Intelligent analysis method for corrosivity of sulfide molecules in transformer oil

Abstract

The invention relates to the technical field of researching or analyzing materials by testing chemical or physical properties of materials, in particular to an intelligent analysis method for corrosivity of sulfide molecules in transformer oil. The influence of the oil environment on the adsorption energy is simulated and corrected in combination with molecular dynamics, a microscopic parameter database is constructed, and a corrosion risk grade classification rule and a generation rate prediction model are established. Oil data are collected in real time and input into the model to output a dynamic risk level, and a circulation rate adjusting or corrosion inhibitor adding instruction is triggered. Model parameters are regularly corrected by combining a closed-loop feedback mechanism, and a microscopic database is expanded, so that transfer learning prediction of unknown sulfide corrosion behaviors is realized. According to the method, the limitation of an existing single experiment condition is broken through, the relevance between a microcosmic molecular mechanism and a macroscopic corrosion behavior is dynamically analyzed, and the prediction precision under the multi-factor coupling effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of researching or analyzing materials by testing the chemical or physical properties of materials, and particularly relates to an intelligent analysis method for the corrosivity of sulfide molecules in transformer oil. Background Art

[0002] The corrosivity of sulfide molecules in transformer oil to copper conductors is mainly determined by their molecular structures and the interaction mechanisms with the metal surface. Research shows that dibenzyl disulfide (DBDS) is prone to break the disulfide bond (-S-S-) in the molecule during the adsorption process on the copper surface to generate benzylthio radicals, form stable S-Cu bonds with copper through high adsorption energy, and cause a significant decrease in the Fermi level, thus showing the strongest corrosivity; while diphenyl sulfone (DPS) and diphenyl sulfoxide (DPSO) only form weak O-Cu bonds because the sulfur atoms are in a high oxidation state or the steric hindrance prevents sulfur from directly contacting copper, so they have no significant corrosivity. The corrosivity intensity is positively correlated with the sulfide concentration, and high temperature (such as 150 °C) and long-term aging (such as 72 hours) will accelerate the reaction kinetics and promote the formation of copper sulfide (Cu2S). Experiments and density functional theory (DFT) analysis show that the adsorption energy and the change in the electron density of states distribution of sulfides are directly related to the corrosivity, providing a theoretical basis for the prediction and protection of sulfur corrosion in transformer oil.

[0003] In the field of power equipment, the corrosivity assessment of sulfides in transformer oil is crucial for equipment reliability. Existing methods mostly measure the corrosion behavior of specific sulfides based on single experimental conditions, and it is difficult to dynamically analyze the correlation between the molecular structure, adsorption mechanism, and macroscopic corrosion characteristics. Existing experimental means are limited by time-consuming, high cost, and the inability to predict the corrosion kinetics under the coupling action of multiple factors (such as concentration gradient, temperature change, aging duration) in real time. In particular, there is a lack of intelligent analysis of microscopic parameters such as the adsorption energy and electron density of sulfide molecules on the copper surface, resulting in the lag and experience dependence of corrosion risk assessment, and it is difficult to accurately guide the optimization of anti-corrosion strategies for transformer oil. This technical bottleneck limits the initiative and adaptability of corrosion protection for power equipment, and there is an urgent need to develop an intelligent analysis method that integrates microscopic molecular mechanisms and macroscopic corrosion behaviors. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides an intelligent analysis method for the corrosivity of sulfide molecules in transformer oil. The present invention solves the problems that the existing sulfide corrosivity assessment methods cannot dynamically analyze the correlation between microscopic molecular mechanisms and macroscopic corrosion behaviors, resulting in lagging corrosion risk assessment, relying on experience, and being difficult to predict the corrosion kinetics under the coupling action of multiple factors in real time.

[0005] To solve the above technical problems, the specific technical solutions of the present invention are as follows: Intelligent analysis method for the corrosivity of sulfide molecules in transformer oil, comprising: Step S1, obtaining the adsorption energy, electron density of states distribution, and bond-breaking characteristics of sulfide molecules on the copper surface, constructing a microscopic parameter database of sulfide molecules on the copper conductor surface based on the adsorption energy, electron density of states distribution, and bond-breaking characteristics of sulfide molecules on the copper surface, the microscopic parameter database including the oxidation state of sulfide molecules, the bond-breaking mode, and the copper surface interaction strength; Step S2, setting the sulfide concentration gradient, temperature gradient, and aging time gradient of the orthogonal experimental matrix according to the adsorption energy threshold and bond-breaking characteristics in the microscopic parameter database, and generating a multi-factor coupling experimental data set; Step S3, observing the coverage of cuprous sulfide on the copper surface by scanning electron microscopy, and analyzing the change in the absorbance of the oil by combining ultraviolet and visible spectra, extracting the corrosion characteristics of the copper conductor surface and the absorbance value of the oil; Step S4, training a corrosion rate prediction model by the random forest algorithm based on the adsorption energy threshold in the microscopic parameter database, the electron density of states distribution, and the temperature-time product factor in the multi-factor coupling experimental data set, the corrosion rate prediction model taking the adsorption energy threshold, Fermi level shift amount, and environmental factor as input variables and outputting the corrosion risk level; Step S5, collecting the sulfide concentration and temperature data in the transformer oil in real time, and inputting them into the corrosion rate prediction model to generate a dynamic corrosion risk level and a cuprous sulfide generation rate curve; Step S6, triggering the optimization of the protection strategy according to the corrosion risk level, regularly detecting the coverage of cuprous sulfide on the copper conductor surface by scanning electron microscopy, calculating the error with the predicted value of the corrosion rate prediction model, and if the error exceeds the preset range, expanding the multi-factor coupling experimental data and updating the parameter weights of the model and the microscopic parameter database.

[0006] Furthermore, for the intelligent analysis method for the corrosivity of sulfide molecules in transformer oil of the present invention, the step S1 includes: Simulating the adsorption configuration of sulfide molecules on the copper surface by density functional theory, and calculating the initial adsorption energy and electron density of states distribution; Based on molecular dynamics, simulating the diffusion path and solvation effect of sulfide in transformer oil, and correcting the environmental influence parameters of the initial adsorption energy according to the energy barrier distribution of the diffusion path and the change in solvation free energy; Integrating the corrected adsorption energy, electron density of states distribution, and sulfide molecule bond-breaking characteristics into structured parameters, and associating them with the sulfide type to form a data set including the molecular oxidation state, bond-breaking mode, and copper surface interaction strength.

[0007] Further, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil according to the present invention, step S2 includes: According to the adsorption energy threshold and bond-breaking characteristics stored in the microscopic parameter database, set the sulfide concentration gradient, temperature gradient, and aging time gradient in the orthogonal experimental matrix; Observe the coverage of cuprous sulfide on the copper surface through a scanning electron microscope, and analyze the absorbance change of the oil through ultraviolet and visible spectroscopy. Quantify the corrosion intensity according to the linear relationship between the coverage and the absorbance change rate; Correlate and map the product factor of the temperature gradient and the aging time gradient in the experimental conditions with the adsorption energy threshold and the electron density of states distribution in the microscopic parameter database to generate a training sample including the product factor of temperature and time and the corrosion rate.

[0008] Further, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil according to the present invention, step S4 includes: Perform eigenvector splicing on the adsorption energy threshold and the Fermi level shift of the electron density of states distribution in the microscopic parameter database with the product factor of temperature and time in the multi-factor coupling experimental dataset as the input features of the random forest algorithm; Conduct feature importance analysis by calculating the Gini coefficient, identify the non-linear correlation between the adsorption energy and the corrosion rate, and establish a corrosion risk level classification rule based on the decision tree splitting rule; Deploy the corrosion risk level classification rule as a dynamic prediction module of the corrosion rate prediction model, and output a prediction result including the corrosion risk level and the cuprous sulfide generation rate curve.

