CN-121980205-A - Method and system for predicting service life of radioactive separation resin based on machine learning

CN121980205ACN 121980205 ACN121980205 ACN 121980205ACN-121980205-A

Abstract

The invention provides a radioactive separation resin life prediction method and a radioactive separation resin life prediction system based on machine learning, and relates to the technical field of radioactive waste liquid treatment, wherein the method comprises the steps of acquiring real-time online monitoring data of radioactive separation resin in the operation process, wherein the real-time online monitoring data comprise local adsorption capacity, local temperature and local irradiation dose acquired from three points of the upper part, the middle part and the lower part which are axially distributed on a resin column; and carrying out feature fusion on the real-time on-line monitoring data and the microstructure degradation quantization index under the current working condition of the resin to construct a fusion feature set. The invention ensures the continuous, stable and safe operation of the radioactive waste liquid treatment and nuclide separation system.

Inventors

GU LONG
Su Xingkang
WANG GUAN

Assignees

福建睿斯科医疗技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (10)

1. A method for predicting the lifetime of a radioactive separation resin based on machine learning, the method comprising: Acquiring real-time online monitoring data of the radioactive separation resin in the running process, wherein the real-time online monitoring data comprise local adsorption capacity, local temperature and local irradiation dose acquired from three points of the upper part, the middle part and the lower part which are axially distributed on a resin column; Based on real-time on-line monitoring data, calculating to obtain a microstructure degradation quantization index of the resin under the current working condition through a constructed microstructure degradation proxy model; performing feature fusion on the real-time online monitoring data and the microstructure degradation quantization index to construct a fusion feature set; Inputting the fusion feature set into a pre-trained extreme working condition generalization enhanced machine learning model to obtain a reference residual life predicted value of the resin under the current working condition; judging whether the current working condition belongs to an extreme working condition in real time, wherein the extreme working condition comprises instantaneous strong irradiation, feed liquid acidity mutation and competitive ion concentration mutation, and if so, constructing a state distribution curve along the axial direction of the resin column according to the local adsorption capacity, local temperature and local irradiation dose of three points of the upper part, the middle part and the lower part; and (3) carrying out curvature calculation and inflection point detection on the state distribution curve to obtain local degradation correction factors of each axial section, generating a comprehensive space correction coefficient through weighted fusion, correcting the reference residual life predicted value by using the comprehensive space correction coefficient to obtain a final residual life predicted value, and if the final residual life predicted value does not belong to the standard residual life predicted value, directly taking the reference residual life predicted value as the final residual life predicted value.
2. The machine learning based radioactive separation resin life prediction method of claim 1, wherein acquiring real-time on-line monitoring data of the radioactive separation resin during operation, the real-time on-line monitoring data including local adsorption capacity, local temperature and local irradiation dose acquired from three points of upper, middle and lower portions of the resin column axial distribution, comprises: An adsorption capacity sensor, a temperature sensor and an irradiation dose detector are respectively arranged at the upper part, the middle part and the lower part of the resin column, and local adsorption capacity, local temperature and local irradiation dose original signals of all points are acquired in real time; Performing analog-to-digital conversion on the acquired original signals to generate corresponding time sequence data, and performing time stamp synchronous alignment on the time sequence data of the local adsorption capacity, the local temperature and the local irradiation dose of the three points to obtain synchronized data; And detecting and removing abnormal values of the synchronized data by adopting a sliding window method to obtain the preprocessed real-time online monitoring data.
3. The machine learning-based radioactive separation resin life prediction method of claim 2, wherein the calculation of the microstructure degradation quantization index of the resin under the current working condition by the constructed microstructure degradation proxy model based on real-time online monitoring data comprises the following steps: Inputting the preprocessed real-time online monitoring data into a microstructure degradation agent model which is built in advance, wherein the microstructure degradation agent model is a regression model which is obtained by training by adopting a deep neural network based on resin offline detection results under different working conditions in historical operation data; And carrying out multi-layer nonlinear mapping calculation on input data through a microstructure degradation proxy model, and outputting microstructure degradation quantization indexes, wherein the microstructure degradation quantization indexes comprise pore channel collapse coefficients used for representing collapse degrees of pore channels in the resin and functional group retention rates used for representing falling degrees of functional groups, the pore channel collapse coefficients are indirectly calculated through attenuation rates of specific surface areas of the resin, and the functional group retention rates are indirectly calculated through attenuation rates of ion exchange capacities.
4. The machine learning-based radioactive separation resin life prediction method of claim 3, wherein feature fusion is performed on real-time on-line monitoring data and microstructure degradation quantization indexes, and a fusion feature set is constructed, comprising: and performing feature layer splicing on the local adsorption capacity, the local temperature and the local irradiation dose in the preprocessed real-time online monitoring data, the pore collapse coefficient and the functional group retention rate to generate an initial fusion feature vector, and performing normalization processing on the initial fusion feature vector to obtain a standardized fusion feature set.
5. The machine learning-based radioactive separation resin life prediction method of claim 4, wherein inputting the fusion feature set into a pre-trained extreme condition generalization enhanced machine learning model to obtain a reference residual life prediction value of the resin under the current condition comprises: The standardized fusion feature set is input into a polar working condition generalization enhancement type machine learning model, the model is constructed based on a deep neural network, the fusion feature set is received through an input layer, nonlinear feature transformation and mapping are carried out through a plurality of hidden layers, finally a one-dimensional numerical value is generated through an output layer and is used as a reference residual life prediction value under the current working condition, and the polar working condition generalization enhancement type machine learning model is optimized in a training stage through introducing an antagonism sample and physical consistency constraint.
