CN103822960B - Polarography many concentration of metal ions online test method - Google Patents

Abstract

The present invention relates to concentration of metal ions detection field, more specifically relate to polarography many concentration of metal ions online test method.The present invention proposes the real time measure that a kind of improvement Wavelet Neural Network Method based on state branching algorithm (STA) is used for many concentration of metal ions.First the method makes Polarographic Curves, then adopts wavelet transform to ask for the first order derivative of polarographic signals, extracts corresponding unique point with this, unique point quantity by 3 times of survey species of metal ion quantity, as the input of wavelet neural network.When using training set to train wavelet neural network, in order to make network model more accurate, state branching algorithm is used for the optimization of wavelet neural network parameter, local extremum of having avoided network to be absorbed in.Finally, the real-time online of training the network model obtained to be used for many concentration of metal ions is detected.The present invention tests algorithm for the zinc of reality, cobalt polarogram overlapped signal, and the result obtained obviously is better than traditional curve and BP neural network algorithm.

Description

Polarography many concentration of metal ions online test method

Technical field

The present invention relates to concentration of metal ions detection field, more specifically relate to a kind of polarography many concentration of metal ions online test method.

Background technology

In zinc wet smelting process, there is other plurality of impurities metallic ions.These foreign metal ions are by adding unclassified stores to remove in purification process, material adds too much can cause the wasting of resources; Deficiency of adding material will cause foreign metal ion remaval not thorough, there is very large harm by follow-up electrolysis process.For a long time, in zinc wet smelting process, many concentration of metal ions adopt the method for artificial offline inspection, this testing process is complicated and retardation time is long, can not feed back concentration information in time, cause the wasting of resources, pollutant emission exceeds standard and the serious consequence such as energy luxus consumption.Therefore, the on-line analysis of the many metal components for hydrometallurgy process need be improved, in order to optimizing process parameter and realize energy-saving and emission-reduction consumption reduction.

Polarographic method is used for the detection of concentration of metal ions, being the concentration that volt-ampere characteristic by producing calculates detected ion, not needing the conversion carrying out signal, just can direct record, has highly sensitive, that speed is fast, easy to operate feature.But in the process of the many concentration of metal ions of polarography on-line checkingi simultaneously, for the component that characteristic is similar, because exciting voltage is close to obtaining overlapping signal; In addition, the bulk composition in polycomponent due to its concentration high, also there is interference by other signal in the signal of generation, also will produce the superposition of signal.Traditional chemical is measured and normal is adopted loaded down with trivial details tediously long Chemical Decomposition, shelter and improve instrument the method such as resolution to obtain the relevant information of one-component.But under many circumstances, the resolution only with chemical method and raising instrument is differentiated very difficult often to multicomponent system.

Summary of the invention

(1) technical matters that will solve

The technical problem to be solved in the present invention is exactly realize gained signal when many concentration of metal ions real-time online detects for polarography to there is the problem of overlap peak, proposes a kind of improvement Wavelet Neural Network Method based on state branching algorithm (STA) and measures for the real-time online of many concentration of metal ions.

(2) technical scheme

First the inventive method makes Polarographic Curves, then adopts wavelet transform to ask for the first order derivative of gained polarogram overlapped signal, extracts the input of corresponding unique point as wavelet neural network using this.Then employing wavelet neural network sets up the relation between polarogram current signal and concentration of metal ions.When using training set to train wavelet neural network, in order to make network model more accurate, then employing state branching algorithm (STA) is for the off-line training process of wavelet neural network parameter, local extremum of having avoided network to be absorbed in.Finally, the real-time online of training the network model obtained to be used for many concentration of metal ions is detected.

First the inventive method obtains corresponding eigenwert from the polarographic signals collected.When adopting polarography to carry out ion concentration detection, in the voltage range of scanning, generally will adopt several ten thousand data, if using these data all as the input of wavelet neural network, calculation of complex certainly will be made, and there is information redundancy.The principle of carrying out ion concentration detection by polarography is known, and ion concentration is closely related with the size of current at spectrum peak place.Therefore, the input of relevant extreme point as wavelet neural network of peak-to-peak signal and the first order derivative thereof occurred in origin pole spectrum signal is chosen.

