CN101813932A - Method for component content prediction and optimization operation in wet-process metallurgic extraction process - Google Patents

Abstract

The hydrometallurgical extraction process component content prediction and optimization operation method adopts the hydrometallurgical extraction process of the multi-stage extraction tank, and realizes the real-time prediction of the raffinate component content through the mixed modeling of the hydrometallurgical extraction process. And provide online optimization operation guidance for the extraction process. Including data collection, auxiliary variable selection and standardization processing, establishment of mixed model, correction of mixed model, determination of optimization operation guidance and other steps. The invention can greatly increase the leaching rate, maintain the production in the best damaged state, reduce the consumption of raw materials and energy, and prolong the operation period of the equipment.

Description

Component Content Prediction and Optimal Operation Method in Hydrometallurgical Extraction Process

technical field

The invention belongs to the field of hydrometallurgy, and in particular provides a method for predicting and optimizing the content of extraction components based on a mixed model, that is, providing a method for predicting the concentration of raffinate components in real time and providing guidance for optimal operations.

Background technique

Hydrometallurgy technology is a new technology that is gradually mature and urgently needs industrialization. Compared with traditional pyrometallurgy, hydrometallurgy technology has the advantages of high efficiency, cleanness, and suitable for recovery of low-grade complex metal mineral resources. Especially in view of the characteristics of my country's mineral resources, which are rich in lean ore, complex symbiosis, and high impurity content, the industrialization of hydrometallurgical processes is of great significance for improving the comprehensive utilization of mineral resources, reducing solid waste production, and reducing environmental pollution.

In recent years, research on hydrometallurgical technology and equipment has progressed rapidly. However, the hydrometallurgical process is complicated, the equipment types are diverse, and the process conditions are harsh, such as high temperature, high pressure, strong corrosion, etc. Therefore, the hydrometallurgical process must realize the improvement of the automatic control level of large-scale industrialization in order to ensure safe, stable and continuous production. Only by running can the product quality and output be guaranteed.

The process flow of hydrometallurgical cascade extraction and impurity removal process is shown in Figure 1. The entire extraction production line is composed of multi-stage extraction tanks in series. segment and anti-iron segment. Each stage of extraction tank is composed of a mixing chamber and a clarification chamber. During the extraction and separation process, through the unique structural design of the extraction tank and the power of the stirring motor in the mixing chamber during the extraction process, the organic phase and the water phase are separated in the mixing chamber and clarification chamber. parallel flow, but the whole produces reverse flow. The organic phase always flows from left to right and the aqueous phase always flows from right to left. The solution in the clarification chamber is divided into two layers according to the principle that the organic extractant and water are immiscible: the upper layer is the organic phase, and the lower layer is the water phase. According to the cascade extraction principle, the operator controls the flow of organic, feed liquid, washing liquid, stripping liquid and anti-iron liquid according to a certain balance ratio. The function of the extraction section is to extract most of the impurity metals and a small amount of valuable metals in the aqueous phase into the organic phase; the function of the washing section is to wash back most of the valuable metals through the contact between the washing liquid and the organic phase. The water phase; and the role of the stripping section and the anti-iron section is to return the impurity metals to the water phase, so that the organic phase can be regenerated.

In order to ensure the quality of the product, increase the yield of metals, reduce consumption, and give full play to the production capacity of the equipment, it is necessary to test the concentration of product components in the aqueous phase outlet (raffinate) of the extraction production line during the production process. In actual production, the concentration of the components cannot be measured online, but obtained by offline laboratory analysis, but the offline analysis lags for several hours, and the analysis sampling frequency is small (1 time/day), which is far from meeting the control requirements. There are two ways to solve this problem, one is to use online analyzer; the other is to realize the prediction of component concentration by modeling the process. Since the function of the former is not perfect, and the investment is large and difficult to maintain, it cannot fully meet the continuous on-line detection requirements of the hydrometallurgical extraction and separation production process; therefore, the best solution is to use the second approach, that is, to establish an extraction process group The prediction model of component content can predict the concentration of each component online without increasing investment.

At present, there is no report on the mixing modeling method and optimization operation guidance of the component content in the hydrometallurgical extraction process. The method adopted by the factory is to manually test the concentration of these components and obtain them through off-line analysis. The operator adjusts the flow rate setting value according to the test value of the component concentration to ensure the product purity and material liquid treatment specified in the production target. quantity. The disadvantage of this method is that the artificial assay has a large lag, which can reach several hours; in addition, due to the cost of the assay, the sampling period is long, so it is difficult for these assay values to be directly used for quality control. Operators mainly rely on their own experience to adjust, so that it is difficult to guarantee the first pass rate of the product, the consumption of auxiliary materials increases, and the cost of the product increases.

Contents of the invention

The invention provides a hydrometallurgical extraction process component content prediction and optimization operation method, through the mixed modeling of the hydrometallurgical extraction process, real-time prediction of raffinate component content is realized, and online optimization operation is provided for the extraction process guide.

The object of the present invention is to seek a kind of method for component content prediction and optimal operation in hydrometallurgical extraction separation production process, and it is used to solve following problem:

(1) Provide component content monitoring data for the automatic control of the hydrometallurgical extraction and impurity removal process, and realize the optimal operation guidance for the extraction and impurity removal process;

(2) By simulating the actual fluctuations of variable factors such as feed liquid composition and flow rate, grasp the impact of different fluctuation ranges on product quality, provide timely and reasonable guidance on optimizing operations, ensure product quality, and realize optimal control of the extraction process ;

(3) The soft sensing method of the present invention not only considers the advantages of the mechanism model, but also integrates the characteristics of the data model, and can simulate the production process of extraction and impurity removal, grasp the consumption of auxiliary materials in the production process, and formulate a reasonable production plan.

