CN101398314A - Differential pressure transmitter data fusion device and method based on DSP - Google Patents

Abstract

The invention provides a data melting device and a method used for measured quantity and output value of a differential pressure transmitter on the basis of DSP; the device comprises a micro-processor DSP, a E<2>PROM, a signal conditioning module, an AD converter and a power module; the device is characterized in that the output of the differential pressure transmitter is connected to the micro-processor DSP by the signal conditioning module and the AD converter; the micro-processor DSP is connected with the E<2>PROM; the output of the micro-processor DSP is connected with a serial port; the power module provides working power for all devices; the method adopts the device and completes the disposal of arithmetic by the DSP, improves the measurement precision and intelligentizes the differential pressure transmitter. The device has high integration, small volume and good real-time performance and disposal capability.

Description

Data fusion device and method based on the differential pressure transmitter of DSP

Technical field

The present invention relates to the data fusion device and the method for measured and output valve in the intelligent field, particularly differential pressure transmitter of differential pressure transmitter.

Background technology

Differential pressure transmitter is the isoparametric field instrument of flow, differential pressure that is used for detecting production run fluid in service.From low, the heavy big displacement mercury float-type differential pressure gauge of early stage precision, to big, the poor reliability of volume the 1950's, force balance type differential pressure transmitter that precision is low.To the seventies in 20th century, along with the development of technology, occurred the 3rd generation micrometric displacement electronic type transmitter.Passed through the development course of decades, differential pressure transmitter of new generation is simple in structure, volume is little, precision is high, good reliability.Since the nineties in 20th century, because electronic technology, the high speed development of computer technology and the appearance of micro-electronic mechanical skill (MicroElectro Mechanical Technology is called for short MEMT) make sensor do littler and littler, function is also more and more stronger.

Along with further developing of control system, that the development of control system more and more trends towards is complicated, high precision int and intellectuality, and the system requirements sensor can be formed high precision, powerful measurement and control network.Sensor is to realize measuring and the primary link of controlling, it is the sensing unit that non electrical quantity is become electric weight, making the intellectuality of low layer instrument, making its measuring accuracy higher is the problem that urgently will solve, the sensor of now also has very big gap aspect intelligent, if can be under the constant situation of sensitive element, promote the measuring accuracy of sensor, then can improve the accuracy of control system information needed to a great extent.By means of the fast development of microprocessor performance, make these imaginations to come true.

Differential pressure transmitter directly contacts with measured medium, moves in the rugged surroundings of being everlasting, and therefore the degree of accuracy to transmitter has very high requirement.And the main cause that influences the differential pressure transmitter measuring accuracy is: the output-input characteristics of sensor has nonlinear characteristic and drifts about along with the variation of time; Differential pressure signal is subjected to the variation of environment temperature, static pressure etc. easily and produces drift.The most traditional linearizing method of nonlinear transducer characteristic is the hardware compensating method, and this method is difficult to accomplish full compensation, and the drift of compensation hardware can influence the complete machine precision.

Modeling method to the sensor input-output characteristic has the earliest: least-squares line method, separate straight lines method etc., but can not satisfy high-precision requirement, and too complicated again with the high order curve match, can not satisfy request for utilization.Development along with neural network, methods such as BP neural network, Elman neural network all have been used in the modeling of sensor input-output characteristic, but they have shortcomings such as being absorbed in local minimum, generalization ability difference easily, along with improving constantly of accuracy requirement, need seek better modeling method.

Summary of the invention

The object of the present invention is to provide a kind of differential pressure transmitter data fusion device and method based on DSP, for differential pressure transmitter provides a kind of embeddable data fusion device based on support vector machine, carry out the processing and the realization of algorithm by DSP, improve the precision of measuring and make the differential pressure transmitter intellectuality.

For achieving the above object, the present invention adopts following technical proposals:

A kind of differential pressure transmitter data fusion device based on DSP comprises microprocessor DSP, E ²PROM, signal condition module, AD converter, power module is characterized in that the output of differential pressure transmitter is connected to microprocessor DSP through signal condition module and AD converter, and microprocessor DSP connects E ²PROM, the output of microprocessor DSP is connected to a serial ports, provides working power by power module for each device.