[0009] Further, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil according to the present invention, step S5 includes: Monitor the sulfide concentration through a fiber optic spectral sensor, and achieve real-time synchronization with the temperature sensor data through timestamp alignment; Input the synchronized real-time data into the corrosion rate prediction model to generate a corrosion risk level and a corrosion prevention strategy suggestion; When the corrosion risk level exceeds the preset threshold, adjust the oil parameters by adjusting the oil circulation rate or adding a corrosion inhibitor, and trigger an early warning signal.

[0010] Further, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil according to the present invention, step S6 includes: Regularly detect the coverage of cuprous sulfide on the surface of the copper conductor through a scanning electron microscope, and calculate the root mean square error between the measured coverage and the predicted value of the corrosion rate prediction model; When the error exceeds the preset range, increase the experimental data samples under different solvation environments, and optimize the feature weights by retraining the decision tree node splitting threshold in the random forest model; Supplement the adsorption energy correction parameters and bond-breaking characteristics corresponding to the newly added sulfide types to the microscopic parameter database, and update the molecular oxidation state classification index.

[0011] Furthermore, the intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to the present invention further includes: Migrate the electron state density distribution characteristics of known sulfides to the corrosion rate prediction model through feature space mapping for predicting the corrosion rate of unknown sulfide types; Align the time stamp sequence of multi-source sensor data with the adsorption energy correction parameters obtained by molecular dynamics simulation in the time dimension; Establish a feedback mechanism between the prediction result and the experimental design. According to the residual analysis result of the corrosion rate prediction model, dynamically adjust the temperature gradient setting of subsequent multi-factor coupling experiments, and update the new experimental data to the microscopic parameter database in reverse.

[0012] Advantages of the present invention; By constructing a microscopic parameter database for the interaction between sulfide molecules and the copper surface, combining multi-factor coupling experimental design and random forest algorithm to train the corrosion rate prediction model, the present invention realizes the dynamic correlation analysis of the microscopic adsorption mechanism and the macroscopic corrosion behavior, and solves the technical defects of the existing evaluation methods that rely on single experimental conditions and cannot predict the multi-factor coupling effect in real time. Based on density functional theory to calculate the adsorption energy and electron state density distribution, molecular dynamics simulation to correct the environmental parameters, orthogonal experiment to generate macroscopic characteristics such as the product factor of temperature and time, establish the corrosion risk classification rules through feature fusion and Gini coefficient analysis, the dynamic prediction module outputs the risk level and generates the rate curve in real time, and combines the closed-loop feedback optimization mechanism to regularly correct the model parameters and the database, significantly improving the real-time performance and accuracy of corrosion rate prediction, reducing the dependence on empirical judgment, and providing an intelligent solution for the evaluation of the corrosivity of sulfides in transformer oil. Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for the embodiments will be briefly introduced below. Obviously, for those of ordinary skill in the art, other drawings can be obtained according to the drawings without creative efforts.

[0014] Figure 1 It is a flowchart of the intelligent analysis method for the corrosivity of sulfide molecules in transformer oil provided by the embodiment of the present invention. Detailed Embodiments

[0015] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention. The following will describe in detail the technical solutions provided by each embodiment of the present invention with reference to the drawings. To better understand the objectives of the present invention, the present invention will be further described in detail below.

[0016] Please refer to Figure 1 , the intelligent analysis method for the corrosivity of sulfide molecules in transformer oil of the present invention includes: Step S1, obtaining the adsorption energy, electron density of states distribution, and bond-breaking characteristics between sulfide molecules and the copper surface, and constructing a microscopic parameter database of sulfide molecules on the copper conductor surface based on the adsorption energy, electron density of states distribution, and bond-breaking characteristics between sulfide molecules and the copper surface. The microscopic parameter database includes the oxidation state of sulfide molecules, the bond-breaking mode, and the copper surface interaction strength; In step S1, the adsorption configuration of sulfide molecules on the copper surface is simulated by density functional theory, and the spatial arrangement between sulfide molecules and copper atoms is optimized by the first-principles calculation method to determine the most stable adsorption site and molecular orientation. Based on quantum mechanics calculations, the initial adsorption energy and electron density of states distribution are obtained, where the adsorption energy characterizes the binding strength between sulfide molecules and the copper surface, and the electron density of states distribution reflects the degree of interaction between the molecular orbit and the electron structure of the metal surface. This process reveals the microscopic adsorption mechanism of sulfide molecules on the copper conductor surface through quantum chemistry calculations, providing a theoretical data basis for subsequent parameter correction.

[0017] Combined with molecular dynamics to simulate the diffusion behavior of sulfide in transformer oil, and simulate the movement trajectory and solvation effect of sulfide molecules in the oil medium. By analyzing the energy barrier distribution of the diffusion path, the activation energy required for sulfide molecules to cross different solvation layers is calculated; at the same time, the change in solvation free energy is evaluated to quantify the influence of the oil environment on the stability of sulfide molecules. Based on the diffusion kinetics and thermodynamic parameters, the environmental influence factor of the initial adsorption energy is corrected to make the adsorption energy data more conform to the molecular interaction characteristics in the actual oil environment.

[0018] Integrate the corrected adsorption energy, electron density of states distribution, and bond-breaking characteristics of sulfide molecules into structured parameters to construct a microscopic parameter database for sulfide molecules on the copper surface. The bond-breaking characteristics are analyzed by molecular dynamics trajectories to obtain the bond-breaking probability and sites of chemical bonds in sulfide molecules. Combining the oxidation state classification and bond-breaking modes, a mapping relationship between the sulfide type and the interaction strength on the copper surface is established. The structured parameters in the database are indexed by sulfide molecule type, associating their oxidation states, bond-breaking modes, and corrected adsorption energy values to form a multi-dimensional data system, providing microscopic mechanism support for multi-factor coupling experimental design and corrosion prediction model training.

[0019] Step S2: According to the adsorption energy threshold and bond-breaking characteristics in the microscopic parameter database, set the sulfide concentration gradient, temperature gradient, and aging time gradient of the orthogonal experimental matrix to generate a multi-factor coupling experimental dataset. In step S2, based on the microscopic parameter database constructed in step S1, the adsorption energy threshold and bond-breaking characteristics serve as the core basis for orthogonal experimental design. The adsorption energy threshold is used to screen the critical value of the adsorption activity of sulfide molecules on the copper surface and determine the upper and lower limits of the sulfide concentration gradient. For example, sulfides with an adsorption energy higher than 1.5 eV are set with a concentration gradient of 0.1 - 1.0 ppm to cover the actual working conditions corresponding to their high corrosion activity. The bond-breaking characteristics, by analyzing the bond-breaking mode and stability of sulfide molecules and combining with the interaction strength on the copper surface, set the experimental conditions for the temperature gradient and aging time gradient. For example, sulfides with a lower bond-breaking energy are set with a temperature gradient of 80 - 150 °C and an aging time gradient of 24 - 72 hours to simulate the corrosion kinetics under different environmental stresses.