Corresponding unique point described in the method chooses the extreme point that corresponding to of occurring in the Polarographic Curves peak point that institute surveys species of metal ion quantity and quantity in its corresponding first order derivative curve are above-mentioned peak point 2 times, using above-mentioned total quantity by the unique point of survey species of metal ion quantity 3 times as the input of wavelet neural network.

Wavelet transform is adopted to ask for the first order derivative of signal, when namely adopting a certain wavelet mother function δ (x) to carry out the decomposition of j layer to polarographic signals, the decomposition marble collection of generation with between there is very little skew, according to discrete data differential formulas, utilize this skew to ask the approximate derivative of each corresponding point, obtain signal C ⁽⁰⁾first order derivative X ⁽¹⁾, its expression formula can be write as into: $X^{\left(1\right)}=C_{D2(m+1)}^{\left(j\right)}-C_{D_{2m}}^{\left(j\right)}.$

After obtaining corresponding eigenwert by foregoing description, using the input of income value as wavelet neural network.The wavelet neural network that described method adopts is based on wavelet transformation theory, replaces common neuron nonlinear activation function and the neural network model that forms with nonlinear wavelet base.This wavelet neural network needs the parameter of training to comprise weight factor W _k, the contraction-expansion factor a of wavelet basis _kwith shift factor b _k.Classic method utilizes gradient descent method to train network, and due to comparatively large to the dependence of initial network parameter, algorithm is easily absorbed in local optimum, is difficult to reach global optimum.Therefore, in order to find global optimum rapidly, the present invention adopts the structural parameters of state branching algorithm (STA) Optimization of Wavelet neural network.

Described metallic ion is the metallic ion existed in zinc hydrometallurgy process.

(3) beneficial effect

Method proposed by the invention is adopted to carry out training and testing, and contrast with the result adopting traditional gaussian curve approximation method and BP neural network algorithm to obtain, this method has better measuring accuracy compared with other two kinds of methods, and the result obtained obviously is better than traditional curve fitting method and BP neural network algorithm.

Accompanying drawing explanation

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.

Fig. 1 is zinc, cobalt coexists linear polarogram sweep current-voltage curve;

The first order derivative curve map of Fig. 2 is zinc, cobalt coexists linear polarogram sweep current-voltage curve;

Fig. 3 is zinc, cobalt independent and the linear polarogram sweep current-voltage curve that coexists.

Embodiment

Below in conjunction with drawings and Examples, embodiments of the present invention are described in further detail.Following examples for illustration of the present invention, but can not be used for limiting the scope of the invention.

State branching algorithm (STA) carries out random search reach optimizing object based on the thought of state space transfer.Similar with genetic algorithm, state branching algorithm adopts alternative manner to process equally, and its searching process is state transfer, and the process upgrading optimum solution is instantly state migration procedure.State migration procedure has following form:

$\left\{\begin{array}{c}{x}_{k+1}=A_k{x}_k+B_k{u}_k\\ y_{k+1}=f\left(x_{k+1}\right)\end{array}\right.---\left(1\right)$

Wherein, x _k∈ R ⁿrepresent a state, corresponding to a solution of optimization problem; A _k∈ R ^{n × n}and B _k∈ R ^{n × m}be state-transition matrix, the operation operator of optimized algorithm can be considered to, u _k∈ R ^mbe the function relevant to state xk and historic state, and f is fitness computing function.

State branching algorithm mainly contains following core operation: rotate (Rotationtransformation), transfer (Translationtransformation), expansion (Expansiontransformation), translation (Axesiontransformation).

(1) rotation operator

$x_{k+1}=x_k+\mathrm{\α}\frac{1}{n{\left|\right|x_k\left|\right|}_2}{R}_r{x}_r---\left(2\right)$

Wherein α is a positive number, is called twiddle factor; be obey equally distributed stochastic matrix between [-1,1], this operator makes searching algorithm think x at one _kcenter, α be radius hypersphere in carry out optimizing.

(2) transition operator

$x_{k+1}=x_k+{\mathrm{\βR}}_t\frac{x_k-x_{k-1}}{{\left|\right|x_k-x_{k-1}\left|\right|}_2}---\left(3\right)$

Wherein, β is a positive number, is called transfer factor; the random number between [0,1]. this operator makes algorithm prolong to search for along gradient direction, and step-size in search is β to the maximum.