A kind of differential pressure transmitter data fusion method based on DSP adopts said apparatus to carry out data fusion, it is characterized in that the step of moving is as follows:

1) initialization of system: E ²The initialization of PROM and microprocessor DSP;

2) reception of signal: the data of differential pressure transmitter output are input among the microprocessor DSP through signal condition module, AD converter;

3) microprocessor DSP sets up and the setting initialization model according to support vector machine is theoretical, comprise the parameter of initialization model and the kernel function of initialization model, according to the data that received initialization model is trained, obtain the support vector machine and the parameter of training pattern, eliminate the interference of temperature, pressure etc., improved the precision of differential pressure value.The step of whole process is as follows:

1. set up and initialization model: comprise the parameter of initialization model, selected kernel function and parameter thereof;

2. given input value of normalization and desired destination value.Formula is as follows:

$\stackrel{\&OverBar;}{u_i}=2\&times;\frac{u_i-u_{\min }}{u_{\max }-u_{\min }}-1$

In the formula:

Differential pressure pick-up output voltage after the-normalization; u _i-differential pressure pick-up output voltage; u _Min-differential pressure pick-up output voltage minimum value; u _Max-differential pressure pick-up output voltage maximal value;

3. the initialization model of usefulness foundation and sample data are according to formula $f^{*}\left(x\right)=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{i}k(x_i,x)+b$ Calculate differential pressure value; In the formula: β _i:-kernel function coefficient; B :-threshold value; K (x _i, x) :-kernel function; X :-input quantity; x _i:-support vector; f ^*(x) :-regression function; K :-support vector machine number;

4. obtain the deviation that calculates differential pressure value and expectation differential pressure value;

5. the parameter in Model parameter and the kernel function is learnt;

Return 2. double counting of step, till error meets the demands;

6. determine parameter, kernel function and the parameter thereof of training pattern;

4) microprocessor DSP receives new data, utilizes step 3) to set up training pattern and calculates differential pressure value;

5) the DSP microprocessor deposits the data of handling well in E ²Export among the PROM and through serial ports; And to step 4).

The present invention has following conspicuous outstanding substantive distinguishing features and remarkable advantage compared with prior art:

1. based on the modeling method of support vector machine theory, can under the situation of existing information, obtain optimum solution and need not experimental knowledge and the internal mechanism of understanding differential pressure transmitter in depth, only need to use the input and output data just can realize the data fusion of differential pressure transmitter; The fitting precision height, generalization ability is strong, does not have the study phenomenon.

2. based on the differential pressure transmitter support vector machine data fusion device of DSP, the software and hardware encapsulation is complete, the integrated level height, and volume is little, and real-time, processing power are good.

Description of drawings

Fig. 1 has provided the general study block diagram of neural network.

Fig. 2 has provided based on support vector machine gamma correction process.

Fig. 3 has provided the hardware block diagram of system.

Fig. 4 has provided the process flow diagram based on the data fusion method of support vector machine.

Fig. 5 has provided the workflow diagram of system.

Embodiment

A preferred embodiment of the present invention accompanying drawings is as follows:

Referring to Fig. 3, this differential pressure transmitter support vector machine data fusion device based on DSP comprises microprocessor DSP 6, E ²PROM 4, signal condition module 2, AD converter 3, power module 5, the output of described differential pressure transmitter 1 is connected to microprocessor DSP 6 through signal condition module 2 and AD converter 3, and microprocessor DSP6 connects E ²PROM 4 also outputs to a serial ports 7.Provide working power by power module 5 for each device.

