CN101825622A - Water quality prediction method and device - Google Patents

Abstract

The invention discloses a water quality prediction method and a device. The method includes the steps that: raw water data is acquired, and according to the raw water data, a prediction model is established; the relative error of the prediction model is calculated, the prediction model is an effective model when the relative error of the prediction model is smaller than or equal to a predetermined relative error, and water quality prediction is carried out according to the prediction model; and the establishment of the prediction model includes the steps that: an original GM(1,1) model is established according to the raw water data, and an original GM(1,1) model value is calculated; according to the original GM(1,1) model value, primary residual error data is determined, and is utilized to establish a primary residual error GM(1,1) model; according to the primary residual error GM(1,1) model, the original GM(1,1) model is corrected, a corrected GM(1,1) model is established, and a corrected GM(1,1) model value is calculated; and according to the corrected GM(1,1) model value, secondary residual error data is determined, and is utilized to establish a BP neural network model.

Description

A kind of water quality prediction method and device

Technical field

The present invention relates to the water quality prediction field, relate in particular to a kind of water quality prediction method and device.

Background technology

Current, we can be divided into two classes by applied water quality prediction model substantially: the one, and by the various factors of analyzing influence water quality, the mode that biodynamic changes in the Simulated Water and the model set up; The 2nd, the model that utilizes the water quality historical data to set up.The oxygen balance model that the former was proposed by U.S. slip-stick artist Streeter and Phelps such as nineteen twenty-five, its basic foundation is that DO concentration depends on BOD reflection and reoxygenation process, the reflection that is participated in by anaerobe belongs to the viewpoint of first _ order kinetics equation, moreover as the QUAL-I water quality unified model that proposes of EPA in 1970 etc.The former complexity of model is higher, generally only is used for the simulation to water quality, and the latter mainly contains linear regression model (LRM) (statistics pattern type), gray theory model (non-statistical pattern type), neural network model and corresponding correction model.

The research focus of water quality prediction also is gray theory model and neural network model at present.About the gray theory application of model, comprise that the wide territory of use Grey Decision Method is set up the Water Quality Evaluation model, Gray Dynamic GM (1,1) model group sets up the water quality prediction of river model, and utilize gray relative analysis method to carry out water analysis or the like; And, then predict downstream water quality and the following water quality of current water quality prediction according to upstream water quality such as the prediction that utilizes the BP neural network to reservoir water quality, use neural network for neural network model, and the water quality prediction model of adaptive neural network or the like.Application to water quality prediction conforms to above-mentioned current gray theory with neural network, all is with form independently water quality to be carried out modeling and forecasting, and utilizes correlation theory that model is improved with expectation to obtain higher precision of prediction.

The problem that above-mentioned water quality prediction model mainly exists is: described various models only at water quality data grey characteristics and time growth property, or only at its undulatory property (comprising nonlinear characteristic), grey characteristics, time growth property and undulatory property are not combined consideration, thereby can not obtain very high precision of prediction.

Summary of the invention

In view of this, the invention provides a kind of water quality prediction method and device, utilize GM (1,1) and BP neural network to carry out the associating modeling, both considered grey characteristics, the time growth property of water quality data, considered its nonlinear characteristic again, simultaneously GM (1,1) is revised, and revising GM (1,1) utilizes the BP neural network to carry out the secondary correction on the model basis, thereby obtained very high water quality prediction precision.

Technical solution of the present invention is as follows:

A kind of water quality prediction method comprises:

Obtain former water number certificate, and according to described former water number according to setting up forecast model;

Calculate the relative error of described forecast model, when the relative error of described forecast model was less than or equal to predetermined relative error, described forecast model was a valid model, and carried out water quality prediction according to described forecast model;

The described forecast model of setting up comprises:

According to setting up original GM (1,1) model, and calculate original GM (1,1) model value according to former water number;

Determine residual error data one time by original GM (1,1) model value, and utilize a described residual error data to set up a Residual GM (1,1) model;

According to a Residual GM (1,1) model original GM (1,1) model is once revised, set up and revise GM (1,1) model, and calculate and revise GM (1,1) model value;

Determine the quadratic residue data by revising GM (1,1) model value, and utilize described quadratic residue data to set up the BP neural network model.

Preferably, a described residual error data obtains according to the difference with GM (1,1) model value by former water number.