[0020] The design of the orthogonal experimental matrix covers the combined effects of sulfide concentration, temperature, and aging time through multi-factor coupling. For example, the L9(3^4) orthogonal table is used to arrange the experimental groups to systematically characterize the corrosion behavior under the interaction of various variables. By observing the coverage of cuprous sulfide on the copper surface with a scanning electron microscope and analyzing the change in the absorbance of the oil solution with ultraviolet-visible spectroscopy, a linear regression model between the coverage and the absorbance decay rate is established. For example, for every 10% increase in coverage, the absorbance decay rate increases by 0.15, quantifying the corrosion intensity and generating calibrated experimental data labels. The temperature gradient and aging time gradient in the experimental conditions are integrated into an environmental stress parameter through a product factor. For example, the product factor of 150 °C and 72 hours is 10^4, characterizing the synergistic acceleration effect of high-temperature and long-time working conditions, and performing feature fusion with the adsorption energy threshold and electron density of states distribution in the microscopic parameter database to form a multi-dimensional training sample containing the product factor of temperature and time, adsorption energy threshold, and electron density of states.

[0021] The above steps realize the systematic mapping from molecular mechanisms to macroscopic corrosion behaviors through the logical association of micro-parameters guiding experimental design, multi-factor coupling to generate datasets, and feature fusion, providing matching data for input features and output labels of the random forest model.

[0022] Step S3: Observe the coverage of cuprous sulfide on the copper surface through a scanning electron microscope, and combine the ultraviolet and visible spectroscopy to analyze the change in the absorbance of the oil, and extract the corrosion characteristics of the copper conductor surface and the absorbance value of the oil. In step S3, the morphology of the coverage of cuprous sulfide on the copper conductor surface is observed through a scanning electron microscope, and the backscattered electron imaging technology is used to obtain the microscopic distribution characteristics of cuprous sulfide. The scanning electron microscope analyzes the coverage area and thickness through high-resolution imaging. For example, when the coverage area ratio of cuprous sulfide exceeds 20%, it is determined as the medium corrosion grade, and when the thickness exceeds 5 microns, it is determined as the high corrosion grade, which is a direct physical index for quantifying the corrosion degree of the copper surface. Combine the ultraviolet-visible spectroscopy to analyze the change in the absorbance of sulfides in the transformer oil, and calculate the dynamic change rate of the sulfide concentration in the oil through the intensity of the characteristic absorption peak (such as the absorbance attenuation at a wavelength of 280 nm). For example, for every 0.1 decrease in the absorbance attenuation rate, the sulfide concentration decreases by 0.05 ppm, and establish the correlation between the change in the oil component and the corrosion behavior of the copper surface.

[0023] The observation data of the scanning electron microscope and the analysis results of the ultraviolet-visible spectroscopy are fused through a linear regression model. For example, for every 10% increase in the coverage area of cuprous sulfide, the absorbance attenuation rate increases by 0.15, forming a quantitative relationship between the corrosion intensity and the sulfide consumption rate of the oil. The extracted coverage and absorbance values are used as corrosion feature labels. For example, the coverage-absorbance combined index is used to calibrate the corrosion rate under different sulfide types and experimental conditions, providing experimental verification data for subsequent model training. The above data integration process ensures the accuracy and consistency of corrosion feature extraction through the collaborative verification of physical observation and chemical analysis, supporting the effectiveness of the multi-factor coupling experimental dataset.

[0024] Step S4: Based on the adsorption energy threshold and the electron state density distribution in the micro-parameter database and the temperature-time product factor in the multi-factor coupling experimental dataset, train a corrosion rate prediction model through the random forest algorithm. The corrosion rate prediction model takes the adsorption energy threshold, Fermi level shift amount, and environmental factor as input variables and outputs the corrosion risk level. In step S4, the adsorption energy threshold in the microscopic parameter database constructed in step S1, the Fermi level shift of the electron state density distribution, and the temperature-time product factor in the multi-factor coupled experimental dataset generated in step S2 are subjected to eigenvector splicing. The adsorption energy threshold characterizes the binding strength between sulfide molecules and the copper surface. The Fermi level shift reflects the driving effect of interfacial electron transfer on the corrosion reaction. The temperature-time product factor quantifies the acceleration effect of environmental stress on the reaction kinetics. The eigenvectors eliminate the dimensional differences through data standardization to form a multi-dimensional input feature set, providing training data that integrates microscopic mechanisms and macroscopic conditions for the random forest algorithm.

[0025] Through the Gini coefficient calculation in the random forest algorithm, the characteristic importance weights of the adsorption energy, Fermi level shift, and temperature-time product factor on the corrosion rate are evaluated. The Gini coefficient analysis identifies the non-linear correlation law between the adsorption energy and the corrosion rate. For example, the classification contribution degree of the adsorption energy threshold in the range of 1.2 - 1.8 eV to the corrosion risk level is the highest. Based on the characteristic importance ranking, the decision tree splitting rule recursively divides the input feature space to generate the classification threshold of the corrosion risk level. For example, the combination of high adsorption energy, low Fermi level shift, and high temperature-time product factor is classified as a high-risk level, forming the logical judgment condition of the classification rule.

[0026] Deploy the corrosion risk level classification rule to the dynamic prediction module. After inputting the sulfide concentration and temperature data collected in real time, the module outputs the corresponding corrosion risk level by traversing the splitting conditions of the decision tree nodes. The regression tree branch calculates the instantaneous value and cumulative trend of the copper sulfide generation rate based on the adsorption energy threshold, Fermi level shift, and temperature-time product factor, generating a time-dependent corrosion rate curve. The dynamic prediction module updates the decision tree weights through an online learning mechanism to adapt to the changes in the oil environment parameters, realizing the real-time classification of the corrosion risk level and the dynamic correction of the rate curve, and supporting the timely adjustment of the protection strategy. The above steps are logically connected in series through feature fusion, non-linear correlation mining, and model deployment.

[0027] Step S5, collect the sulfide concentration and temperature data in the transformer oil in real time, and input them into the corrosion rate prediction model to generate a dynamic corrosion risk level and a copper sulfide generation rate curve; In step S5, the characteristic absorption spectrum of sulfide in transformer oil is collected in real time by an optical fiber spectral sensor. The sulfide concentration is analyzed based on the change in absorbance at a specific wavelength. For example, the attenuation rate of absorbance at a wavelength of 280 nm has a linear relationship with the sulfide concentration. The temperature sensor synchronously monitors the oil temperature data, and the concentration and temperature data are aligned in time series through timestamp synchronization technology. For example, a sliding window algorithm is used to eliminate the acquisition delay of the sensor, generating a multi-source data set with consistent time dimensions. The synchronized real-time data is input into the corrosion rate prediction model. Based on the decision tree splitting rules and regression tree structure trained in step S4, the adsorption energy threshold, Fermi level shift, and temperature-time product factor are spliced into an input feature vector, and the dynamic corrosion risk level and the copper sulfide generation rate curve are output.

[0028] When the corrosion risk level exceeds the preset threshold, the system changes the diffusion flux of sulfide on the copper surface by adjusting the oil circulation rate. For example, the circulation rate is increased to more than 2.5 m / s to reduce the local sulfide concentration. At the same time, the corrosion inhibitor automatic addition module releases a suitable corrosion inhibitor according to the real-time sulfide type and concentration. For example, 0.05% benzotriazole is added for dibenzyl disulfide (DBDS) to inhibit the adsorption and bond-breaking reaction of sulfide molecules on the copper surface. The warning signal is triggered by an audible and visual alarm or a remote monitoring platform, prompting the operator to perform equipment maintenance or parameter review. The real-time data and operation records are synchronously stored in the database for subsequent model iteration optimization and closed-loop verification of the protection strategy, forming a complete technical link from data acquisition to dynamic response.