(4) Replace manual laboratory analysis to achieve the purpose of timely and accurate detection of production status.

The method for predicting and optimizing the content of components in the hydrometallurgical extraction process provided by the present invention includes: (1) process data collection, (2) selection and standardization of auxiliary variables, (3) establishment of a mixing model, (4) mixing Calibration of the model, (5) Determination of the optimization operation guidance and other steps.

(1) Process data acquisition

The device of the present invention includes an extraction process component content prediction and optimization operating system, a host computer, a PLC, and on-site sensing and transmission parts, as shown in FIG. 2 . Among them, the field sensing and transmission part includes pH value, temperature, flow and other detection instruments. The detection instrument is installed on site during the extraction process, and the detection instrument sends the collected signal to the PLC through the Profibus-DP bus, and the PLC regularly transmits the collected signal to the host computer through Ethernet, and the host computer transmits the received data to the extraction process component content prediction With the optimization of the operating system, real-time prediction of the content of raffinate components is performed, and online optimization operation guidance is provided.

The functions of each part of the device of the present invention:

(A) On-site sensing and transmission part: including pH value, temperature, flow and other detection instruments are composed of sensors, responsible for the collection and transmission of process data;

(B) PLC: Responsible for A/D conversion of the collected signal, and transmit the signal to the host computer through Ethernet;

(C) Host computer: collect local PLC data, send it to the extraction process component content prediction and optimization operating system, and provide online optimization operation guidance.

(2) Selection and standardization of auxiliary variables

The auxiliary variables selected by the present invention include,

(A) concentration x ₁ of the extracted component in the feed liquid;

(B) The flow rate x ₂ of feed liquid;

(C) flow rate x ₃ of washing liquid;

(D) Flow rate of organic phase x ₄

(D) pH value x ₅ of raffinate;

(E) Temperature x ₆ of feed liquid.

In order to prevent the impact of each detection variable on the data model due to different units, the collected sensor measurement data is first standardized.

${\stackrel{<span\; class="notranslate">\; ~</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; max</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; min</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

-第i个数据样本，第j个传感器测量值的标准化值；In the formula

- the i-th data sample, the normalized value of the j-th sensor measurement;

x _j (i) - i-th data sample, j-th sensor measurement;

- the minimum value of the measured value of the jth sensor;

-第j个传感器测量值的最大值。

- the maximum value of the jth sensor measurement.

(3) Establishment of mixed model

I. The structure of the mixed model

Mixed use of multiple modeling methods to establish the mathematical model of the object can achieve the effect of various methods complementing each other, and has become a research hotspot at present. If the system has prior physical knowledge that can be used, it should be used as much as possible to transform the black box model into a gray box model, so as to combine the mechanism method with the data method. The data method can extract the complex information inside the object that cannot be explained by the mechanism method, and the mechanism model can improve the generalization ability of the statistical model. Combination methods are generally divided into parallel and serial.

(A) Serial combination method: first use the mechanism method to obtain a model structure with parameters, and then use the data method to determine those parameters;

(B) Parallel combination method: use the data method to determine a compensator to compensate the results obtained by the mechanism model.

The application of prior knowledge improves the accuracy of the model, enhances the generalization ability of the model, reduces the data required for parameter estimation, and reduces the amount of calculation compared with the black-box model established purely based on data.

In many cases, simply using the mechanistic model is not enough to describe all the characteristics of the process. Due to the complicated relationship between the measurable variables and the leading variables in some processes, it is difficult to include them all in the mechanistic model; Factors will also reduce the prediction accuracy of the mechanism model. At this time, the data model can be used to compensate the unmodeled dynamics in the mechanism model to improve the prediction accuracy of the model. Aiming at the characteristics of the extraction process, the present invention adopts the structure of the parallel mixing model, as shown in FIG. 3 . The invention adopts the parallel hybrid model structure based on nonlinear PLS to compensate the unmodeled dynamics in the mechanism model.

II. Mechanism model

After a comprehensive understanding of the reaction mechanism of the process, the relevant balance equations can be written, the mathematical relationship between the unmeasurable leading variable and the measurable secondary variable can be determined, and the mechanism model for estimating the leading variable can be established. Mechanism modeling requires an in-depth understanding of specific objects and a comprehensive grasp of the basic laws involved in the actual process, including the equation of state in thermodynamics, phase balance, reaction kinetics, material balance, energy balance in physical chemistry, and polymer Chemistry and many other aspects of knowledge. Mechanism modeling has achieved success in many aspects. After the mechanism model is built, it can be used to simulate the operation of the actual system, deepen the understanding of the actual process, and improve the operation level; at the same time, through model simulation, it can help to grasp the dynamics of the object. characteristics, laying the foundation for process optimization and control. However, establishing a mechanism model usually requires a lot of energy, and is not suitable for industrial processes whose mechanism is not completely clear. Therefore, the present invention simplifies the mechanism model through certain assumptions.

The mechanism model adopted in the present invention is composed of m+n stage mixing clarifiers, wherein m stage is used for extraction and n stage is used for washing, and its equivalent structure is shown in Fig. 4 . Usually, due to the high efficiency of stripping and anti-iron, the concentration of each metal ion in the fresh organic phase after the stripping section can be regarded as zero. Therefore, in order to simplify the mechanism model structure, only the extraction section and the washing section are considered to be modeled.