Preferably, described quadratic residue data obtain according to the difference with correction GM (1,1) model value by former water number.

Preferably, described: as to calculate the relative error of described forecast model, specifically comprise:

Obtain the true predictive value of revising GM (1,1) model;

Obtain the true predictive value of BP neural network model;

Two predicted values that superpose obtain the true predictive value of forecast model;

Utilize the true predictive value of described forecast model and the relative error that described former water number it is calculated that described forecast model.

Preferably, the described true predictive value of revising GM (1,1) model of obtaining specifically comprises:

Extrapolate to revising GM (1,1) model, obtain generating the number predicted value;

Carry out once tired subtracting to generating the number predicted value, obtain to revise the true predictive value of GM (1,1) model.

Preferably, the true predictive value of the described BP of obtaining neural network model specifically comprises:

With the end data of quadratic residue data as first constantly the input data, in order to predict second constantly the predicted value;

With second constantly the predicted value as the input data, in order to predict the 3rd constantly predicted value;

By that analogy, finally obtain the true predictive value of described BP neural network model.

A kind of water quality prediction device comprises:

Acquiring unit is used to obtain former water number certificate;

Model unit is used for according to described former water number according to setting up forecast model;

Computing unit is used to calculate the relative error of described forecast model;

Test film unit, be used to verify when the relative error of described forecast model is less than or equal to predetermined relative error, determine that described forecast model is a valid model;

Predicting unit is used for carrying out water quality prediction according to described forecast model;

Described modeling unit specifically comprises:

The master pattern subelement is used for according to former water number according to setting up original GM (1,1) model, and calculates original GM (1,1) model value;

Residual error model subelement is used for determining residual error data one time by original GM (1,1) model value, and utilizes a described residual error data to set up a Residual GM (1,1) model;

The correction model subelement is used for according to a Residual GM (1,1) model original GM (1,1) model once being revised, and sets up and revises GM (1,1) model, and calculate and revise GM (1,1) model value;

The network model subelement is used for determining the quadratic residue data by revising GM (1,1) model value, and utilizes described quadratic residue data to set up the BP neural network model.

Preferably, described computing unit comprises:

First acquiring unit is used to obtain the true predictive value of revising GM (1,1) model;

Second acquisition unit is used to obtain the true predictive value of BP neural network model;

Superpositing unit, two predicted values that are used to superpose obtain the true predictive value of forecast model;

Computation subunit is used to utilize the true predictive value of described forecast model and the relative error that described former water number it is calculated that described forecast model.

Preferably, described first acquiring unit comprises:

The extrapolation unit is used for extrapolating to revising GM (1,1) model, obtains generating the number predicted value;

Tired subtract the unit, be used for carrying out once tired subtracting, obtain to revise the true predictive value of GM (1,1) model generating the number predicted value.

Preferably, the effect of described second acquisition unit is specially:

With the end data of quadratic residue data as first constantly the input data, in order to predict second constantly the predicted value;

With second constantly the predicted value as the input data, in order to predict the 3rd constantly predicted value;

By that analogy, finally obtain the true predictive value of described BP neural network model.

From above-mentioned technical scheme as can be seen, the embodiment of the invention utilizes GM (1,1) model and BP neural network model to realize combined type stack forecast model, and raw water quality is predicted.Adopt GM (1,1) model that the water quality time series is carried out modeling earlier, utilize a residual error data to set up Residual GM (1 again, 1) model is revised GM (1, the 1) model of raw data, and predict, can know the general trend of change of water quality from this correction model; Because water quality not only has grey characteristics, time growth property, but also have certain fluctuation, fluctuation itself has comprised nonlinear transformations (water quality data is subjected to the influence of various physical reactions and chemical reaction) again, in order effectively to handle nonlinear transformations, the present invention is revising GM (1,1) on the model based, adopt the BP neural network that the water quality nonlinear characteristic is compensated again, promptly use the BP neural network to revising GM (1,1) the quadratic residue data of model are trained, and acquisition can be handled the BP neural network model of nonlinear characteristic; Finally, the present invention is to revising GM (1,1) predicted value of model and BP neural network model superposes, obtain the predicted value of forecast model, by calculating the relative error of forecast model, and this error compared with predetermined relative error, when this error is less than or equal to predetermined relative error, determine that described forecast model is a valid model.By the present invention, make water quality prediction reach the better prediction effect, improved precision of prediction greatly.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art, to do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below, apparently, accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