[0029] Step S6: Trigger the optimization of the protection strategy according to the corrosion risk level, and regularly detect the coverage of copper sulfide on the surface of the copper conductor by scanning electron microscopy, and calculate the error with the predicted value of the corrosion rate prediction model. If the error exceeds the preset range, expand the multi-factor coupling experimental data and update the parameter weights and microscopic parameter database of the model.

[0030] In step S6, the optimization of the protection strategy is triggered according to the risk level output by the corrosion rate prediction model. For example, when the risk level is high risk, the oil circulation rate is automatically adjusted to more than 2.5 m / s or 0.05% suitable corrosion inhibitor is added to inhibit sulfide adsorption. The coverage of copper sulfide on the surface of the copper conductor is regularly detected by scanning electron microscopy, and the backscattered electron imaging technology is used to quantify the coverage area and thickness. For example, when the coverage area exceeds 25% or the thickness exceeds 5 microns, it is determined as the measured high corrosion state. The deviation between the measured coverage and the model prediction value is calculated by the root mean square error formula. For example, when the error exceeds 5%, it is determined that the model prediction accuracy is insufficient and data expansion and parameter update are required.

[0031] When the root mean square error exceeds the preset range, expand the multi-factor coupling experimental data to cover more solvation environments (such as changes in oil polarity or additive concentration gradients). For example, design a new orthogonal experimental matrix, increase the temperature gradient to 200 °C or the sulfide concentration gradient to 2.0 ppm to generate a supplementary data set. The new data is used to retrain the random forest model, re-evaluate the feature importance weights of the adsorption energy threshold and the temperature-time product factor based on the Gini coefficient, and optimize the decision tree node splitting threshold. For example, increase the contribution weight of the adsorption energy threshold to classification from 0.35 to 0.45 to enhance the model's prediction ability for high-error regions.

[0032] The adsorption energy correction parameters corresponding to the new sulfide types are recalculated by molecular dynamics simulation and density functional theory. For example, simulate the diffusion path energy barrier and solvation free energy change of unknown sulfide molecules in oil to correct the adsorption energy threshold and supplement the bond-breaking characteristics. The updated microscopic parameter database expands the sulfide molecular oxidation state classification index. For example, add the mapping relationship between the sulfide type and the interaction strength with the copper surface to form a dynamic closed-loop feedback mechanism for experimental data, model parameters, and the database.

[0033] In step S1, calculate the initial adsorption energy and electron density of states distribution of sulfide molecules and copper surface by density functional theory, simulate the adsorption configuration of sulfide molecules on the copper surface, and obtain the interaction parameters between the molecule and the metal interface. Combine the molecular dynamics simulation of the diffusion path and solvation effect of sulfide in transformer oil, analyze the diffusion energy barrier and solvation free energy change, and correct the environmental impact parameters of the adsorption energy to make the microscopic parameter database include the dynamic correlation of sulfide oxidation state, bond-breaking mode, and copper surface interaction. This process uses quantum chemical calculations and molecular kinematics simulations to establish a physical and chemical property data set of sulfide molecules on the copper conductor surface, providing microscopic mechanism support for subsequent experiments and model training.

[0034] In step S2, based on the adsorption energy threshold and bond-breaking characteristics constructed in step S1, use the orthogonal experimental design method to set the sulfide concentration, temperature, and aging time gradients to generate an experimental data set covering multi-factor coupling effects. The design of the orthogonal matrix uses the adsorption energy threshold as the screening condition for corrosion activity, combines the molecular stability index corresponding to the bond-breaking mode to determine the concentration gradient range; the temperature gradient and aging time gradient are set based on the activation energy threshold of the copper surface interaction, and the acceleration effect of environmental stress on reaction kinetics is characterized by the temperature-time product factor. The generation of the experimental data set provides training samples for the mapping between the model input features and corrosion behavior by systematically covering the correlation between microscopic parameters and macroscopic conditions.

[0035] In step S3, a scanning electron microscope is used to observe the morphology of the coverage of cuprous sulfide on the surface of the copper conductor, and the change in the absorbance of sulfides in the oil is analyzed by ultraviolet-visible spectroscopy. The scanning electron microscope quantifies the coverage area and thickness of cuprous sulfide through backscattered electron imaging, and the ultraviolet-visible spectroscopy calculates the attenuation rate of the sulfide concentration in the oil through the intensity of the characteristic absorption peak. The linear relationship between the coverage and the change rate of absorbance is used to establish the correlation between the corrosion characteristics of the copper surface and the change in the oil components, and the extracted corrosion characteristic values are used as the measured data labels for model training and verification.

[0036] In step S4, the adsorption energy threshold in step S1, the Fermi level shift of the electron density of states distribution, and the temperature-time product factor in step S2 are used for eigenvector splicing to construct the input feature set of the random forest algorithm. The characteristic importance weights of the adsorption energy and environmental factors on the corrosion rate are analyzed through the Gini coefficient, and the non-linear correlation rules between the adsorption energy and the corrosion risk are identified. The decision tree splitting rule optimizes the node division threshold based on the characteristic importance and outputs the classification rule of the corrosion risk level and the regression curve of the cuprous sulfide generation rate. The dynamic prediction module of the model realizes the adaptive mapping between the microscopic parameters and the macroscopic corrosion rate through characteristic weight assignment and rule iteration.

[0037] In step S5, the concentration of sulfides in the transformer oil is monitored in real time by a fiber optic spectroscopy sensor, and the environmental data is collected by a temperature sensor. The time series alignment of multi-source data is realized by using the timestamp synchronization technology. After the synchronized data is input into the corrosion rate prediction model, the dynamic corrosion risk level and the cuprous sulfide generation rate curve are generated. When the risk level exceeds the preset threshold, an instruction for adjusting the oil circulation rate or adding a corrosion inhibitor is triggered, and the diffusion and adsorption processes of sulfides are inhibited by adjusting the oil parameters. At the same time, a warning signal is activated to indicate the maintenance requirements.

[0038] In step S6, the coverage of cuprous sulfide on the surface of the copper conductor is regularly detected by a scanning electron microscope, and the root mean square error is calculated between the measured coverage and the model prediction value. If the error exceeds the preset range, the experimental data samples under different solvation environments are expanded, and the characteristic weight assignment is optimized by retraining the decision tree splitting threshold of the random forest model. The adsorption energy correction parameters and bond-breaking characteristics corresponding to the new sulfide types are supplemented to the microscopic parameter database, and the molecular oxidation state classification index is updated to form a closed-loop feedback mechanism of experimental data and model parameters, improving the generalization ability and prediction accuracy of the model.

[0039] Specifically, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil according to the present invention, step S1 includes: Simulating the adsorption configuration of sulfide molecules on the copper surface by density functional theory, and calculating the initial adsorption energy and the electron density of states distribution; Based on molecular dynamics simulation of the diffusion path and solvation effect of sulfides in transformer oil, according to the energy barrier distribution of the diffusion path and the change in solvation free energy, the environmental impact parameters of the initial adsorption energy are corrected; Integrate the corrected adsorption energy, electron density of states distribution, and sulfide molecule bond-breaking characteristics into structured parameters, and correlate them with the sulfide type to form a dataset including molecular oxidation state, bond-breaking mode, and copper surface interaction strength.

[0040] In step S1, the adsorption configuration of sulfide molecules on the copper surface is simulated by density functional theory. The spatial arrangement between sulfide molecules and copper atoms is optimized using the first-principles calculation method to determine the most stable adsorption sites and molecular orientations. Based on quantum mechanics calculations, the initial adsorption energy and electron density of states distribution are obtained, where the adsorption energy characterizes the binding strength between sulfide molecules and the copper surface, and the electron density of states distribution reflects the degree of interaction between molecular orbitals and the electronic structure of the metal surface. This process reveals the microscopic adsorption mechanism of sulfide molecules on the copper conductor surface through quantum chemical calculations, providing a theoretical data basis for subsequent parameter correction.