According to the material balance relationship, for the i-th stage of the extraction section, i=1, 2,...m-1, there is the following material balance relationship:

Vy _i-1 +(L+L′)x _i+1 ＝Vy _i +(L+L′)x _i (2)

And for the m grade (feed grade) of extraction section, i.e. i=m, there is following material balance relation:

Vy _m-1 +Lx ₀ +L′x _m+1 ＝Vy _m +(L+L′)x _m (3)

For the jth stage of the washing section, j=m+1,...,n-1, there is the following material balance relationship:

Vy _j-1 +L′x _j+1 ＝Vy _j +L′x _j (4)

And for the nth stage of the washing section, i.e. j=n, there is the following material balance relationship:

Vy _n-1 +L'x' ₀ =Vy _n +L'x _n (5)

In the formula, V-organic phase flow rate;

L-material liquid flow rate;

L'- washing liquid flow rate;

y _i - the concentration of the extracted component in the organic phase effluent from the i-stage mixing-settler;

x _i - the concentration of the extracted component in the organic phase effluent from the i-stage mixing and clarifying device;

y _j - the concentration of the extracted component in the organic phase effluent from the j-stage mixing-settler;

x _j - the concentration of the extracted component in the organic phase effluent from the j-stage mixing-settler;

_y0 - the concentration of the extracted component in the fresh organic phase;

x ₀ - the concentration of the extracted component in the feed liquid;

x' ₀ - the concentration of the extracted component in the washing solution.

The above m+n material balance relations form a system of equations with 2(m+n) unknowns. In order to solve the above system of equations, it is also necessary to introduce the extraction balance relation equation

y _i =f(x _i ) (6)

In order to establish the unknown functional relationship f(·) between the concentration of metal ions in the organic phase and the concentration of metal ions in the aqueous phase in an ideal equilibrium, an offline extraction equilibrium experiment is required to obtain sample data for identification. First, take out a small amount of fresh organic phase for the extraction equilibrium experiment, prepare the aqueous phases with different metal ion concentrations respectively, and use the separatory funnel for the extraction equilibrium experiment, add the organic phase and the aqueous phase to the separatory funnel according to a certain ratio, and oscillate to mix Stand and stratify for a certain period of time, adjust the pH value of the balanced water phase, analyze the concentration of metal ions in the water phase, and record the experimental data for identification. Through analysis, the model can be embodied as a semi-empirical model

${\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

In the formula, a ₁ , a ₂ - parameters to be identified.

From the physical meaning of formula (7), it can be judged that the values of a ₁ and a ₂ should be greater than 0. Because the results of optimization calculations are often not unique, in order to make the calculation results conform to the actual situation as much as possible, the nonlinear optimization algorithm with constraints is used below, and the values of a ₁ and a ₂ can be limited within a certain range based on experience. Equation (7) for reverse optimization calculation. The specific steps are:

(A) Randomly generate a group of initial values greater than 0 and less than the specified range of a ₁ and a ₂ ;

(B) Substitute a ₁ and a ₂ into formula (6) to calculate the estimated value of y _i

(C) Sum of the sample value and the estimated error square

$\mathrm{<span\; class="notranslate">\; EMS</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\stackrel{\stackrel{<span\; class="notranslate">\; ^</span>}{<span\; class="notranslate">\; \&OverBar;</span>}}{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

As the objective function of the optimization calculation, judge whether the objective function meets the requirements, if the requirements are met, stop the calculation, otherwise enter the next step;

In the formula, y _i - is the observed value of the i-th sample;

-为第i个样本的估计值；

- is the estimated value of the i-th sample;

n- is the number of samples.

(D) Use the BFGS method to perform iterative calculation with constraints to obtain new a ₁ and a ₂ ;

(E) Repeat steps (B) to (D) until the objective function meets the requirements.

III. Data Model

According to the input and output data of the system, the method of establishing a mathematical model equivalent to the external characteristics of the system is called data modeling. Data modeling regards the system as a black box. Without knowing the internal structure and mechanism of the system, a group of secondary variables that are closely related to the leading variable and easy to measure are selected. According to an optimal criterion, the secondary variable is constructed using statistical methods. The mathematical model between the secondary variable and the leading variable.

In the present invention, non-linear PLS (RBF-PLS) is used as the data modeling method to compensate the unmodeled dynamics in the mechanism model. The RBF-PLS method is formed by combining the RBF network and the PLS algorithm, and its model structure is shown in FIG. 5 . The RBF-PLS method usually uses the Gaussian radial basis function to transform the independent variable data matrix X into the activation matrix A. The elements of A can be defined using the following formula:

${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\frac{{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

In the formula, k-the number of data samples;

x _i - the input vector for the ith data sample;

a _ij - the element in row i, column j of A;

c _j - the central parameter of the Gaussian function;

σ _j - the width parameter of the Gaussian function.

In the RBF-PLS method, the center parameter c _j is selected as the input vector for each data sample, namely

c _j =x _j (10)

The width parameter σ _j can be calculated by the following formula:

${\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; e</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

In the formula, e- is greater than the constant of 0, gets 1 among the present invention,

Therefore, matrix A is a k×k-dimensional square matrix with diagonal elements of 1.

After the above transformation, use the PLS algorithm to establish a linear regression model between the matrix A and the output data vector y. If T is a k×h dimensional matrix composed of the first h score vectors, the model can be described by the following formula:

A=TP ^T +E (12)

y=Tq+r=Ab+r (13)

where A-activation matrix;

T-score matrix;

P-loading matrix;

E-residual matrix;

y - output data vector;

q - loading vector;

r - residual vector;

Vector of regression coefficients for b-PLS.

The modeling steps using the above algorithm are as follows:

(A) standardize the training samples;

(B) Calculate the Euclidean distance between each training sample and the center, and use (11) formula to calculate the width parameter σ _j ;

(C) Utilize formula (9) to calculate and obtain activation matrix A;

(D) Establish a linear regression model between the activation matrix A and the output data vector y, and use the PLS algorithm to obtain the regression coefficient b.