Fig. 1 is the process flow diagram of the disclosed a kind of water quality prediction method of the embodiment of the invention;

Fig. 2 is the disclosed process flow diagram of setting up forecast model of the embodiment of the invention;

Fig. 3 is the structural drawing of the disclosed BP neural network of the embodiment of the invention;

Fig. 4 is the synoptic diagram of the disclosed BP train samples of the embodiment of the invention;

Fig. 5 is the synoptic diagram of the disclosed BP neural network of embodiment of the invention iteration one-step prediction method;

Fig. 6 is the structural representation of the disclosed a kind of water quality prediction device of the embodiment of the invention;

Fig. 7 is the structural representation of the disclosed computing unit of the embodiment of the invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the invention, the technical scheme in the embodiment of the invention is clearly and completely described, obviously, described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills belong to the scope of protection of the invention not making the every other embodiment that is obtained under the creative work prerequisite.

The embodiment of the invention discloses a kind of water quality prediction method, utilize GM (1,1) and the BP neural network carry out the associating modeling, both considered grey characteristics, the time growth property of water quality data, considered its nonlinear characteristic again, simultaneously to GM (1,1) revises, and on correction GM (1,1) model basis, utilize the BP neural network to carry out the secondary correction, thereby improved precision of prediction greatly.

Referring to Fig. 1, the process flow diagram for the disclosed a kind of water quality prediction method of the embodiment of the invention specifically comprises the steps:

S11, obtain former water number certificate, and according to described former water number according to setting up forecast model;

The described process of setting up forecast model can specifically comprise referring to schematic flow sheet shown in Figure 2:

S21, according to former water number according to setting up original GM (1,1) model, and calculate original GM (1,1) model value;

This step comprises following substep:

A1, raw data is carried out one-accumulate;

B1, the data after adding up are carried out modeling, that is, set up the albefaction form differential equation of GM (1,1) by gray theory;

C1, ask the parameter of the described differential equation;

D1, ask separating of GM (1, the 1) differential equation, and ask GM (1,1) model value.

Described modeling process also can be realized by modeling process well-known to those skilled in the art, not do qualification in this present invention.

S22, determine residual error data one time by original GM (1,1) model value, and utilize a described residual error data to set up a Residual GM (1,1) model;

A described residual error data can obtain by the difference of raw data and GM (1,1) model value.

S23, original GM (1,1) model is once revised, set up and revise GM (1,1) model, and calculate and revise GM (1,1) model value according to a Residual GM (1,1) model;

In this step, the process of setting up correction GM (1,1) model is not done once more and is given unnecessary details with the similar process of setting up GM (1,1) model among the above-mentioned step S1.

S24, determine the quadratic residue data, and utilize described quadratic residue data to set up the BP neural network model by revising GM (1,1) model value.

The principle of BP neural network model is: for time series, must there be certain funtcional relationship (linearity or nonlinear) in past value constantly with following value constantly, uses function table x _T+1=F (x _t, x _T-1..., x _T-n), go to approach this funtcional relationship with neural network model, can fully improve the nonlinear prediction precision of data.

Wherein, described quadratic residue data obtain by raw data and the difference of revising GM (1,1) model value.

The relative error of S12, the described forecast model of calculating;

Mainly comprise following substep:

A2, obtain the true predictive value of revising GM (1,1) model;

This process mainly comprises:

Extrapolate to revising GM (1,1) model, obtain generating the number predicted value;

Carry out once tired subtracting to generating the number predicted value, obtain to revise the true predictive value of GM (1,1) model.

B2, obtain the true predictive value of BP neural network model;

The prediction of BP neural network can be adopted iteration one-step prediction method, can carry out medium-and long-term forecasting, its forecasting process is, the predicted value of network output is fed back to network as the input data that are used for next step prediction, concrete example is in this way: with the quadratic residue data as first constantly the input data, prediction second predicted value constantly;

Second constantly the predicted value as the input data, is predicted the 3rd constantly predicted value;

By that analogy, finally obtain the true predictive value of BP neural network model.

C2, stack two true predictive values obtain forecast model true predictive value;

D2, the true predictive value of utilizing described forecast model and described former water number it is calculated that the relative error of described forecast model.