Referring to Fig. 4 and Fig. 5, this adopts said apparatus to carry out data fusion based on the differential pressure transmitter support vector machine data fusion method of DSP, and the step of its operation is:

1) initialization of system: E ²The initialization of PROM 4 and microprocessor DSP 6;

2) reception of signal: the data of differential pressure transmitter 1 output are input among the microprocessor DSP 6 through signal condition module 2, AD converter 3;

3) microprocessor DSP 6 sets up and the setting initialization model according to support vector machine is theoretical, comprise the parameter of initialization model and the kernel function of initialization model, according to the data that received initialization model is trained, obtain the support vector machine and the parameter of training pattern, eliminate the interference of temperature, pressure etc., improved the precision of differential pressure value.The step of whole process is as follows:

1. set up and initialization model: comprise the parameter of initialization model, selected kernel function and parameter thereof;

2. given input value of normalization and desired destination value.Formula is as follows:

$\stackrel{\&OverBar;}{u_i}=2\&times;\frac{u_i-u_{\min }}{u_{\max }-u_{\min }}-1\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(1\right)$

In the formula:

Differential pressure pick-up output voltage after the normalization; u _i-differential pressure pick-up output voltage; u _Min-differential pressure pick-up output voltage minimum value; u _Max-differential pressure pick-up output voltage maximal value;

3. calculate differential pressure value with initialization model of setting up and sample data according to formula (9);

4. obtain the deviation that calculates differential pressure value and expectation differential pressure value;

5. the parameter in Model parameter and the kernel function is learnt;

Return 2. double counting of step, till error meets the demands;

6. determine parameter, kernel function and the parameter thereof of training pattern;

4) microprocessor DSP 6 receives new data, utilizes step 3) to set up training pattern and calculates differential pressure value;

5) DSP microprocessor 6 deposits the data of handling well in E ²Among the PROM 4 and through serial ports 7 outputs; And to step 4).

Be described further below:

Data fusion method based on support vector machine

Support vector machine is a kind of learning model, and Fig. 1 has provided the model of neural network learning generally speaking.The basic thought of support vector machine is with the method for nuclear mapping nonlinear problem to be converted into the feature space of higher-dimension, and the structure linear discriminant function is realized the Nonlinear Discriminant Function in the former space in high-dimensional feature space, as shown in Figure 2.

The differential pressure transmitter input/output relation can be expressed as:

q＝f(x)，x∈(a，b)……(2)

Wherein x is the input quantity of differential pressure transmitter, and q is an output quantity, and a and b are the scope of input signal.The input sample is: D={ (x ₁, q ₁), (x ₂, q ₂) ..., (x _k, q _k).

At first, the x territory with a nonlinear transformation x → φ (x), is mapped to the feature space of a higher-dimension with the input space, in the feature space of this higher-dimension, carries out linear regression then, that is:

f ^*(x)＝ω·φ(x)+b……(3)

The dot product of " " expression vector in the formula, ω is the support vector machine coefficient, b is a threshold value.According to the structural risk minimization criterion, under the situation of permissible error, optimization problem is converted into objective function:

$\min \frac{1}{2}{\left|\right|\mathrm{\&omega;}\left|\right|}^2+c\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}({\mathrm{\&xi;}}_i+{\mathrm{\&xi;}}_i^{*})\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(4\right)$

Wherein, c is a given constant, ξ _i ^*, ξ _iIt is slack variable.Adopt ε-insensitive loss function, minimized constraint condition is: f ^*(x _i)-y _i≤ ξ _i ^*+ ε, y _i-f ^*(x _i)≤ξ _i+ ε, ξ _i ^*, ξ _i〉=0.Introduce Lagrangian function:

In the formula, a _i, a _i ^*, r _i, r _i ^*〉=0, be all coefficient.Make ω in the formula (5), b, ξ _i, ξ _i ^*Partial derivative all equal zero, with local derviation as a result substitution formula (4) obtain the antithesis optimization problem.Its a _i, a _i ^*The maximization objective function is:

$\max \mathrm{\&omega;}(a_i,a_i^{*})=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{y}_i(a_i-a_i^{*})-\frac{1}{2}\underset{i,j=1}{\overset{k}{\mathrm{\&Sigma;}}}(a_i-a_i^{*})(a_j-a_j^{*})\left[\mathrm{\&phi;}\left(x_i\right)\mathrm{\&phi;}\left(x_j\right)\right]-\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}(a_i+a_i^{*})\mathrm{\&epsiv;}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(6\right)$