(4) Correction of the mixed model

Because some process data are not reliable enough, the model established by the above method alone is not enough to provide reliable prediction accuracy. Therefore, on the basis of the above model, we also use the prediction error to further correct the mixed model. The recursive algorithm in The correction amount d(t) can be determined by the following formula

d(t)=wd ₀ (t)+(1-w)d(t-1) (14)

In the formula, d(t)-the weighted sum of the previous forecast error and historical error;

d ₀ (t) - current forecast error;

w-weighting coefficient, 0.5 is taken in the patent of the present invention;

And d(0)=0, d ₀ (t) can be calculated by the following formula

d ₀ (t)=y _lab (t-1)-y _mod (t-1) (15)

In the formula, y _lab -offline assay value;

y _mod - model predicted value;

The final model-corrected output can be calculated using the following formula

y _cor (t) = y _mod (t) + d (t) (16)

where _ycor - corrected model prediction.

(5) Determination of optimal operation guidance

The present invention adopts an expert system based on a mixed model to provide online optimization operation guidance for the extraction process, obtains a series of operating experience through on-site communication with the operator, and combines the mixed model to provide guidance for each operating variable through iterative calculations. The algorithm steps are as follows:

(A) setting the concentration requirements of each extracted component in the raffinate;

(B) read the concentration of each extracted component in the feed liquid and the flow rate of the feed liquid;

(C) According to the experience of the operator, select an appropriate organic phase flow rate and the corresponding washing liquid flow rate as the initial value;

(D) Under the above operating conditions, use the mixed model to predict the concentration of each extracted component in the raffinate, and judge whether it meets the standard; if one fails to meet the standard, return to step (C) to re-select an initial organic phase flow rate; if all reach the standard Go to step (E);

(E) Reduce the flow rate of the organic phase by ΔV under the condition of the above initial value, adjust the flow rate of the washing solution accordingly, call the mixing model again, predict the concentration of each extracted component in the raffinate, and judge whether it meets the standard; If one fails to meet the standard, then stop the calculation, and the organic phase flow rate and washing liquid flow rate obtained last time are the optimal operation guidance at the current moment; if both reach the standard, repeat step (E).

The invention can greatly improve the extraction efficiency, maintain the production in the best operating state, effectively reduce the consumption of auxiliary materials and energy, and prolong the operation period of the equipment.

Description of drawings

Fig. 1 is a process flow diagram of hydrometallurgical extraction process, wherein, PHT1-raffinate pH value sensor, PH2-washing raffinate pH value sensor, PH3-reverse raffinate pH value sensor, FT1-feed liquid flow sensor, FT2 - washing liquid flow sensor, FT3- stripping liquid flow sensor, FT4- anti-iron liquid flow sensor, FT5- organic phase flow sensor, TT1- feed liquid temperature sensor;

Fig. 2 is a schematic diagram of the hardware structure of the device of the present invention,

Fig. 3 is a hybrid model structure diagram;

Fig. 4 is the equivalent structure diagram of mechanism model;

Fig. 5 is a structural diagram of the nonlinear PLS model;

Fig. 6 is the curve trend chart of copper extraction raffinate Cu ion concentration test value and predicted value;

Fig. 7 is the interface diagram of copper extraction component concentration prediction and optimization operation guidance;

Fig. 8 is P204 pre-extraction raffinate Cu ion concentration test value and predicted value curve trend figure;

Fig. 9 is P204 pre-extraction raffinate Mn ion concentration test value and predicted value curve trend figure;

Figure 10 is a P204 pre-extraction detection interface diagram;

Figure 11 is the interface diagram of component content prediction in the cobalt hydrometallurgical extraction workshop;

Fig. 12 is an interface diagram of the optimization operation guidance interface of the cobalt hydrometallurgical extraction workshop.

Detailed ways

The following specific examples have been practically applied in the extraction workshop of a cobalt hydrometallurgical production plant, and have achieved remarkable results.

Example 1

Implementation on the copper extraction and impurity removal production line.

There are 5 stages of cascaded extraction tanks for copper extraction on this production line, 2 stages for extraction, 1 stage for washing, 2 stages for stripping, a total of 5 stirring motors, 4 flow meters, 4 pumps, and 4 frequency conversion device, a pH meter, and a thermometer. The extraction process detection system is mainly composed of flow detection, pH value detection and temperature detection.

PLC controller adopts CPU 315-2DP of Siemens 300 series, with Profibus-DP port to connect distributed IO. Equip the PLC with an Ethernet communication module for the host computer to access PLC data. The PLC controller and Ethernet communication module are placed in the PLC cabinet in the central control room.

The pH value of the extraction process is detected online through the glass electrode produced by Cole-parmer Company, and the change of the pH value of the solution is converted into the change of the mV signal. The glass electrode pH measurement system blows the end of a glass tube with a pH-sensitive glass membrane into a bubble shape, and the tube is filled with a 3mol/l KCL buffer solution containing saturated AgCl, with a pH value of 7. The potential difference reflecting the pH value existing on the two sides of the glass film uses the Ag/AgCl conduction system to derive the potential difference, and then uses the mA acquisition instrument to convert the mA value into a pH value and display it.

The invention adopts the pH200 acidity controller produced by Cole-parmer Company to display the pH value on the spot and transmit the detection signal. The pH200 acidity controller uses the pH electrode to generate different DC potentials for the concentration of hydrogen ions in the measured solution, which are input to the A/D converter through the preamplifier to achieve the purpose of pH measurement, and then the pH value is displayed digitally, and the The pH value is converted into a current signal output.

The temperature of the feed liquid in the extraction process is detected by the platinum resistance thermometer produced by SOLUTION. The platinum resistance thermometer uses the characteristics of electrical parameters changing with temperature to detect the temperature.