6. The method for predicting the service life of the radioactive separation resin based on machine learning of claim 5, wherein the method for predicting the service life of the radioactive separation resin based on machine learning is characterized by judging whether the current working condition belongs to an extreme working condition in real time, wherein the extreme working condition comprises instantaneous strong irradiation, feed liquid acidity mutation and competitive ion concentration mutation, and if so, constructing a state distribution curve along the axial direction of the resin column according to the local adsorption capacity, the local temperature and the local irradiation dose of three points of the upper part, the middle part and the lower part, wherein the method comprises the following steps: Judging whether the current working condition belongs to an extreme working condition in real time, wherein the extreme working condition comprises instantaneous strong irradiation, feed liquid acidity mutation and competitive ion concentration mutation; if the method belongs to the field, respectively carrying out normalization treatment on the local adsorption capacity, the local temperature and the local irradiation dose of the three points to obtain an adsorption capacity normalization value, a temperature normalization value and an irradiation dose normalization value; according to the comprehensive state index value, the corresponding axial coordinate values of the upper part, the middle part and the lower part of the resin column are taken as interpolation nodes, the comprehensive state index value of each node is taken as a node function value, a cubic spline interpolation method is adopted to construct a state distribution curve which is continuous along the axial direction of the resin column, specifically, a cubic polynomial is constructed between every two adjacent nodes, all the cubic polynomials have the same function value, first derivative value and second derivative value at the nodes, and the coefficients of the cubic polynomials of each segment are determined by solving a three-bending moment equation set established by the continuity condition of the second derivative at the nodes, so that the state distribution curve formed by the cubic polynomials of the segments is obtained.
7. The method for predicting the life of a radioactive separation resin based on machine learning according to claim 6, wherein the method for predicting the life of a radioactive separation resin based on machine learning is characterized in that curvature calculation and inflection point detection are performed on a state distribution curve to obtain local degradation correction factors of each axial section, a comprehensive space correction coefficient is generated by weighted fusion, and a reference residual life prediction value is corrected by the comprehensive space correction coefficient to obtain a final residual life prediction value, and if the predicted value does not belong to the predicted value, the method for predicting the life of the radioactive separation resin based on machine learning is characterized in that the reference residual life prediction value is directly used as the final residual life prediction value, and the method comprises the steps of: the curvature value is detected, the axial position corresponding to the local maximum point of the curvature is marked as a potential degradation mutation point, and the potential degradation mutation point is taken as an inflection point of the state distribution curve; Dividing the resin column into a plurality of continuous degradation sections according to inflection points, integrating curvature values in each degradation section to obtain accumulated degradation intensity of the sections, taking the ratio of the accumulated degradation intensity to the section length as a local degradation correction factor of the sections; And if the current working condition is judged not to belong to the extreme working condition in real time, directly outputting the reference residual life predicted value as the final residual life predicted value.
8. A machine learning based radioactive separation resin life prediction system implementing the method of any one of claims 1 to 7, comprising: The acquisition module is used for acquiring real-time online monitoring data of the radioactive separation resin in the operation process, wherein the real-time online monitoring data comprise local adsorption capacity, local temperature and local irradiation dose acquired from three points of the upper part, the middle part and the lower part which are axially distributed on the resin column; the calculation module is used for calculating and obtaining the microstructure degradation quantification index of the resin under the current working condition through the constructed microstructure degradation agent model based on the real-time online monitoring data; the fusion module is used for carrying out feature fusion on the real-time online monitoring data and the microstructure degradation quantization index to construct a fusion feature set; The prediction module is used for inputting the fusion feature set into a pre-trained extreme working condition generalization enhancement type machine learning model to obtain a reference residual life prediction value of the resin under the current working condition; The judging module is used for judging whether the current working condition belongs to an extreme working condition in real time, wherein the extreme working condition comprises instantaneous strong irradiation, feed liquid acidity mutation and competitive ion concentration mutation, and if so, a state distribution curve along the axial direction of the resin column is constructed according to the local adsorption capacity, the local temperature and the local irradiation dose of three points of the upper part, the middle part and the lower part; The correction module is used for carrying out curvature calculation and inflection point detection on the state distribution curve to obtain local degradation correction factors of each axial section, generating a comprehensive space correction coefficient through weighted fusion, correcting the reference residual life predicted value by utilizing the comprehensive space correction coefficient to obtain a final residual life predicted value, and if the reference residual life predicted value does not belong to the final residual life predicted value, directly taking the reference residual life predicted value as the final residual life predicted value.
9. A computing device, comprising: One or more processors; Storage means for storing one or more programs which when executed by the one or more processors cause the one or more processors to implement the method of any of claims 1to 7.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a program which, when executed by a processor, implements the method according to any of claims 1 to 7.