[0041] Combined with molecular dynamics simulation of the diffusion behavior of sulfides in transformer oil, the movement trajectories and solvation effects of sulfide molecules in the oil medium are simulated. By analyzing the energy barrier distribution of the diffusion path, the activation energy required for sulfide molecules to cross different solvation layers is calculated; at the same time, the change in solvation free energy is evaluated to quantify the influence of the oil environment on the stability of sulfide molecules. Based on the diffusion dynamics and thermodynamic parameters, the environmental impact factor of the initial adsorption energy is corrected to make the adsorption energy data more consistent with the molecular interaction characteristics in the actual oil environment.

[0042] Integrate the corrected adsorption energy, electron density of states distribution, and sulfide molecule bond-breaking characteristics into structured parameters to construct a microscopic parameter database of sulfide molecules on the copper surface. The bond-breaking characteristics analyze the bond-breaking probability and bond-breaking sites of chemical bonds in sulfide molecules through molecular dynamics trajectories. Combining oxidation state classification and bond-breaking mode, a mapping relationship between sulfide type and copper surface interaction strength is established. The structured parameters in the database are indexed by sulfide molecule type, associating their oxidation state, bond-breaking mode, and adsorption energy correction value to form a multi-dimensional data system, providing microscopic mechanism support for multi-factor coupling experimental design and corrosion prediction model training.

[0043] Specifically, for the intelligent analysis method of the corrosiveness of sulfide molecules in transformer oil described in the present invention, step S2 includes: According to the adsorption energy threshold and bond-breaking characteristics stored in the microscopic parameter database, set the sulfide concentration gradient, temperature gradient, and aging time gradient in the orthogonal experimental matrix; The coverage of cuprous sulfide on the copper surface was observed by scanning electron microscopy, and the absorbance change of the oil was analyzed by ultraviolet and visible spectroscopy. The corrosion intensity was quantified according to the linear relationship between the coverage and the absorbance change rate; The product factor of the temperature gradient and the aging time gradient in the experimental conditions was correlated and mapped with the adsorption energy threshold and the electron density of states distribution in the microscopic parameter database to generate training samples including the product factor of temperature and time and the corrosion rate.

[0044] In step S2, based on the microscopic parameter database constructed in step S1, the adsorption energy threshold and the bond-breaking characteristics were used as the core basis for the orthogonal experimental design. The adsorption energy threshold was used to screen the critical value of the adsorption activity of sulfide molecules on the copper surface and determine the upper and lower limits of the sulfide concentration gradient; the bond-breaking characteristics were used to set the experimental conditions of the temperature gradient and the aging time gradient by analyzing the bond-breaking mode and stability of the chemical bonds in the sulfide molecules and combining the interaction strength on the copper surface. The design of the orthogonal experimental matrix covered the combined effects of sulfide concentration, temperature, and aging time through multi-factor coupling, aiming to systematically characterize the influence law of environmental variables on corrosion kinetics and provide multi-dimensional experimental data for subsequent model training.

[0045] The coverage of cuprous sulfide on the copper conductor surface was quantitatively analyzed by scanning electron microscopy. The microscopic morphology distribution of cuprous sulfide was obtained by backscattered electron imaging technology, and the coverage area and thickness were measured as direct indicators of the corrosion degree. At the same time, the absorbance change of sulfide in the transformer oil was analyzed by ultraviolet-visible spectroscopy, and the dynamic change of the sulfide concentration in the oil was calculated through the intensity attenuation rate of the characteristic absorption peak. The linear relationship between the coverage and the absorbance change rate was established by multiple regression analysis, and the morphology observation data was correlated with the change of the oil component to form a quantitative index of the corrosion intensity, providing calibrated experimental data labels for model training.

[0046] The temperature gradient and the aging time gradient in the orthogonal experiment were integrated into an environmental stress parameter through a product factor to characterize the synergistic effect of temperature and time on the corrosion reaction. This product factor was correlated and mapped with the adsorption energy threshold and the electron density of states distribution in the microscopic parameter database. The adsorption energy threshold reflected the binding strength between sulfide molecules and the copper surface, and the electron density of states distribution revealed the regulation mechanism of interfacial electron transfer on the corrosion process. Through feature fusion, multi-dimensional training samples including the product factor of temperature and time, the adsorption energy threshold, and the electron density of states characteristics were generated, and the mapping relationship between environmental variables and the corrosion rate was established, providing a matching data set of input features and output labels for the random forest model to support the prediction of the corrosion rate and the mining of classification rules.

[0047] Specifically, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil described in the present invention, step S4 includes: Concatenate the adsorption energy threshold and the Fermi level shift of the electronic state density distribution in the microparameter database with the temperature-time product factor in the multi-factor coupling experimental dataset as the input features of the random forest algorithm; Perform feature importance analysis by calculating the Gini coefficient, identify the non-linear correlation between the adsorption energy and the corrosion rate, and establish a corrosion risk level classification rule based on the decision tree splitting rule; Deploy the corrosion risk level classification rule as a dynamic prediction module of the corrosion rate prediction model, and output the prediction results including the corrosion risk level and the copper sulfide generation rate curve.

[0048] In step S4, based on the microparameter database constructed in step S1 and the multi-factor coupling experimental dataset generated in step S2, concatenate the adsorption energy threshold, the Fermi level shift of the electronic state density distribution, and the temperature-time product factor. The adsorption energy threshold characterizes the binding strength between the sulfide molecule and the copper surface, the Fermi level shift reflects the driving effect of interfacial electron transfer on the corrosion reaction, and the temperature-time product factor quantifies the acceleration effect of environmental stress on the reaction kinetics. The concatenation of the feature vectors eliminates the dimensional difference through data standardization, constructs a multi-dimensional input feature set, and provides training data that integrates the micro-mechanism and macroscopic conditions for the random forest algorithm.

[0049] Evaluate the feature importance weights of the adsorption energy, the Fermi level shift, and the temperature-time product factor on the corrosion rate through the Gini coefficient calculation in the random forest algorithm. The Gini coefficient analysis identifies the non-linear correlation law between the adsorption energy and the corrosion rate, such as the high-order interaction effect between the adsorption energy threshold and the corrosion rate. Based on the feature importance ranking, the decision tree splitting rule recursively divides the input feature space to generate the classification threshold of the corrosion risk level. For example, the combination of high adsorption energy, low Fermi level shift, and high temperature-time product factor is classified as a high-risk level, forming the logical judgment condition of the classification rule.

[0050] Deploy the corrosion risk level classification rule into the dynamic prediction module. After inputting the sulfide concentration and temperature data collected in real time, the module outputs the corresponding corrosion risk level by traversing the splitting conditions of the decision tree nodes. At the same time, based on the continuous prediction of the copper sulfide generation rate by the regression tree branch, a time-dependent corrosion rate curve is generated. The dynamic prediction module updates the decision tree weights through an online learning mechanism to adapt to the changes in the oil environment parameters, realizes the real-time classification of the corrosion risk level and the dynamic correction of the rate curve, and supports the timely adjustment of the protection strategy.