Since the organic phase is not conductive, different flow sensors are used for flow detection:

(A) Water phase flow detection: material liquid, acid liquid, and alkali liquid are all conductive and corrosive, and the electromagnetic flowmeter with polytetrafluoroethylene lining produced by KROHNE company is selected for detection. The electromagnetic flowmeter has the advantages of high precision, long service life and convenient maintenance for the detection of non-resistance parts. The local display instrument equipped with the electromagnetic flowmeter can realize the functions of flowmeter local display, flow signal transmission and flow accumulation. The signal output by the electromagnetic flowmeter is a standard current signal.

(B) Organic phase flow detection: The organic phase in the extraction section is non-conductive and cannot use an electromagnetic flowmeter. The flow detection uses a differential pressure flowmeter produced by ELETTA. The principle is that the differential pressure in the pipeline acts on the rubber diaphragm to generate mechanical movement of the diaphragm rod, which acts on the linear potentiometer installed on the circuit board of the monitor, because the monitor has a linear conversion between differential pressure and flow function, so the board can provide a linear flow output signal. The output signal of the differential pressure flowmeter is a standard current signal.

The host computer is a Core 2 DELL computer with WINDOW XP operating system.

The component content prediction and optimization operating system runs on a Core 2 DELL computer, using C#2005 programming software, and the mixed model algorithm uses Matlab 2007a programming software.

The signal transmission software of PLC and component content prediction and optimization operating system adopts C#2005 programming software.

The detection instrument is installed on site during the extraction process, and the detection instrument transmits the collected signal to the PLC through Profibus-DP, and the PLC regularly transmits the collected signal to the host computer through Ethernet, and the host computer transmits the received data to the component content prediction and optimization The operating system performs online real-time prediction of component concentration and provides online optimization operation guidance.

The first step, parameter identification of the mechanism model: determine the extraction balance relationship in the mechanism model according to the extraction balance experiment;

(A) Configure the water phase with different concentrations of the extracted components, and record the volume of the water phase;

(B) Get a certain volume of fresh organic phase, mix the water phase and the organic phase in a separatory funnel according to a certain ratio, stir evenly, and let stand to separate layers;

(C) take out the water phase, and test the concentration of the extracted component in the water phase;

(D) Calculate the concentration of the extracted component in the organic phase by subtraction, repeat the above steps, and record the experimental data;

(E) BFGS algorithm is used to identify unknown parameters in the extraction balance relationship.

The second step, collecting data: collecting the sensor measurement data corresponding to the offline test data, and recording the offline test data values y _lab (1), ..., y _lab (n);

The third step, mechanism model prediction: use the mechanism model to predict the content of the extracted components of the raffinate, and record the prediction results y _mod (1), ..., y _mod (n) corresponding to the offline test data;

The fourth step, compare the predicted result with the offline test value, and calculate the error value d(1),...,d(n) between the predicted value and the real value, where

d(i) = y _lab (i) - y _mod (i), i = 1, ..., n (17)

The fifth step, data model establishment: the above-mentioned error value and its corresponding standardized detection value are combined to form an input-output data pair, and the RBF-PLS method is used for training to obtain the parameters in the data model;

The sixth step is the prediction of the mixed model: the structure of the established mechanism model and the data model are combined into a parallel mixed model, and the mixed model is used to predict the content of the raffinate components in real time;

The seventh step, the correction of the mixed model: use the correction algorithm to correct the predicted value of the mixed model according to the daily offline test value, and output the corrected predicted value;

The eighth step, optimization operation guidance: use the optimization algorithm to provide online optimization operation guidance according to the value of the operation variable.

Table 1 and Figure 6 respectively show the curve trend of the off-line assay value and model prediction value of Cu concentration in the raffinate.

Analyzing the measured data and the predicted value of the mixed model gives a predicted root mean square error of 0.0694, which is within the range of the predicted value of the present invention and fully meets the needs of actual production.

serial number Feed liquid concentration Feed liquid flow lotion flow Organic phase flow Raffinate pH value temperature Raffinate concentration Model predicted concentration 1 14.89 1.25 0.20 3.00 0.51 29.2 0.38 0.36 2 11.31 1.40 0.20 2.90 0.49 28.6 0.50 0.49 3 13.16 1.20 0.18 4.00 0.50 9.2 0.15 0.17 ... ... 28 17.37 1.20 0.20 2.00 0.54 29.5 1.74 1.82 29 18.13 1.20 0.19 3.90 0.53 29.3 2.25 2.18 30 16.79 1.80 0.20 4.00 0.55 27.5 1.79 1.62

Table 1 Comparison of test value and predicted value of Cu ion concentration in copper extraction raffinate

In addition, based on the previously established mixing model, and using the optimization algorithm designed by us, the extraction process was optimized online and compared with the on-site operation according to the optimized operating conditions. The cumulative data comparison results for one month are shown in Table 3:

Organic phase flow rate (cumulative) Washing liquid flow (cumulative) Stripping liquid flow (cumulative) historical value 2592.04 116.64 1296.27

Organic phase flow rate (cumulative) Washing liquid flow (cumulative) Stripping liquid flow (cumulative) Optimized value 2462.40 103.68 1166.37 Savings 129.64 12.96 129.90 percentage 5.00% 11.11% 10.02%

Table 2 Comparison table of optimization results

The present invention combines the copper extraction component content prediction interface with the optimization operation guidance interface, which is coordinated, as shown in Figure 7, which is a unified interface for the extraction process component content prediction and expert system optimization operation guidance.

Example 2

Implementation on the P204 pre-extraction and impurity removal production line.