S13, by the checking, when the relative error of described forecast model is less than or equal to predetermined relative error, determine that described forecast model is a valid model;

Otherwise, then need to rebulid model.

S14, carry out water quality prediction according to described forecast model.

For the easier quilt of technical solution of the present invention is understood, describe in detail with a specific embodiment below.

In the present embodiment, technical scheme of the present invention is divided into two parts is described, comprising: set up the process of forecast model and the process of checking forecast model.

At first be the process of setting up forecast model, specifically be exemplified below:

Suppose certain water quality time series x ⁽⁰⁾=(x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n)), (1) n is a positive integer, and modeling is as follows:

1) data is carried out one-accumulate

Sequence after adding up is:

x ⁽¹⁾＝(x ⁽¹⁾(1)，x ⁽¹⁾(2)；..，x ⁽¹⁾(n))

＝(x ⁽⁰⁾(1)，x ⁽⁰⁾(1)+x ⁽⁰⁾(2)，...，x ⁽⁰⁾(1)+x ⁽⁰⁾(2)+...+x ⁽⁰⁾(n))

＝(x ⁽⁰⁾(1)，x ⁽¹⁾(1)+x ⁽⁰⁾(2)；..，x ⁽¹⁾(n-1)+x ⁽⁰⁾(n))， (2)

2) the albefaction form differential equation of GM (1,1)

Data after adding up are carried out modeling, and by gray theory as can be known, the differential equation is:

$\frac{{\mathrm{dx}}^{\left(1\right)}}{\mathrm{dt}}+{\mathrm{ax}}^{\left(1\right)}=u,$ A wherein, u is an equation parameter;

3) equation parameter of differentiating

The note parameter list is

Try to achieve by least square method

In the formula,

y _N=[x ⁽⁰⁾(2), x ⁽⁰⁾(3) ..., x ⁽⁰⁾(n)] ^T: (3)

Model background value wherein

K=1,2 ..., n-1; According to pertinent literature, the background value after the improvement:

K=2,3 ..., n wherein works as x ⁽¹⁾(k)=x ⁽¹⁾(k-1) time, z ⁽¹⁾(k)=x ⁽¹⁾(k-1).

4) ask GM (1, the 1) differential equation to separate

Integrating step 2), 3) can solve the differential equation separate into:

$\widehat{x^{\left(1\right)}}(k+1)={\mathrm{ce}}^{-\mathrm{ak}}+\frac{u}{a},---\left(4\right)$

Wherein c is a constant.In order more to tally with the actual situation, existing by

Starting condition change into

Finally solving this solution of equation is:

$x^{\left(1\right)}(k+1)=(x^{\left(0\right)}\left(n\right)-\frac{u}{a})e^{-\mathrm{ak}}{e}^{\mathrm{an}}+\frac{u}{a},---\left(5\right)$

K=0,2 ..., n-1, a wherein, u is an equation parameter.

5) set up generation number Residual GM (1,1) model

Generating the number residual sequence is: $e^{\left(1\right)}=(e^{\left(1\right)}\left(k\right)=x^{\left(1\right)}\left(k\right)-{\hat{x}}^{\left(1\right)}\left(k\right),k=\mathrm{1,2},...,m),---\left(6\right)$

In order to make residual sequence satisfy the condition of GM modeling, must all transfer residual values to non-negatively earlier, use here with adding in the residual sequence minimum value and (generally bear, be made as e ⁽¹⁾ _Min) absolute value; Afterwards, utilize non-negative residual error data to set up to generate number residual error model separate for:

$\widehat{e^{\left(11\right)}}(k+1)=(e^{\left(1\right)}\left(n\right)-\frac{u_e}{a_e})e^{-a_{e}k}{e}^{a_{e}n}+\frac{u_e}{a_e},---\left(7\right)$

Must generate number residual error model behind once tired the subtracting is:

$\widehat{e^{\left(1\right)}}(k+1)=\widehat{e^{\left(11\right)}}(k+1)-\widehat{e^{\left(11\right)}}\left(k\right)=(1-e^{a_e})(e^{\left(1\right)}\left(n\right)-\frac{u_e}{a_e})e^{-a_{e}k}{e}^{a_{e}n},---\left(8\right)$

At last, to this residual error model value

Deduct | e _Min| obtain real residual error model value.