(3) enlargement oprator

x _k+1=x _k+γR _ex _k（4）

Wherein, γ is a positive number, is called spreading factor. it is the diagonal matrix (what adopt in the text is obey standardized normal distribution) of Gaussian distributed.Enlargement oprator can carry out expanded search in whole search volume.

(4) translation operator

x _k+1=x _k+δR _ax _k（5）

Wherein, δ is a positive number, is called shift factor; be the diagonal matrix of Gaussian distributed, in matrix, only have the element on a random site non-vanishing.Translation operator can be searched for along axle, thus improves one-dimensional search dynamics.

Wavelet neural network is based on wavelet transformation theory, replaces common neuron nonlinear activation function and the neural network model formed with nonlinear wavelet base.Its corresponding input layer replaced by the warp parameter a of wavelet function and translation parameters b respectively to the weights of hidden layer and hidden layer threshold output valve, and output layer is generally linear neuron, here adopt Morlet morther wavelet, its system of being stretched by the small echo of hidden layer carries out linear superposition and forms Output rusults.The output expression formula of this wavelet neural network is shown in formula (6), wherein for wavelet basis, W _kweight factor, a _k, b _kcontraction-expansion factor and the shift factor of wavelet basis respectively.

$\stackrel{\‾}{y}\left(x\right)=\underset{k=1}{\overset{k}{\mathrm{\Σ}}}{W}_{k}h\left(\frac{x-b_k}{a_k}\right)---\left(6\right)$

This wavelet neural network needs the parameter of training to comprise weight factor W _k, the contraction-expansion factor a of wavelet basis _kwith shift factor b _k.Classic method utilizes gradient descent method to train network, and due to comparatively large to the dependence of initial network parameter, algorithm is easily absorbed in local optimum, is difficult to reach global optimum.Therefore, in order to find global optimum rapidly, adopt the structural parameters of state branching algorithm (STA) Optimization of Wavelet neural network herein.

The present invention is to detect zinc, concentration of cobalt ions in zinc hydrometallurgy process simultaneously, four extreme points (two maximum values, two minimal values, C, D, E, F as shown in Figure 2 tetra-points) choosing peak point (A, B as shown in Figure 1 two points) in Polarographic Curves and its corresponding first order derivative are totally six unique points, input as wavelet neural network.

Asking in first order derivative process, along with the raising of differentiate order, there is the shortcoming that noise is exaggerated in common derivative method, wavelet transformation is applied to signal derivation and can overcomes this deficiency well.For this reason, wavelet transform is adopted to ask for the first order derivative of signal.The principle that wavelet transform is used for signal derivation is when adopting a certain wavelet mother function δ (x) to carry out the decomposition of j layer to polarographic signals, the decomposition marble collection of generation with between there is very little skew, according to discrete data differential formulas, this skew just can be utilized to ask the approximate derivative of each corresponding point.Therefore signal C can be obtained ⁽⁰⁾first order derivative X ⁽¹⁾, its expression formula can be write as into:

$X^{\left(1\right)}=C_{D2(m+1)}^{\left(j\right)}-C_{D_{2m}}^{\left(j\right)}---\left(7\right)$

Wavelet transform is adopted to carry out differentiate gained derivative curve as shown in Figure 2 to the superimposed curves of zinc, cobalt ions in Fig. 1.

After obtaining six eigenwerts by foregoing description, using the input of income value as wavelet neural network.The wavelet neural network that described method adopts is based on wavelet transformation theory, replaces common neuron nonlinear activation function and the neural network model that forms with nonlinear wavelet base.This wavelet neural network needs the parameter of training to comprise weight factor W _k, the contraction-expansion factor a of wavelet basis _kwith shift factor b _k.Classic method utilizes gradient descent method to train network, and due to comparatively large to the dependence of initial network parameter, algorithm is easily absorbed in local optimum, is difficult to reach global optimum.Therefore, in order to find global optimum rapidly, the present invention adopts the structural parameters of state branching algorithm (STA) Optimization of Wavelet neural network.