There are 20 cascaded extraction tanks for P204 pre-extraction on this production line, 10 for extraction, 3 for washing, 3 for stripping, 3 for anti-iron, a total of 20 stirring motors, 5 flow meters , 5 pumps, 5 inverters, 3 pH meters, and 1 thermometer. The extraction process detection system is mainly composed of flow detection, pH value detection and temperature detection.

serial number Feed liquid concentration Feed liquid flow lotion flow Organic phase flow Raffinate pH value temperature Raffinate concentration Model predicted concentration 1 4.60 3.0 0.22 2.4 4.64 15.1 0.20 0.21 2 6.35 2.0 0.25 3.0 4.45 12.4 0.22 0.21 3 4.94 2.0 0.24 2.6 4.50 9.2 0.21 0.20 ... ... 28 5.71 2.5 0.20 2.8 4.37 9.5 0.22 0.23 29 2.92 4.0 0.15 1.6 4.51 10.1 0.22 0.19 30 3.17 3.5 0.26 3 4.60 9.6 0.15 0.16

Table 3 Raffinate Cu ion concentration test value and predicted value comparison

Tables 3 and 4 and Figures 8 and 9 respectively show the curve trends of the off-line assay values and model prediction values of the concentrations of Cu and Mn in the raffinate.

The root mean square errors of the analysis measurement data and the prediction value of the mixed model are respectively 0.0231 and 0.0059, both of which are within the range of the prediction value of the present invention, fully meeting the actual production needs.

serial number Feed liquid concentration Feed liquid flow lotion flow Organic phase flow Raffinate pH value temperature Raffinate concentration Model predicted concentration 1 0.39 3.0 0.22 2.4 4.64 15.1 0.001 0.003 2 0.70 2.0 0.25 3.0 4.45 12.4 0.020 0.021 3 0.51 2.0 0.24 2.6 4.50 9.2 0.021 0.023 ... ... 28 1.27 2.5 0.20 2.8 4.37 9.5 0.02 0.018 29 0.73 4.0 0.15 1.6 4.51 10.1 0.022 0.019 30 1.83 3.5 0.26 3 4.60 9.6 0.046 0.041

Table 4 Raffinate Mn ion concentration test value and predicted value comparison

In addition, based on the previously established mixing model and the optimization algorithm designed by us, the extraction process was optimized online and compared with the on-site operation according to the optimized operating conditions. The results of the accumulated one-month data comparison are shown in Table 3:

Organic phase flow rate (cumulative) Washing liquid flow (cumulative) Strip flow (cumulative) Anti-ferrous liquid flow (cumulative) historical value 1259.64 64.82 1310.40 1346.58 Optimized value 1135.37 57.60 1195.36 1188.49 Savings 124.27 7.22 115.04 158.09 percentage 9.87% 11.13% 8.78% 11.74%

Table 5 Comparison table of optimization results

The present invention combines the P204 pre-extraction component content prediction interface with the optimization operation guidance interface, which is coordinated, as shown in Figure 10, which is a unified interface for the extraction process component content prediction and expert system optimization operation guidance.

When the present invention predicts the component content of a certain hydrometallurgical extraction workshop and provides optimized operation guidance, a friendly human-computer interaction interface is also essential. The present invention also fully takes this requirement into consideration, and lists the overall prediction results and the optimization operation guidance at the same time, as shown in Figures 11 and 12, which is more convenient for the operator to monitor the entire extraction section.

Claims (2)

1. wet-process metallurgic extraction process component content prediction and Optimizing operation method, be the hydrometallurgical extraction technology that adopts known employing multitple extraction groove, it is characterized in that: by hybrid modeling wet-process metallurgic extraction process, realize the real-time estimate of raffinate component concentration, and provide online Optimizing operation to instruct to extraction process, what comprise that the correction of foundation, mixture model of the selection of data acquisition, auxiliary variable and standardization, mixture model and Optimizing operation instruct step such as determines;

1) process data collection

The hardware unit that the present invention uses comprises extraction process component content prediction and Optimizing operation system, host computer, PLC, on-the-spot sensing becomes send part, wherein on-the-spot sensing becomes send part to comprise the pH value, measuring instrument such as temperature and flow, in the on-the-spot installation and measuring instrument of extraction process, measuring instrument is delivered to PLC with the signal of gathering by the Profibus-DP bus, PLC realizes the A/D conversion with the signal of gathering, and regularly acquired signal is sent to host computer by Ethernet, host computer passes to extraction process component content prediction and Optimizing operation system to the local plc data of accepting, carry out the real-time estimate of raffinate component concentration, and provide online Optimizing operation to instruct;

2) selection of auxiliary variable and standardization

The selected auxiliary variable of the present invention comprises:

1. in the feed liquid by collection component concentrations x ₁

2. the flow X of feed liquid ₂

3. the flow X of washing lotion ₃

4. the flow X of organic phase ₄

5. the pH value X of raffinate ₅

6. the temperature X of feed liquid ₆

At first the sensor measurement data that collect are carried out standardization:

${\tilde{x}}_j\left(i\right)=\frac{x_j\left(i\right)-x_j\left(\min \right)}{x_j\left(\max \right)-x_j\left(\min \right)}$

In the formula -Di i data sample, the standardized value of j measurement value sensor;

x _j(i)-and an i data sample, j measurement value sensor;

The minimum value an of-Di j measurement value sensor;

The maximal value an of-Di j measurement value sensor;

3) set up mixture model

The present invention adopts the structure of parallel mixture model, adopts the parallel mixture model structure based on non-linear PLS that the not modeling in the mechanism model is dynamically compensated;