6) revise GM (1,1) model

Be added to and generate digital-to-analogue pattern (5) and deduct again generating number residual error modular form (8) | e _Min|, obtain revised GM (1,1) model and be:

$\widehat{x^{\left(1\right)}}(k+1)=(x^{\left(0\right)}\left(n\right)-\frac{u}{a})e^{-\mathrm{ak}}{e}^{\mathrm{an}}+\frac{u}{a}+(1-e^{a_e})(e^{\left(1\right)}\left(n\right)-\frac{u_e}{a_e})e^{-a_{e}k}{e}^{a_{e}n}-\left|e_{\min }\right|---\left(9\right)$

7) data are once tired subtracts, and asks model value

By step 2) as can be known, step 6) is obtained is correction model value after raw data adds up, so need carry out reverse operating promptly once tired subtracting to these data:

$\widehat{x^{\left(0\right)}}\left(1\right)=\widehat{x^{\left(1\right)}}\left(1\right)-\widehat{x^{\left(1\right)}}\left(0\right),$

$\widehat{x^{\left(0\right)}}\left(2\right)=\widehat{x^{\left(1\right)}}\left(2\right)-\widehat{x^{\left(1\right)}}\left(1\right),$

$\widehat{x^{\left(0\right)}}\left(3\right)=\widehat{x^{\left(1\right)}}\left(3\right)-\widehat{x^{\left(1\right)}}\left(2\right),$

...

$\widehat{x^{\left(0\right)}}\left(n\right)=\widehat{x^{\left(1\right)}}\left(n\right)-\widehat{x^{\left(1\right)}}(n-1)$

The model value sequence of finally trying to achieve is: $\widehat{x^{\left(0\right)}}=(\widehat{x^{\left(0\right)}\left(1\right)},\widehat{x^{\left(0\right)}\left(2\right)},...,\widehat{x^{\left(0\right)}\left(n\right)})---\left(10\right)$

8) the quadratic residue data are set up the BP neural network model.

Referring to BP neural network structure figure shown in Figure 3, the BP neural network comprises input layer, hidden layer and output layer, and wherein input layer and hidden layer are necessary layers, and the input layer number is the input number of samples, and the output layer neuron number is the output sample number; Hidden layer is a variable layer, needs hidden layer generally speaking, and its neuron number is determined by particular problem, needs could determine after constantly adjustment reaches precision.In the embodiment of the invention, be to predict next output valve constantly according to preceding s input sample, so the input neuron number is s, the output neuron number is 1, as shown in Figure 3.Wherein, the input sample is residual sequence E=(e ₁, e ₂..., e _n) combination, i.e. e _I-(s-1), e _I-(s-2)..., e _i, i=s, s+1 ..., n-1, total n-s group input sample data, the output sample sequence is e _S+1, e _S+2..., e _n, n-s organizes altogether.Constantly adjust weights W through the BP neural network BP training algorithm _IjWith threshold value Vj and hidden layer neuron number, reach the final weights W of determining behind the model accuracy _IjValue with threshold value Vj and hidden layer neuron number.

Suppose that the quadratic residue data are E=(e ₁, e ₂..., e _n), the quadratic residue data can be by computing formula: residual error=raw data-correction GM (1,1) model value calculates, present embodiment for example, and the data sequence that can be deducted formula (1) by the data sequence of formula (10) obtains.Simultaneously, predict next value constantly with s value constantly, then funtcional relationship becomes: e _I+1=F (e _I-(s-1), e _I-(s-2) ..., e _i), i=s, s+1 ..., n-1, total n-s group sample data referring to Fig. 4, is the training sample (residual sample) of BP neural network, comprises the input sample, network model value and desired output.The input sample enters from input layer, the network model value is the continuous calculating back acquisition of input sample through network, and next of desired output input sample is worth constantly, is the comparison other of BP neural network model value, BP neural network model value need reach desired output as far as possible, and this network is just calculated convergence.Again as shown in Figure 3, neural network comprises input layer, hidden layer and output layer, present embodiment is that a preceding s water quality data is predicted next following value constantly, so know the input layer of this neural network by inference s neuron arranged, output layer has 1 neuron, and hidden layer hierachy number and neuron number thereof need to decide according to practical application, can constantly adjust in the network training process.