[0051] Specifically, for the intelligent analysis method of the corrosivity of sulfide molecules in transformer oil described in the present invention, step S5 includes: Monitoring the sulfide concentration through an optical fiber spectral sensor and achieving real-time synchronization with the temperature sensor data by aligning them through timestamps; Input the synchronized real-time data into the corrosion rate prediction model to generate corrosion risk levels and suggestions for protection strategies; When the corrosion risk level exceeds the preset threshold, adjust the oil parameters by adjusting the oil circulation rate or adding corrosion inhibitors, and trigger a warning signal.

[0052] In step S5, the characteristic absorption spectrum of sulfide in transformer oil is collected in real time through an optical fiber spectral sensor, and the sulfide concentration is analyzed based on the change in absorbance at a specific wavelength. The fiber optic sensor uses multi-channel spectral acquisition technology to detect the intensity of the characteristic peaks of sulfide in the oil through grating spectroscopy and a photodiode array, and calculates the concentration value by combining the calibration curve. The temperature sensor synchronously collects the oil temperature data, and uses timestamp alignment technology to match the concentration and temperature data in time sequence. For example, the acquisition delay is eliminated through time series interpolation or sliding window data fusion to achieve real-time synchronization of multi-source data.

[0053] After the synchronized sulfide concentration and temperature data are input into the corrosion rate prediction model, the model makes predictions based on the classification rules and regression tree structure trained in step S4. The input features include the real-time sulfide concentration, temperature value, and the product factor of temperature and time. By traversing the splitting conditions of the decision tree nodes to match the current environmental parameters, the corresponding corrosion risk level is output. The regression tree branches calculate the instantaneous value and cumulative trend of the copper sulfide formation rate based on the adsorption energy threshold, Fermi level shift, and the product factor of temperature and time, and generate a time-dependent corrosion rate curve. The protection strategy suggestions output by the model are adaptively generated based on the risk level. For example, a low risk level corresponds to a conventional monitoring mode, and a high risk level activates an active intervention command.

[0054] When the corrosion risk level exceeds the preset threshold, the system changes the diffusion flux of sulfide on the copper surface by adjusting the oil circulation rate. For example, the oil flow rate is increased by controlling the pump valve to reduce the local sulfide concentration. At the same time, the corrosion inhibitor automatic addition module releases a suitable corrosion inhibitor according to the sulfide type and concentration to inhibit the adsorption and bond-breaking reactions of sulfide molecules on the copper surface. The warning signal is triggered through an audible and visual alarm or a remote monitoring platform to prompt the operator to perform equipment maintenance or parameter review. The real-time data and operation records are synchronously stored in the database for subsequent model iteration optimization and closed-loop verification of the protection strategy, forming a technical link of data acquisition, prediction, response, and feedback.

[0055] Specifically, for the intelligent analysis method of the corrosiveness of sulfide molecules in transformer oil according to the present invention, step S6 includes: Regularly detect the coverage of cuprous sulfide on the surface of the copper conductor by scanning electron microscopy, and calculate the root mean square error between the measured coverage and the predicted value of the corrosion rate prediction model; When the error exceeds the preset range, increase the experimental data samples under different solvation environments, and optimize the feature weights by retraining the decision tree node splitting threshold in the random forest model; Supplement the adsorption energy correction parameters and bond-breaking characteristics corresponding to the newly added sulfide types to the microscopic parameter database, and update the molecular oxidation state classification index.

[0056] In step S6, periodically detect the coverage of cuprous sulfide on the surface of the copper conductor by scanning electron microscopy, use backscattered electron imaging technology to obtain microscopic morphology data of the coverage in different regions, and calculate the statistical average of the coverage area and thickness in combination with image processing algorithms as a quantitative index of the measured corrosion degree. The root mean square error formula is used to calculate the deviation between the measured coverage and the predicted value of the corrosion rate prediction model to evaluate the prediction accuracy of the model in a specific solvation environment. The error calculation result is used to judge whether the generalization ability of the model meets the preset threshold. If it exceeds the threshold range, it indicates that the training sample coverage of the current model is insufficient or there is a deviation in the environmental parameter correlation.

[0057] When the root mean square error exceeds the preset range, it is necessary to expand the experimental data samples to enhance the adaptability of the model. By designing orthogonal experiments under different solvation environments (such as different oil polarities and additive concentrations), simulate the diffusion and adsorption behaviors of sulfides under complex working conditions, and obtain the mapping relationship between the new temperature-time product factor and the corrosion rate. The newly added experimental data is optimized by the decision tree node splitting threshold in the random forest algorithm, and the weight distribution of features such as adsorption energy and Fermi level shift in the model is adjusted. For example, the importance of features is re-evaluated through the Gini coefficient, and the decision tree splitting rule is corrected to improve the prediction robustness of the model for multi-solvation environments.

[0058] The adsorption energy correction parameters corresponding to the newly added sulfide types are recalculated by molecular dynamics simulation and density functional theory, combined with the statistical analysis of bond-breaking characteristics, and the molecular oxidation state classification index in the microscopic parameter database is updated. The database update includes the bond-breaking mode of the newly added sulfide, the change in solvation free energy, and the corrected adsorption energy threshold, and is associated with the corresponding experimental data set and model parameters through the index. The updated database provides extended microscopic mechanism support for subsequent model training and prediction, forming a dynamic closed-loop optimization link of experimental data, model parameters, and database, and continuously improving the accuracy and adaptability of corrosion risk prediction.

[0059] Specifically, the intelligent analysis method for the corrosivity of sulfide molecules in transformer oil described in the present invention further includes: Transfer the electron density distribution characteristics of known sulfides to the corrosion rate prediction model through feature space mapping for predicting the corrosion rate of unknown sulfide types; Align the time stamp sequences of multi-source sensor data with the adsorption energy correction parameters obtained from molecular dynamics simulations in the time dimension; Establish a feedback mechanism between the prediction results and the experimental design. According to the residual analysis results of the corrosion rate prediction model, dynamically adjust the temperature gradient setting of subsequent multi-factor coupling experiments, and update the new experimental data back to the microscopic parameter database.

[0060] Project the electron density distribution characteristics of known sulfide molecules into a unified high-dimensional feature space through feature space mapping technology, and use transfer learning algorithms to extract the common laws of the interaction between sulfide molecular orbitals and metal surfaces. Based on the electron density distribution data of known sulfides, construct a feature mapping model, map the electron density distribution of unknown sulfides to the same feature space through kernel function or neural network dimensionality reduction methods, and use similarity metrics to match the corrosion behavior patterns of known sulfides to achieve the prediction of the corrosion rate of unknown sulfide types. This mapping process maintains the correlation between the microscopic mechanism and the macroscopic corrosion behavior by retaining the peak position and energy difference characteristics of the electron density distribution.

[0061] Align the sulfide concentration, temperature, and oil flow rate data collected by multi-source sensors through time stamp sequences. Use a sliding window mechanism to slice the sensor data stream in the time dimension and perform time series matching with the adsorption energy correction parameters generated by molecular dynamics simulations. The adsorption energy correction parameters are dynamically updated according to the simulation results of the solvation free energy change and the diffusion path energy barrier. Eliminate the time series deviation between the experimental data and the simulation parameters through time stamp alignment, for example, interpolate the simulation parameters to the sensor data acquisition points at time steps to ensure the consistency of the feature vectors input to the model in the time dimension. The time-aligned dataset supports the real-time prediction of the corrosion rate in a dynamic environment by the model.