The mechanism model that adopts among the present invention mixes limpid device by the m+n level and forms, and wherein the m level is used for extraction, the n level is used for washing; Usually,, can be considered 0, therefore, only consider extraction section and washing section are carried out modeling for simplifying the mechanism model structure through the concentration of each metallic ion in the fresh organic phase behind the stripping section because back extraction, anti-iron efficient are higher,

According to the mass balance relation, for extraction section i level, i=1,2 ... there is following mass balance relation in m-1:

Vy _i-1+(L+L′)x _i+1＝Vy _i+(L+L′)x _i

And be charging level i=m for extraction section m level, there is following mass balance relation:

Vy _m-1+Lx ₀+L′x _m+1＝Vy _m+(L+L′)x _m

For washing section j level, j=m+1 ..., there is following mass balance relation in n-1:

Vy _j-1+L′x _j+1＝Vy _j+L′x _j

And for washing section n level, i.e. there is following mass balance relation in j=n:

Vy _n-1+L′x′ ₀＝Vy _n+L′x _n

V-organic phase flow in the formula;

L-feed liquid flow;

L '-eluent flow;

y _i-flow out the organic phase by the collection component concentrations from i level mixer-settler extractor;

x _i-flow out the organic phase by the collection component concentrations from i level mixer-settler extractor;

y _j-flow out the organic phase by the collection component concentrations from j level mixer-settler extractor;

x _j-flow out the organic phase by the collection component concentrations from j level mixer-settler extractor;

y ₀Quilt collection component concentrations in the-fresh organic phase;

x ₀Quilt collection component concentrations in the-feed liquid;

X ' ₀Quilt collection component concentrations in the-washing lotion;

Above-mentioned m+n mass balance relation formed a system of equations with the individual unknown number of 2 (m+n), in order to find the solution above-mentioned system of equations, also needs to introduce the extraction equilibrium relation equation:

y _i＝f(x _i)

The concentration of metallic ion and the unknown function between the aqueous metal ion concentration concern f () in the organic phase when setting up balance ideally, need carry out the experiment of off-line extraction equilibrium to obtain the sample data that is used for identification; At first take out a small amount of fresh organic phase and be used for the extraction equilibrium experiment, prepare water respectively with different metal ion concentration, carry out the extraction equilibrium experiment with separating funnel, organic phase and water are by certain adding separating funnel of comparing, vibration mixes, standing demix, aqueous pH values behind the adjustment is analyzed the aqueous metal ion concentration, and record is used for the experimental data of identification; By analyzing, model can be embodied as semiempirical model:

${\stackrel{\&OverBar;}{y}}_i=f\left({\stackrel{\&OverBar;}{x}}_i\right)=\frac{a_1\&CenterDot;{\stackrel{\&OverBar;}{x}}_i}{a_2+{\stackrel{\&OverBar;}{x}}_i}$

A in the formula ₁, a ₂-parameter to be identified;

From the physical significance of above-mentioned radius empirical model, can judge a ₁, a ₂Value should be greater than 0; Because the result of computation optimization often is not only, tally with the actual situation in order to make result of calculation as far as possible, adopt the nonlinear optimization algorithm of belt restraining below, can be rule of thumb with a ₁, a ₂Value be limited within certain scope, to above-mentioned radius empirical model ${\stackrel{\&OverBar;}{y}}_i=f\left({\stackrel{\&OverBar;}{x}}_i\right)=\frac{a_1\&CenterDot;{\stackrel{\&OverBar;}{x}}_i}{a_2+{\stackrel{\&OverBar;}{x}}_i}$ Carry out reverse computation optimization; Concrete steps are:

1. produce one group at random greater than 0, less than a ₁, a ₂The initial value of specialized range;

2. with a ₁, a ₂The substitution following formula calculates y _iEstimated value

3. with sample value and estimated error sum of squares

$\mathrm{EMS}=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{({\stackrel{\&OverBar;}{y}}_i-{\stackrel{\widehat{\&OverBar;}}{y}}_i)}^2$

As the objective function of computation optimization, judge whether objective function meets the requirements, as meet the demands, then stop to calculate, otherwise enter next step;

Y in the formula _i-be the observed reading of i sample;

-be the estimated value of i sample;

N-is the number of sample;

4. carry out the iterative computation of belt restraining condition with the BFGS method, obtain new a ₁, a ₂

5. repeat 2.～4. step, meet the demands up to objective function;

Adopt non-linear PLS (RBF-PLS) dynamic as the not modeling in the data modeling method compensatory michanism model among the present invention, the RBF-PLS method is by RBF network and PLS algorithm be combined into,

What Optimizing operation instructed determines

The present invention adopts the system based on mixture model to instruct for extraction process provides online Optimizing operation, by exchanging of on-the-spot and operator, obtain the sequence of operations experience, and in conjunction with mixture model, provide the guidance of each performance variable by interative computation, its specific algorithm step is as follows:

1. set each quilt collection component concentrations requirement in the raffinate;

2. read the flow of each come together component concentrations and feed liquid in the feed liquid;

3. according to operator's experience, select a suitable organic phase flow, and eluent flow correspondingly is as initial value;

4. under the aforesaid operations condition, utilize each quilt collection component concentrations in the mixture model prediction raffinate, judge whether up to standard; If do not have one up to standardly, return that 3. the step is reselected an initialization organic phase flow; If all up to standard entering 5. goes on foot;

5. the organic phase flow is reduced Δ V under the condition of above-mentioned initial value, eluent flow adjusts accordingly simultaneously, calls mixture model again, to each is predicted by the collection component concentrations in the raffinate, judges whether up to standard; If do not have one up to standardly, then stop to calculate, have valency phase flow rate and the eluent flow of last time trying to achieve are the Optimizing operation guidance of current time; If all up to standard, then 5. the step repeats.