When using neural metwork training, at first whole samples are made normalized, get one group of sample then at random and offer network, calculate the neuronic weights and the threshold value of each hidden layer and output layer, next sample of picked at random carries out same training then, finishes up to n-s sample training, chooses one group of sample training again from n-s sample, up to network global error E less than predetermined minimal value, i.e. network convergence; If frequency of training is greater than predefined value, then network can't be restrained, and after training finishes, gets a model sequence:

Be the process of checking forecast model then, this process need obtains to revise GM (1,1) the true predictive value of model, true predictive value with the BP neural network model, again two true predictive values are superposeed and obtain the true predictive value of forecast model, calculate the relative error of forecast model then,, be proved to be successful when described relative error during smaller or equal to predetermined relative error.

In the present embodiment, the true predictive value of revising GM (1,1) model can draw by following step:

Correction GM (1,1) model by step 6) is extrapolated, that is,

$\widehat{x^{\left(1\right)}}(k+1)=(x^{\left(0\right)}\left(n\right)-\frac{u}{a})e^{-\mathrm{ak}}{e}^{\mathrm{an}}+\frac{u}{a}+(1-e^{a_e})(e^{\left(1\right)}\left(n\right)-\frac{u_e}{a_e})e^{-a_{e}k}{e}^{a_{e}n}-\left|e_{\min }\right|,---\left(11\right)$

K=n wherein, n+1 ...;

Carry out once tired subtracting to generating the number predicted value again, obtain last true predictive value.

Iteration one-step prediction method is adopted in the prediction of BP neural network model, referring to Fig. 5, is the synoptic diagram of BP neural network iteration one-step prediction method.Its process is: the predicted value of network output fed back to network is used for next step prediction as input, that is, and when predicting, quadratic residue data e _N-s+1..., e _N-1, e _nAs the input data, output be next predicted value of n+1 constantly

And

As import data next time, predict n+2 value constantly

Other obtain the predicted value of BP neural network model by that analogy at last.

At last, superpose to revising GM (1,1) predicted value and BP Neural Network model predictive value, obtain the true predictive value of forecast model, (relative error is by the absolute value of the difference of raw data and true predictive value to calculate the relative error of forecast model again, obtain divided by raw data again), by relatively, when the relative error of forecast model during smaller or equal to predetermined relative error, illustrate that this forecast model is a valid model, can put into production, otherwise, need modeling again.

The embodiment of the invention also provides a kind of water quality prediction device, and the apparatus structure synoptic diagram referring to shown in Figure 6 comprises: acquiring unit 601, model unit 602, computing unit 603, test form unit 604 and predicting unit 605.Wherein:

Acquiring unit 601 is used to obtain former water number certificate;

Model unit 602 is used for according to described former water number according to setting up forecast model;

Computing unit 603 is used to calculate the relative error of described forecast model;

Test form unit 604, be used to verify when the relative error of described forecast model is less than or equal to predetermined relative error, determine that described forecast model is a valid model;

Predicting unit 605 is used for carrying out water quality prediction according to described forecast model

Modeling unit 604 specifically comprises:

Master pattern subelement 606 is used for according to former water number according to setting up original GM (1,1) model, and calculates original GM (1,1) model value;

Residual error model subelement 607 is used for determining residual error data one time by original GM (1,1) model value, and utilizes a described residual error data to set up a Residual GM (1,1) model;

Correction model subelement 608 is used for according to a Residual GM (1,1) model original GM (1,1) model once being revised, and sets up and revises GM (1,1) model, and calculate and revise GM (1,1) model value;

Network model subelement 609 is used for determining the quadratic residue data by revising GM (1,1) model value, and utilizes described quadratic residue data to set up the BP neural network model.

Referring to Fig. 7, computing unit 603 also comprises:

First acquiring unit 701 is used to obtain the true predictive value of revising GM (1,1) model;

Second acquisition unit 702 is used to obtain the true predictive value of BP neural network model;

The function of this unit specifically comprises:

With the end data of quadratic residue data as first constantly the input data, in order to predict second constantly the predicted value;

With second constantly the predicted value as the input data, in order to predict the 3rd constantly predicted value;

By that analogy, finally obtain the true predictive value of described BP neural network model.