Embodiment

Lower mask body is to detect zinc, concentration of cobalt ions in zinc hydrometallurgy process simultaneously, and adopt the accurate polarographic analyze of the JP-06A type of Chengdu Instruement Factory to detect, the polarographic signals obtained as shown in Figure 3.Experimental technique is the preparation zinc of series concentration and the blending ingredients solution (wherein zinc ion concentration scope 0.17 ~ 0.85g/L, concentration of cobalt ions scope is 0.002mg ~ 0.2mg/L) of cobalt.Testing environment is: solution temperature 20 DEG C, PH=8.26.Scan within the scope of-0.9 ~ 1.4V, record 23000 current values altogether.Be configured with altogether 14 groups of mixed solutions by gradient, train with wherein 9 groups (see table 1), obtain the correlation parameter of wavelet neural network as off-line training collection.All the other 5 groups are carried out testing (see table 2) as online data.

Table 1 training set

Table 2 test set

By the on-line checkingi of state transfer-wavelet neural network (STA-Wavenet) for ion concentration, first need to carry out off-line training and obtain a wavelet-neural network model determined, then the network trained is used for the on-line checkingi of ion concentration.First the method obtains required network parameter by off-line training wavelet neural network, then the network trained is used for the on-line checkingi of ion concentration.Off-line training process is 4 extreme points in the first order derivative curve first determining that 2 peak points in primary curve and discrete wavelet calculate, using the input of the result after process as wavelet neural network, various ion concentration is as the output of network, set up and relation in polycomponent between each ion concentration, using the network parameter values that obtains after training as the network model of on-line analysis, thus realize simple and effective on-line analysis process.

Adopt method (STA-Wavenet) proposed by the invention to carry out training and testing, and contrast with the result adopting traditional gaussian curve approximation method and BP neural network algorithm to obtain, three kinds of method application result contrasts are as shown in table 3.The iterations that wherein BP neural network algorithm and method proposed by the invention are in operation is 1000 times, and the parameter alpha value in state branching algorithm is that β, γ, δ get fixed value 1 from 1 to 1e-4 linear decrease.Can find out that this method can have better measuring accuracy compared with other two kinds of methods from the result table, be more suitable for many concentration of metal ions polarography on-line checkingi with overlap peak.

The test result of table 3 curve, BP neural network, STA-Wavenet

Above embodiment is only for illustration of the present invention, but not limitation of the present invention.Although with reference to embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that, various combination, amendment or equivalent replacement are carried out to technical scheme of the present invention, do not depart from the spirit and scope of technical solution of the present invention, all should be encompassed in the middle of right of the present invention.

Claims (2)

1. polarography many concentration of metal ions online test method, it is characterized in that, first make Polarographic Curves, then adopt wavelet transform to ask for the first order derivative of gained polarogram overlapped signal, extract the input of corresponding unique point as wavelet neural network using this; Then employing wavelet neural network sets up the relation between polarogram current signal and concentration of metal ions; Then adopt state branching algorithm (STA) for the off-line training process of wavelet neural network parameter; Finally, the real-time online of training the network model obtained to be used for many concentration of metal ions is detected;

Corresponding unique point described in the method chooses the extreme point that corresponding to of occurring in the Polarographic Curves peak point that institute surveys species of metal ion quantity and quantity in its corresponding first order derivative curve are above-mentioned peak point 2 times, using total quantity by the above-mentioned unique point of survey species of metal ion quantity 3 times as the input of wavelet neural network;

Wavelet transform is adopted to ask for the first order derivative of signal, when namely adopting a certain wavelet mother function δ (x) to carry out the decomposition of j layer to polarographic signals, the decomposition marble collection of generation with between there is very little skew, according to discrete data differential formulas, utilize this skew to ask the approximate derivative of each corresponding point, obtain signal C ⁽⁰⁾first order derivative X ⁽¹⁾, its expression formula can be write as into: $X^{\left(1\right)}=C_{D2(m+1)}^{\left(j\right)}-C_{D_{2m}}^{\left(j\right)};$

The wavelet neural network that the method adopts is based on wavelet transformation theory, replaces common neuron nonlinear activation function and the neural network model that forms with nonlinear wavelet base, and this wavelet neural network needs the parameter of training to comprise weight factor W _k, the contraction-expansion factor a of wavelet basis _kwith shift factor b _k; The structural parameters of employing state branching algorithm (STA) Optimization of Wavelet neural network, find global optimum rapidly.

2. polarography many concentration of metal ions online test method according to claim 1, is characterized in that, described metallic ion is the metallic ion existed in zinc hydrometallurgy process.