2. wet-process metallurgic extraction process component content prediction and Optimizing operation method based on mixture model according to claim 1 are to implement on copper collection removal of impurities production line, it is characterized in that:

Copper extraction has 5 grades of cascade extraction grooves on the production line, and 2 grades are used for extraction, and 1 grade is used for washing, and 2 grades are used for back extraction, totally 5 stirring motors, and 4 flowmeters, 4 pumps, 4 frequency converters, 1 pH meter, 1 thermometer is formed; The extraction process detection system mainly is made of flow detector, pH value detector, temperature monitor;

The PLC controller adopts the CPU 315-2DP of Simens 300 series, has the Profibus-DP mouth and connects distributed I/O; For PLC is equipped with the ethernet communication module, be used for host computer visit plc data; PLC controller and ethernet communication module are placed in the PLC cabinet in the central control room;

The pH value of extraction process is to carry out the online detection of pH value by the glass electrode that Cole-parmer company produces, and the variation of pH value of solution value is converted into the variation of mV signal; Glass electrode PH measuring system is blown out blister with a glass tube end for the glass-film of pH sensitivity, and casing pack has the 3mol/l KCL buffer solution that contains saturated AgCl, and the pH value is 7; The potential difference (PD) Ag/AgCl conducting system that is present in the reflection pH value of two of glass-films is derived potential difference (PD), with the mA acquisition instrument mA number is converted into the pH value then and shows;

The pH200 type acidity controller that the present invention adopts Cole-parmer company to produce carries out the change of showing on the spot of pH value and detection signal and send pH200 type acidity controller to utilize the different DC potential of pH generation in the pH electrode pair detected solution, be input to A/D converter by prime amplifier, to reach the purpose that pH measures, show the pH value by numeral then, simultaneously the pH value is converted to current signal output;

The temperature of extraction process feed liquid is to detect by the platinum-resistance thermometer that SOLUTION company produces, and platinum-resistance thermometer utilizes the temperature variant characteristic of electric parameter to come detected temperatures;

Because organic phase is non-conductive, therefore adopt different flow sensors for flow detection:

1. water flow detection: feed liquid, acid solution, alkali lye all conduct electricity and have corrosivity, and the teflon-lined electromagnetic flowmeter that has of selecting for use KROHNE company to produce detects; Electromagnetic Flow is counted non-resistance spare and is detected; The Displaying Meter on the spot that electromagnetic flowmeter is equipped with can realize that flowmeter shows on the spot, the flow signal change is sent and the flux cumulating function; The signal of electromagnetic flowmeter output is the current signal of standard;

2. the organic phase flow detects: the extraction section organic phase is non-conductive can not to adopt electromagnetic flowmeter, the differential pressure flowmeter that the detection of flow selects for use ELETTA company to produce, and the signal of the output of differential pressure flowmeter is the current signal of standard;

Host computer is selected Core 2 DELL computing machines for use, adopts WINDOW XP operating system;

Component content prediction and Optimizing operation system operate on the Core 2 DELL computing machines, adopt the C#2005 programming software, and the mixture model algorithm adopts Matlab 2007a programming software;

It is to adopt the C#2005 programming software that the signal of PLC and component content prediction and Optimizing operation system transmits software;

In the on-the-spot installation and measuring instrument of extraction process, measuring instrument is sent to the signal of gathering among the PLC by Profibus-DP, PLC regularly sends acquired signal to host computer by Ethernet, the data of accepting are passed to component content prediction to host computer and the Optimizing operation system carries out the prediction of concentration of component online in real time, and provide online Optimizing operation to instruct;

The first step, mechanism model parameter identification: determine that according to the extraction equilibrium experiment extraction equilibrium in the mechanism mould concerns:

1. configuration has variable concentrations the come together water of component and the volume of record water;

2. get the fresh organic phase of certain volume, water is mixed by certain comparing in separating funnel with organic phase, stir standing demix;

3. take out water, the chemical examination aqueous phase is by the component concentrations of coming together;

4. calculate quilt collection component concentrations in the organic phase with minusing, repeat above-mentioned steps, and the record experimental data;

5. utilize the unknown parameter in the BFGS algorithm identification extraction equilibrium relation;

Second step, collection data: collect and off-line analysis data corresponding sensor measurement data, and record off-line analysis data value y _1ab(1) ..., y _1ab(n);

The 3rd step, mechanism model prediction: utilize mechanism model that the component concentration that come together of raffinate is predicted, and record the predict the outcome y corresponding with the off-line analysis data _Mod(1) ..., y _Mod(n);

The 4th step, will predict the outcome and the off-line laboratory values compares, and calculate error amount d (1) between predicted value and the actual value ..., d (n), wherein

d(i)＝y _1ab(i)-y _mod(i)，i＝1，...，n

The 5th step, data model are set up: the detected value after the standardization that above-mentioned error amount is corresponding with it, and it is right to form inputoutput data, utilizes the RBF-PLS method to train, and obtains the parameter in the data model;

The prediction of the 6th step, mixture model: the structure of foundation mechanism model and data model is formed parallel mixture model, utilize mixture model that the raffinate component concentration is carried out real-time estimate;

The correction of the 7th step, mixture model: utilize correcting algorithm that the predicted value of mixture model is carried out on-line correction according to the off-line laboratory values of every day, and the predicted value behind the output calibration;

The 8th step, Optimizing operation instruct: utilize optimized Algorithm, according to the performance variable value, provide online Optimizing operation to instruct;

System interface comprises: A reads the main interface that respectively extracts workshop section's off-line analysis data and show predicted data in real time; B provides online Optimizing operation to instruct the interface; C reads the interface of important parameter in the model; The technological process interface that D shows predicted data in real time and provides online Optimizing operation to instruct.

Images