[0062] Based on the residual analysis results of the corrosion rate prediction model, establish a feedback correlation between the prediction error and the experimental condition setting. Residual analysis calculates the deviation distribution between the predicted value and the measured coverage to identify the sensitivity of the temperature gradient setting to the model error. According to the residual distribution characteristics, dynamically adjust the temperature gradient range of subsequent multi-factor coupling experiments, for example, increase the temperature sampling point density in high-error regions, and optimize the experimental design to cover the weak links of the model prediction. Recalculate the adsorption energy correction parameters for the new experimental data through molecular dynamics simulations, supplement them to the microscopic parameter database, and update the sulfide molecular oxidation state classification index and the bond breakage mode mapping table to form a closed-loop optimization link of experimental design - model prediction - data update, improving the model's adaptability to complex oil environments.

[0063] Explanation of the technical features in the technical solution of the present invention is as follows: Density Functional Theory (DFT): A computational method based on quantum mechanics, used to simulate the adsorption configuration of sulfide molecules on the copper surface, calculate the initial adsorption energy and the electron density of states distribution between the molecule and the metal interface. The adsorption energy characterizes the binding strength between the sulfide molecule and the copper surface, and the electron density of states distribution reflects the driving effect of interfacial electron transfer on the corrosion reaction.

[0064] Molecular Dynamics Simulation (MD): By simulating the movement trajectory and solvation effect of sulfide molecules in transformer oil, analyzing the energy barrier distribution of the diffusion path and the change of solvation free energy, and correcting the environmental influence parameters of the initial adsorption energy to make it more conform to the molecular interaction characteristics in the actual oil environment.

[0065] Adsorption energy threshold: The lowest energy critical value for significant adsorption of sulfide molecules on the copper surface, used to screen the corrosion activity of different sulfides. For example, sulfides with adsorption energy higher than the threshold (such as DBDS) have stronger corrosivity.

[0066] Fermi level shift of the electron density of states distribution: Characterizes the change in the energy position of the Fermi level on the copper surface caused by sulfide adsorption, reflects the regulation mechanism of interfacial electron transfer on the corrosion reaction, and the larger the shift, the stronger the corrosion driving force.

[0067] Orthogonal experiment matrix: A multi-factor experimental design method. By setting sulfide concentration gradients (such as 0.1 - 1.0 ppm), temperature gradients (such as 80 - 150 °C), and aging time gradients (such as 24 - 72 hours), an experimental data set covering the interaction of multiple variables is generated for training the corrosion rate prediction model.

[0068] Temperature-time product factor: Integrates temperature and aging time into a single environmental stress parameter to quantify the synergistic acceleration effect of the two on the corrosion reaction kinetics. For example, the product factor of 150 °C and 72 hours is used to characterize the corrosion risk under high temperature and long time conditions.

[0069] Observation of the coverage of cuprous sulfide by Scanning Electron Microscope (SEM): Measures the coverage area and thickness of cuprous sulfide on the copper surface through backscattered electron imaging technology, as a direct quantitative index of the corrosion degree. For example, when the coverage exceeds 30%, it is determined as a high-risk level.

[0070] Ultraviolet-Visible spectroscopy analysis of the absorbance of the oil: Based on the change in the intensity of the characteristic absorption peak of sulfide at a specific wavelength (such as 280 nm), calculates the attenuation rate of the sulfide concentration in the oil. The linear relationship between the absorbance change rate and the coverage is used to calibrate the corrosion intensity.

[0071] Random Forest Algorithm: An ensemble learning model that conducts feature fusion and splitting rule training on the adsorption energy threshold, Fermi level shift, and temperature-time product factor through multiple decision trees, and outputs the classification of corrosion risk levels and the regression prediction of the copper sulfide generation rate.

[0072] Gini Coefficient Analysis: Used to evaluate the classification contribution of input features (such as adsorption energy, temperature-time product factor) to the corrosion rate, and identify the non-linear association rules between adsorption energy and corrosion risk. For example, the adsorption energy threshold in the range of 1.2 - 1.8 eV has the greatest contribution to classification.

[0073] Real-time Monitoring of Fiber Optic Spectral Sensor: Analyzes the characteristic absorption spectra of sulfides in oil by multi-channel spectral acquisition technology, and calculates the sulfide concentration in real-time in combination with the calibration curve. For example, when the detected concentration exceeds 0.5 ppm, a warning signal is triggered.

[0074] Closed-loop Feedback Optimization: Regularly detects the copper surface coverage by scanning electron microscopy, calculates the model prediction error (such as the root mean square error exceeds 5%), expands the experimental data under different solvation environments, retrains the model and updates the microscopic parameter database. For example, new sulfide types are added to map the electronic state density characteristics through transfer learning, improving the generalization ability of the model.

[0075] Density Functional Theory (DFT) Model: Density functional theory is a computational model based on quantum mechanics, used to simulate the adsorption configuration of sulfide molecules on the copper surface, and calculate the initial adsorption energy and the distribution of electronic state density between the molecule and the metal interface. In the present invention, DFT optimizes the spatial arrangement of sulfide molecules on the copper surface through first-principles calculations, and determines the most stable adsorption sites and orientations. The adsorption energy value characterizes the binding strength between the sulfide molecule and the copper surface, while the distribution of electronic state density reveals the interaction between the molecular orbit and the electronic structure of the metal surface, such as the hybridization degree between the p orbit of the sulfur atom and the d orbit of the copper, providing a theoretical basis for the construction of the microscopic parameter database.

[0076] Molecular Dynamics (MD) Simulation Model: Molecular dynamics simulation is used to analyze the diffusion behavior and solvation effect of sulfides in transformer oil. By simulating the movement trajectory of molecules in the oil, the energy barrier distribution of the diffusion path (such as the activation energy required to cross the solvation layer) and the change in solvation free energy are calculated, quantifying the correction effect of the oil environment on the adsorption energy. For example, a highly polar oil will increase the solvation energy of sulfides, reducing the probability of their diffusion to the copper surface, thereby correcting the value of the initial adsorption energy to make it more in line with the actual working conditions.

[0077] Orthogonal experimental design method: Orthogonal experimental design is a multi-factor coupling experimental planning method. By setting gradient combinations of sulfide concentration, temperature, and aging time (such as concentration gradient 0.1 - 1.0 ppm, temperature gradient 80 - 150 °C, time gradient 24 - 72 hours), it systematically covers the corrosion behavior under the interaction of multiple variables. The design of the experimental matrix is based on the adsorption energy threshold and bond-breaking characteristics in the microscopic parameter database. For example, sulfides with an adsorption energy higher than 1.5 eV need to set a higher temperature gradient to simulate the accelerated corrosion process, thus generating a multi-dimensional dataset for machine learning model training.

[0078] Random forest algorithm: Random forest is an ensemble learning algorithm. By constructing multiple decision trees to train the splitting rules for input features (adsorption energy threshold, Fermi level shift, temperature-time product factor), it outputs the classification of corrosion risk levels and the regression prediction of the formation rate of cuprous sulfide. The algorithm uses the Gini coefficient to evaluate the importance of features. For example, the adsorption energy threshold has the highest contribution to corrosion risk. The decision tree optimizes the node splitting threshold according to the feature importance to form classification rules (such as when the adsorption energy ≥ 1.5 eV and the temperature-time product factor ≥ 10^4, it is determined as high risk). The dynamic prediction module of the model updates the decision tree weights through an online learning mechanism to adapt to the changes in oil fluid environment parameters.

[0079] Transfer learning technology: Transfer learning is used to map the electronic state density distribution characteristics of known sulfide types to unknown sulfides. By constructing a high-dimensional feature space (such as kernel function or neural network dimensionality reduction), it extracts the common laws of the interaction between sulfide molecular orbitals and the copper surface. For example, the peak position and energy difference of the electronic state density of sulfur atoms. The electronic state density distribution of unknown sulfides is matched with known types through similarity metrics (such as cosine similarity) to achieve cross-type prediction of corrosion behavior and reduce the dependence on experimental data of unknown sulfides.