Superpositing unit 703, two predicted values that are used to superpose obtain the true predictive value of forecast model;

Computation subunit 704 is used to utilize the true predictive value of described forecast model and the relative error that described former water number it is calculated that described forecast model.

Wherein, first acquiring unit 701 also comprises:

The extrapolation unit is used for extrapolating to revising GM (1,1) model, obtains generating the number predicted value;

Tired subtract the unit, be used for carrying out once tired subtracting, obtain to revise the true predictive value of GM (1,1) model generating the number predicted value.

The embodiment of the invention utilizes GM (1,1) model and BP neural network model to realize combined type stack forecast model, and raw water quality is predicted.Because the influence factor of water quality data is a lot, the relation that will find out accurately between the various factors is quite difficult, again because water quality data is the growth property time series that changes (grey characteristics) within the specific limits, at these characteristics and consider GM (1,1) applicability of gray theory forecast model, the present invention adopts the basic model single argument single order GM (1 in the gray theory forecast model, 1) earlier the water quality time series is carried out modeling, improved the starting condition of the background value and the model of model simultaneously, utilize a residual error data to set up Residual GM (1 again, 1) model, revise the GM (1 of raw data, 1) model, and predict, can know the general trend of change of water quality from this correction model; But this correction model can't reach precision of prediction, mainly be because water quality not only has grey characteristics, the time growth property, but also have certain fluctuation, fluctuation itself has comprised nonlinear transformations (water quality data is subjected to the influence of various physical reactions and chemical reaction) again, so in order effectively to handle nonlinear transformations, the present invention is revising GM (1,1) on the model based, adopt the BP neural network that the water quality nonlinear characteristic is compensated again, promptly use the BP neural network to revising GM (1,1) (actual value deducts and revises GM (1 residual error data, 1) model value) train, acquisition can be handled the BP neural network prediction model of the quadratic residue data of nonlinear characteristic; Finally, the present invention is to revising GM (1,1) predicted value of model and BP neural network model superposes, obtain the predicted value of forecast model, by calculating the relative error of forecast model, and this error compared with predetermined relative error, when this error is less than or equal to predetermined relative error, determine that described forecast model is a valid model.By the present invention, make water quality prediction reach the better prediction effect, improved precision of prediction greatly.Simultaneously, compare with residual error correction neural network combined type model etc. with in the past single GM (1,1) forecast model or neural network model and original GM (1,1) and have certain innovation.

Above-mentioned is description to various embodiments of the present invention.Each embodiment adopts the mode of going forward one by one to describe in this instructions, and what each embodiment stressed all is and the difference of other embodiment that identical similar part is mutually referring to getting final product between each embodiment.For the disclosed device of embodiment, because it is corresponding with the embodiment disclosed method, so description is fairly simple, relevant part partly illustrates referring to method and gets final product.

It will be understood by those skilled in the art that and to use many different technologies and in the technology any one to come expression information, message and signal.For example, the message of mentioning in the above-mentioned explanation, information can be expressed as voltage, electric current, electromagnetic wave, magnetic field or magnetic particle, light field or above combination in any.

The professional can also further recognize, the unit and the algorithm steps of each example of describing in conjunction with embodiment disclosed herein, can realize with electronic hardware, computer software or the combination of the two, for the interchangeability of hardware and software clearly is described, the composition and the step of each example described prevailingly according to function in the above description.These functions still are that software mode is carried out with hardware actually, depend on the application-specific and the design constraint of technical scheme.The professional and technical personnel can use distinct methods to realize described function to each specific should being used for, but this realization should not thought and exceeds scope of the present invention.

The method of describing in conjunction with embodiment disclosed herein or the step of algorithm can directly use the software module of hardware, processor execution, and perhaps the combination of the two is implemented.Software module can place the storage medium of any other form known in random access memory (RAM), internal memory, ROM (read-only memory) (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or the technical field.

To the above-mentioned explanation of the disclosed embodiments, make this area professional and technical personnel can realize or use the present invention.Multiple modification to these embodiment will be conspicuous concerning those skilled in the art, and defined herein General Principle can realize under the situation that does not break away from the spirit or scope of the present invention in other embodiments.Therefore, the present invention will can not be restricted to these embodiment shown in this article, but will meet and principle disclosed herein and features of novelty the wideest corresponding to scope.

