al. | Predicting the Impacts and Socio‐Economic Consequences of Climate Change on Global Marine Ecosystems and Fisheries: The QUEST_Fish Framework | |
| Winiarski et al. | Spatially explicit model of wintering common loons: conservation implications | |
| Chaalali et al. | From species distributions to ecosystem structure and function: a methodological perspective | |
| Lee et al. | Neural network for the prediction and supplement of tidal record in Taichung Harbor, Taiwan | |
| Welch et al. | Implications of climate change on fisheries resources of northern Australia Part 1: Vulnerability assessment and adaptation options | |
| Rounds | Development of a neural network model for dissolved oxygen in the Tualatin River, Oregon | |
| Jirakittayakorn et al. | Temporal kNN for short-term ocean current prediction based on HF radar observations | |
| Dauji et al. | Prediction of ocean currents with artificial neural networks | |
| Curnock et al. | Increased use levels, effort, and spatial distribution of tourists swimming with dwarf minke whales at the Great Barrier Reef | |
| Herrera et al. | Variability of the western Galician upwelling system (NW Spain) during an intensively sampled annual cycle. An EOF analysis approach | |
| Chasco et al. | Evaluating the accuracy and precision of multiple abundance estimators using state-space models: a case study for a threatened population of chinook salmon in Johnson Creek, Idaho | |
| Campagna et al. | A species approach to marine ecosystem conservation | |
| Papagera et al. | Simulation and prediction of water allocation using artificial neural networks and a spatially distributed hydrological model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |