CN-122020125-A - Nonlinear error correction method of online analysis instrument based on manifold learning

CN122020125ACN 122020125 ACN122020125 ACN 122020125ACN-122020125-A

Abstract

The invention belongs to the technical field of data processing, and particularly relates to a nonlinear error correction method of an online analysis instrument based on manifold learning, which comprises the steps of obtaining a high-dimensional characteristic vector set containing original physical quantity readings and auxiliary environment parameters; and calculating signal transient response entropy and environment coupling stress index of each dimension data, and further deducing manifold shear space distortion rate representing the bending degree of the data structure. On the basis, a distortion weighted distance measure is constructed to replace the traditional Euclidean distance, the improved local linear embedding algorithm is utilized to map the high-dimensional features to the low-dimensional space, and finally an error prediction model is established through a least square support vector machine. By introducing physical perception measurement, the problems of data manifold curling and Euclidean distance invalidation caused by environmental drastic change are reduced, neighborhood selection errors are avoided, and the measurement accuracy and stability of the instrument under the multi-physical field coupling environment are improved.

Inventors

LIU FUHUA
WANG WENTONG
WANG XIN
ZHANG DAOQI
LIU HAONAN
LIU KEZENG

Assignees

青岛三华泰工程技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260210

Claims (10)

1. The nonlinear error correction method of the online analysis instrument based on manifold learning is characterized by comprising the steps of obtaining a high-dimensional characteristic vector set containing original physical quantity readings and auxiliary environment parameters; The manifold cut space distortion rate of each dimension data of each high-dimensional feature vector is calculated, wherein the manifold cut space distortion rate is positively correlated with an environment coupling stress index, the environment coupling stress index is positively correlated with the product of the signal transient response entropy of the corresponding dimension data, the environment sensitivity coefficient and the time change rate of the auxiliary environment parameter, and the signal transient response entropy comprises the shannon entropy of probability distribution of all values of the corresponding dimension data in a set window and is positively correlated with the ratio of the variance and the mean of all values in the window; Calculating the comprehensive manifold cut space distortion rate of each high-dimensional feature vector, wherein the comprehensive manifold cut space distortion rate is the arithmetic average value of the manifold cut space distortion rates of all the dimensional data under the same high-dimensional feature vector; using a local linear embedding algorithm to replace Euclidean distance with the distortion weighted distance, and mapping the high-dimensional feature vector set to a low-dimensional space to obtain a low-dimensional feature vector sequence; The method comprises the steps of training a prediction model based on a low-dimensional feature vector sequence, acquiring real-time original physical quantity readings, and acquiring corrected measured values based on the real-time original physical quantity readings and the prediction model.
2. The online analysis meter nonlinear error correction method based on manifold learning according to claim 1, wherein the signal transient response entropy satisfies the expression: ; In the formula, Signal transient response entropy representing the c-th dimension data of the i-th sampling point; Representing a sliding time window length; the ith sample point is represented by the ith sample point within the sliding time window Probability distribution density of the c-th dimension data of the sampling points; representing the variance of the c-th dimension data within the sliding time window of the i-th sampling point; Representing the mean value of the c dimension data in the sliding time window of the i sampling point; Representing a natural logarithmic function; Representing a minute value.
3. The online analytical instrument nonlinear error correction method based on manifold learning according to claim 1, wherein the environmental coupling stress index satisfies the expression: ; In the formula, An environmental coupling stress index representing the c-th dimension data of the i-th sampling point; signal transient response entropy representing the c-th dimension data of the i-th sampling point; representing an environmental sensitivity coefficient; a time rate vector representing the normalized auxiliary environmental parameter sequence component of the i-th sampling point; modulo length symbols representing vectors; Representing the normalization function.
4. The online analysis meter nonlinear error correction method based on manifold learning according to claim 1, wherein the manifold cut space distortion rate satisfies the expression: ; In the formula, Manifold cut space distortion rate of the c dimension data representing the i-th sampling point; an environmental coupling stress index representing the c-th dimension data of the i-th sampling point; representing instantaneous distortion weights; Representing the cumulative distortion weight; A number of sampling points representing a memory period; environmental coupling stress index of the c-th dimension data of the i-m-th sampling point representing the memory period of the i-th sampling point.
5. The online analysis meter nonlinear error correction method based on manifold learning according to claim 1, wherein the distortion weighted distance satisfies the expression: ; In the formula, Representing a distortion weighted distance between an ith sample point and a jth sample point; Represent the first High-dimensional feature vector of each sampling point And the first High-dimensional feature vector of each sampling point Squaring the Euclidean distance between the two; And (3) with Respectively representing the integrated manifold tangential space distortion rate of the ith sampling point and the jth sampling point; modulo long symbols representing vectors.
6. The online analysis meter nonlinear error correction method based on manifold learning according to claim 1, wherein the training of the prediction model based on the low-dimensional feature vector sequence comprises: the prediction model is a least square support vector machine, the kernel function is a radial basis kernel function, the low-dimensional feature vector sequence is taken as input by the prediction model, and the difference value between the original physical quantity reading and the standard true value at the corresponding sampling moment is taken as output by the prediction model.
7. The method for nonlinear error correction of an on-line analysis meter based on manifold learning according to claim 1, wherein the obtaining real-time raw physical quantity readings, based on the real-time raw physical quantity readings and a predictive model, obtains corrected measured values, comprises: In the real-time running process of the instrument, acquiring a real-time original physical quantity reading sequence and a real-time auxiliary environment parameter sequence, and constructing a real-time high-dimensional feature vector; and (3) performing difference between the real-time original physical quantity reading sequence of the instrument and the prediction error to obtain a corrected measured value.
8. The online analysis meter nonlinear error correction method based on manifold learning of claim 7, wherein the constructing a real-time low-dimensional feature vector based on the real-time high-dimensional feature vector comprises: calculating distortion weighted distance of each vector in the real-time high-dimensional feature vector and the high-dimensional feature vector set, and selecting The nearest-neighbor of the one is, Is a preset value; And carrying out linear weighted combination on the low-dimensional feature vector sequence corresponding to the nearest neighbor by utilizing the reconstructed weight vector to obtain the real-time low-dimensional feature vector.
9. The online analysis meter nonlinear error correction method based on manifold learning of claim 7, wherein the obtaining the prediction error comprises: And inputting the real-time low-dimensional feature vector into an error prediction model, and outputting a prediction error.
10. The online analysis meter nonlinear error correction method based on manifold learning according to claim 4, wherein the acquisition of the memory period comprises: Placing an online analysis instrument in a constant temperature environment, and stabilizing the indication; adjusting the set temperature to raise the ambient temperature Recording the change from temperature to final steady state change in output signal amplitude The time period is taken as a memory period.