Description

Method and system for predicting service life of radioactive separation resin based on machine learning Technical Field The invention relates to the technical field of radioactive waste liquid treatment, in particular to a radioactive separation resin life prediction method and system based on machine learning. Background In the fields of nuclear industry spent fuel post-treatment and radioactive waste liquid deep purification, the radioactive separation resin is a core consumable material for adsorbing and separating key radionuclides such as cesium, strontium and the like, and the service life of the radioactive separation resin directly influences the operation safety of a system and the nuclide separation effect; in the prior art, the method still mainly uses periodic off-line sampling to detect the whole exchange capacity of the resin and artificial experience to judge the service life, obtains the average performance index of the resin, can not carry out real-time on-line monitoring on the local adsorption capacity, temperature and irradiation dose of the upper, middle and lower axial points of the resin column, does not establish a quantitative characterization method for microstructure degradation such as resin pore collapse and functional group falling off, does not establish a machine learning prediction model with generalization enhancement capability aiming at extreme working conditions such as instantaneous strong irradiation, feed liquid acidity mutation and competitive ion concentration mutation, can not be combined with an axial state distribution curve to identify local inflection point and correct the service life prediction result, has the technical defects of on-line monitoring deficiency, unquantifiable microstructure degradation, poor adaptability of the extreme working conditions, unaccounted axial space degradation non-uniformity and low service life prediction precision, and is extremely easy to cause resin service life misjudgment to reduce separation efficiency and even resin failure leakage. Disclosure of Invention The invention provides a radioactive separation resin life prediction method and a radioactive separation resin life prediction system based on machine learning, which ensure continuous, stable and safe operation of a radioactive waste liquid treatment and nuclide separation system. In order to solve the technical problems, the technical scheme of the invention is as follows: In a first aspect, a method for predicting the lifetime of a radioactive separation resin based on machine learning, the method comprising: Acquiring real-time online monitoring data of the radioactive separation resin in the running process, wherein the real-time online monitoring data comprise local adsorption capacity, local temperature and local irradiation dose acquired from three points of the upper part, the middle part and the lower part which are axially distributed on a resin column; Based on real-time on-line monitoring data, calculating to obtain a microstructure degradation quantization index of the resin under the current working condition through a constructed microstructure degradation proxy model; performing feature fusion on the real-time online monitoring data and the microstructure degradation quantization index to construct a fusion feature set; Inputting the fusion feature set into a pre-trained extreme working condition generalization enhanced machine learning model to obtain a reference residual life predicted value of the resin under the current working condition; judging whether the current working condition belongs to an extreme working condition in real time, wherein the extreme working condition comprises instantaneous strong irradiation, feed liquid acidity mutation and competitive ion concentration mutation, and if so, constructing a state distribution curve along the axial direction of the resin column according to the local adsorption capacity, local temperature and local irradiation dose of three points of the upper part, the middle part and the lower part; and (3) carrying out curvature calculation and inflection point detection on the state distribution curve to obtain local degradation correction factors of each axial section, generating a comprehensive space correction coefficient through weighted fusion, correcting the reference residual life predicted value by using the comprehensive space correction coefficient to obtain a final residual life predicted value, and if the final residual life predicted value does not belong to the standard residual life predicted value, directly taking the reference residual life predicted value as the final residual life predicted value. Further, acquiring real-time on-line monitoring data of the radioactive separation resin in the operation process, wherein the real-time on-line monitoring data comprises local adsorption capacity, local temperature and local irradiation dose acquired from three points of the upper part, the