Claims (10)

1. a water quality prediction method is characterized in that, comprising:

Obtain former water number certificate, and according to described former water number according to setting up forecast model;

Calculate the relative error of described forecast model, when the relative error of described forecast model was less than or equal to predetermined relative error, described forecast model was a valid model, and carried out water quality prediction according to described forecast model;

The described forecast model of setting up comprises:

According to setting up original GM (1,1) model, and calculate original GM (1,1) model value according to former water number;

Determine residual error data one time by original GM (1,1) model value, and utilize a described residual error data to set up a Residual GM (1,1) model;

According to a Residual GM (1,1) model original GM (1,1) model is once revised, set up and revise GM (1,1) model, and calculate and revise GM (1,1) model value;

Determine the quadratic residue data by revising GM (1,1) model value, and utilize described quadratic residue data to set up the BP neural network model.

2. method according to claim 1 is characterized in that, a described residual error data obtains according to the difference with GM (1,1) model value by former water number.

3. method according to claim 1 is characterized in that, described quadratic residue data obtain according to the difference with correction GM (1,1) model value by former water number.

4. method according to claim 1 is characterized in that, and is described: calculate the relative error of described forecast model, specifically comprise:

Obtain the true predictive value of revising GM (1,1) model;

Obtain the true predictive value of BP neural network model;

Two predicted values that superpose obtain the true predictive value of forecast model;

Utilize the true predictive value of described forecast model and the relative error that described former water number it is calculated that described forecast model.

5. method according to claim 4 is characterized in that, the described true predictive value of revising GM (1,1) model of obtaining specifically comprises:

Extrapolate to revising GM (1,1) model, obtain generating the number predicted value;

Carry out once tired subtracting to generating the number predicted value, obtain to revise the true predictive value of GM (1,1) model.

6. method according to claim 4 is characterized in that, the true predictive value of the described BP of obtaining neural network model specifically comprises:

With the end data of quadratic residue data as first constantly the input data, in order to predict second constantly the predicted value;

With second constantly the predicted value as the input data, in order to predict the 3rd constantly predicted value;

By that analogy, finally obtain the true predictive value of described BP neural network model.

7. a water quality prediction device is characterized in that, comprising:

Acquiring unit is used to obtain former water number certificate;

Model unit is used for according to described former water number according to setting up forecast model;

Computing unit is used to calculate the relative error of described forecast model;

Test form unit, be used to verify when the relative error of described forecast model is less than or equal to predetermined relative error, determine that described forecast model is a valid model;

Predicting unit is used for carrying out water quality prediction according to described forecast model;

Described modeling unit specifically comprises:

The master pattern subelement is used for according to former water number according to setting up original GM (1,1) model, and calculates original GM (1,1) model value;

Residual error model subelement is used for determining residual error data one time by original GM (1,1) model value, and utilizes a described residual error data to set up a Residual GM (1,1) model;

The correction model subelement is used for according to a Residual GM (1,1) model original GM (1,1) model once being revised, and sets up and revises GM (1,1) model, and calculate and revise GM (1,1) model value;

The network model subelement is used for determining the quadratic residue data by revising GM (1,1) model value, and utilizes described quadratic residue data to set up the BP neural network model.

8. device according to claim 7 is characterized in that, described computing unit comprises:

First acquiring unit is used to obtain the true predictive value of revising GM (1,1) model;

Second acquisition unit is used to obtain the true predictive value of BP neural network model;

Superpositing unit, two predicted values that are used to superpose obtain the true predictive value of forecast model;

Computation subunit is used to utilize the true predictive value of described forecast model and the relative error that described former water number it is calculated that described forecast model.

9. device according to claim 8 is characterized in that, described first acquiring unit comprises:

The extrapolation unit is used for extrapolating to revising GM (1,1) model, obtains generating the number predicted value;

Tired subtract the unit, be used for carrying out once tired subtracting, obtain to revise the true predictive value of GM (1,1) model generating the number predicted value.

10. device according to claim 8 is characterized in that, the effect of described second acquisition unit is specially:

With the end data of quadratic residue data as first constantly the input data, in order to predict second constantly the predicted value;

With second constantly the predicted value as the input data, in order to predict the 3rd constantly predicted value;

By that analogy, finally obtain the true predictive value of described BP neural network model.

Images