Description

Nonlinear error correction method of online analysis instrument based on manifold learning Technical Field The invention relates to the technical field of data processing. More particularly, the invention relates to an online analysis meter nonlinear error correction method based on manifold learning. Background The on-line analysis instrument is used as key equipment for industrial process control, is widely applied to the fields of petrochemical industry, steel smelting, environmental protection monitoring and the like, and the real-time property and accuracy of measurement data directly determine the optimal control and safe operation level of the production process. In the application scenario of the actual industrial field, the analysis instrument is placed in a complex environment of multi-physical field coupling for a long time. When the sensor works, the sensor responds to the concentration change of the component to be measured, and is also inevitably subjected to the cooperative impact of external interference factors such as abrupt change of the temperature of the measuring chamber, fluctuation of the sample gas pressure, unstable carrier gas flow, change of the ambient humidity and the like. These environmental parameters tend to exhibit dynamic, non-linear, varying characteristics, rather than constant values in an ideal state. However, existing error correction techniques still have limitations in addressing such issues. Particularly when environmental parameters of an industrial site fluctuate strongly, the sensor tends to enter a strongly nonlinear response area, resulting in severe bending or folding of the data manifold that describes the state of the system. Under the specific scene, the Euclidean distance measurement mode adopted in the traditional manifold learning algorithm cannot sense the physical stress and the history memory effect of the sensor caused by environmental drafts, so that the algorithm can easily cross manifold bending parts by mistake when searching for a reference neighborhood for correction, and the geometric distance of space coordinates is close, but sample points with different actual physical evolution states are misjudged to be similar neighbors. The error prediction model can not accurately reconstruct the current nonlinear drift characteristic due to the error of the neighborhood selection caused by the measurement failure, and finally, the measured value correction failure of the instrument is caused, so that the reliability of the monitoring data is seriously affected. Disclosure of Invention In order to solve the technical problem of poor nonlinear error correction effect of the online analysis instrument, the invention provides an online analysis instrument nonlinear error correction method based on manifold learning, which comprises the following steps: The method comprises the steps of obtaining a high-dimensional feature vector set containing original physical quantity readings and auxiliary environment parameters, calculating manifold cut space distortion rate of each dimensional data of each high-dimensional feature vector, wherein the manifold cut space distortion rate is positively correlated with environment coupling stress indexes, the product of the environment coupling stress indexes and signal transient response entropy of corresponding dimensional data, environment sensitivity coefficients and time change rates of the auxiliary environment parameters is positively correlated, the signal transient response entropy comprises shannon entropy of probability distribution of all values of the corresponding dimensional data in a set window and is positively correlated with the ratio of variance and mean value of all values in the window, calculating comprehensive manifold cut space distortion rate of each high-dimensional feature vector, the comprehensive manifold cut space distortion rate is an arithmetic average value of manifold cut space distortion rate of all dimensional data under the same high-dimensional feature vector, constructing distortion weighted distance based on the distance of the high-dimensional feature vector set and the product of the corresponding comprehensive manifold cut space distortion rate, using a local linear embedding algorithm to replace Euclidean distance to be weighted distance, obtaining a low-dimensional feature vector based on the low-dimensional feature vector set, and obtaining a real-time feature vector prediction model based on the low-dimensional feature vector, and obtaining the original physical feature vector. According to the method, the original physical quantity reading and the auxiliary environment parameter are fused, the comprehensive manifold shear space distortion rate reflecting the bending degree of the data structure is calculated, and the distortion weighted distance is constructed according to the distortion rate to replace the Euclidean distance. The method and the device can sense physical