Its constraint condition is $\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}(a_i-a_i^{*})=0,$ a _i, a _i ^*φ (x among the ∈ [0, c], formula (5) _i) φ (x _j) be the dot-product operation of high-dimensional feature space, make kernel function K (x, x _i)=φ (x _i) φ (x _j), therefore, objective function becomes

$\mathrm{\&omega;}(a_i,a_i^{*})=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{y}_i(a_i-a_i^{*})-\frac{1}{2}\underset{i,j=1}{\overset{k}{\mathrm{\&Sigma;}}}(a_i-a_i^{*})(a_j-a_j^{*})K(x,x_i)-\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}(a_i+a_i^{*})\mathrm{\&epsiv;}\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(7\right)$

Separating formula earlier in constraint condition obtains

$\mathrm{\&omega;}=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}(a_i-a_i^{*})\mathrm{\&phi;}\left(x_i\right)\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(8\right)$

Make β=a _i-a _i ^*, β ≠ 0 is corresponding to support vector machine.Regression function becomes

$f^{*}\left(x\right)=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{i}k(x_i,x)+b\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\&CenterDot;\left(9\right)$

Claims (3)

1. the differential pressure transmitter data fusion device based on DSP comprises microprocessor DSP (6), E ²PROM (4), signal condition module (2), AD converter (3), power module (5), it is characterized in that: the output of differential pressure transmitter (1) is connected to microprocessor DSP (6) through signal condition module (2) and AD converter (3), and microprocessor DSP (6) connects E ²PROM (4), microprocessor DSP (6) output is connected to a serial ports (7), provides working power by power module (5) for each device.

2. the differential pressure transmitter data fusion method based on DSP adopts claims 1 described differential pressure transmitter data fusion device based on DSP to carry out data fusion, it is characterized in that the step of moving is as follows:

1) initialization of system: E ²The initialization of PROM (4) and microprocessor DSP (6);

2) reception of signal: the data of differential pressure transmitter (1) output are input among the microprocessor DSP (6) through signal condition module (2), AD converter (3);

3) microprocessor DSP (6) sets initialization model, initialization model parameter according to theoretical foundation of support vector machine, carry out the training of model according to the data that received, obtain the support vector machine and the parameter of training pattern, eliminate temperature, pressure interference, improve the precision of measured value;

4) microprocessor DSP (6) receives new data, and the training pattern of utilizing step 3) to set up is calculated output valve;

5) microprocessor DSP (6) deposits the data of handling well in E ²Export among the PROM (4) and through serial ports (7); And to step 4).

3. according to the differential pressure transmitter data fusion method of claims 2 described DSP, it is characterized in that the concrete operations step of described step 3) is:

1. set up and initialization model: comprise the parameter of initialization model, selected kernel function and parameter thereof;

2. given input value of normalization and desired destination value, formula is as follows:

$\stackrel{\&OverBar;}{u_i}=2\&times;\frac{u_i-u_{\min }}{u_{\max }-u_{\min }}-1$

In the formula:

: the differential pressure pick-up output voltage the after-normalization; u _i-differential pressure pick-up output voltage; u _Min-differential pressure pick-up output voltage minimum value; u _Max-differential pressure pick-up output voltage maximal value;

3. the initialization model of usefulness foundation and sample data are according to formula $f^{*}\left(x\right)=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{i}k(x_i,x)+b$ Calculate differential pressure value;

In the formula:

β _i:-kernel function coefficient; B :-threshold value; K (x _i, x) :-kernel function; X :-input value; x _i:-support vector; f ^*(x) :-regression function; K: a support vector machine number;

4. obtain the deviation that calculates differential pressure value and expectation differential pressure value;

5. the parameter in Model parameter and the kernel function is learnt;

Return 2. double counting of step, till error meets the demands;

6. determine parameter, kernel function and the parameter thereof of training pattern.

Images