[0080] Closed-loop feedback optimization mechanism: This mechanism is based on the error analysis of the measured coverage by scanning electron microscopy and the model prediction value (such as root mean square error calculation), and dynamically adjusts the experimental design and model parameters. If the error exceeds the preset threshold (such as 5%), then expand the experimental data under different solvation environments (such as changes in additive concentration), retrain the decision tree splitting threshold of the random forest model, and optimize the feature weight allocation. The adsorption energy correction parameters of new sulfide types are updated to the microscopic database through molecular dynamics simulation, forming a closed-loop link of "data collection - model prediction - feedback update" to improve the generalization ability of the model.

[0081] Based on the problems of lag and experience dependence in the corrosion assessment of transformer oil sulfides in the background technology, combined with the actual application scenarios of power equipment, through the dynamic correlation between microscopic parameters and macroscopic behaviors and the coupling analysis of multiple factors, the intelligent prediction and active protection of the corrosion rate are realized. In the implementation process, first, the adsorption configuration of sulfide molecules on the copper surface is simulated by density functional theory, and the initial adsorption energy and electron density of states distribution are calculated. For example, the adsorption energy of dibenzyl disulfide (DBDS) forming a stable S-Cu bond on the copper surface is significantly higher than that of diphenyl sulfone (DPS). Combining molecular dynamics simulation of the diffusion path of sulfides in the oil and the change of solvation free energy, the environmental influence parameters of the initial adsorption energy are corrected. For example, the calculation of the activation energy required for sulfides to cross the solvation layer in the oil quantifies the correction amplitude of the oil polarity on the adsorption energy threshold. The corrected adsorption energy, electron density of states distribution, and bond-breaking characteristics are integrated into a structured database to correlate the interaction strength between the sulfide type and the copper surface, supporting subsequent experimental design and model training.

[0082] Based on the adsorption energy threshold and bond-breaking characteristics in the microscopic parameter database, an orthogonal experimental matrix is designed to cover the combined effects of sulfide concentration, temperature, and aging time. For example, when the adsorption energy threshold is higher than the preset critical value, the sulfide concentration gradient is set to 0.1 - 1.0 ppm, the temperature gradient is 80 - 150 °C, and the aging time gradient is 24 - 72 hours to generate a multi-factor coupling experimental data set. The coverage of cuprous sulfide on the copper surface is observed by scanning electron microscopy, and the change in the absorbance of the oil is analyzed by ultraviolet-visible spectroscopy to establish a linear regression model of the coverage and the absorbance decay rate to quantify the corrosion intensity. The product factor of temperature and time in the experimental conditions is feature-fused with the adsorption energy threshold and Fermi level shift in the microscopic parameter database to generate training samples for the input random forest model. For example, the combination of a high temperature (150 °C) and a long time (72 hours) product factor and a high adsorption energy (≥1.5 eV) corresponds to a high-risk corrosion level.

[0083] The adsorption energy threshold, Fermi level shift, and temperature-time product factor are concatenated into a feature vector and input into a random forest algorithm to train a corrosion rate prediction model. The non-linear effect of adsorption energy on corrosion rate is identified through Gini coefficient analysis. For example, the classification contribution of the adsorption energy threshold in the range of 1.2 - 1.8 eV to the corrosion risk level is the highest. After the model is deployed, the concentration and temperature data of sulfides in the oil are collected in real time. Through the timestamp alignment technology of the fiber optic spectroscopy sensor and the temperature sensor, the data is input into the model to generate a dynamic corrosion risk level and a copper sulfide formation rate curve. When the risk level exceeds the threshold, the oil circulation rate is adjusted to more than 2.5 m / s or 0.05% corrosion inhibitor is added to inhibit sulfide adsorption. The coverage is regularly detected by scanning electron microscopy, and the model prediction error is calculated. If the root mean square error exceeds 5%, the experimental data under different solvation environments is expanded, the model is retrained, and the microscopic parameter database is updated. For example, new sulfide types are added to map the electronic state density characteristics through transfer learning to predict the corrosion behavior of unknown sulfides. The above implementation process forms a closed-loop technical link of experimental design - model prediction - feedback optimization, solving the problem of real-time analysis of multi-factor coupled corrosion kinetics.

[0084] The present invention solves the deficiencies of the existing sulfide corrosion evaluation method through the following technical solutions: First, a microscopic parameter database is constructed based on density functional theory and molecular dynamics simulation to quantify the adsorption energy, electron state density distribution, and bond-breaking characteristics of sulfide molecules on the copper surface, and establish a correlation model between the microscopic adsorption mechanism and the macroscopic corrosion behavior. Through the orthogonal experimental matrix, multi-factor coupled experiments are designed to cover the combined effects of sulfide concentration, temperature, and aging time, generating a training data set containing microscopic parameters and macroscopic environmental variables. The integration of the microscopic parameter database and experimental data provides input features for the machine learning model, dynamically analyzing the coupling effect of microscopic indicators such as adsorption energy threshold and Fermi level shift and macroscopic environmental parameters such as temperature-time product factor, breaking through the dependence of the existing method on a single experimental condition.

[0085] Second, a random forest algorithm is used to train a corrosion rate prediction model. The importance of features is analyzed through Gini coefficient to identify the non-linear correlation rule between adsorption energy and corrosion rate. The model concatenates microscopic parameters and macroscopic experimental data into a feature vector, establishes a decision tree splitting rule, and outputs a corrosion risk level and a copper sulfide formation rate curve. The concentration and temperature data of sulfides are collected in real time and synchronously input into the model through timestamp alignment technology to achieve dynamic prediction and immediate response of the protection strategy. The model updates the feature weights through an online learning mechanism to adapt to the changes in the oil environment, solving the deficiencies of lag and experience dependence of the existing evaluation methods.

[0086] Finally, a closed-loop feedback optimization mechanism is established. The corrosion coverage is regularly detected by scanning electron microscopy, and the prediction error of the calculation model is calculated. When the error exceeds the limit, the experimental data under different solvation environments are extended and the model is retrained to optimize the decision tree node splitting threshold. The adsorption energy correction parameters of the new sulfide type and the broken bond characteristics are supplemented to the microscopic database, and the oxidation state classification index is updated. The electron density of states characteristics of known sulfides are mapped to unknown types through transfer learning, and the experimental design is dynamically adjusted in combination with residual analysis to form a closed-loop link of "data acquisition - model prediction - feedback update", improving the real-time prediction accuracy and adaptability of multi-factor coupled corrosion kinetics.

Claims

1. An intelligent analysis method for the corrosivity of sulfide molecules in transformer oil, characterized in that, Including: Step S1: Obtain the adsorption energy, electron density of states distribution, and bond-breaking characteristics of sulfide molecules on the copper surface. Based on the adsorption energy, electron density of states distribution, and bond-breaking characteristics of sulfide molecules on the copper surface, construct a microscopic parameter database of sulfide molecules on the copper conductor surface. The microscopic parameter database includes the oxidation state of sulfide molecules, the bond-breaking mode, and the copper surface interaction strength. Step S2: According to the adsorption energy threshold and bond-breaking characteristics in the microscopic parameter database, set the sulfide concentration gradient, temperature gradient, and aging time gradient of the orthogonal experimental matrix to generate a multi-factor coupling experimental dataset. Step S3: Observe the coverage of cuprous sulfide on the copper surface by scanning electron microscopy, and analyze the change in oil absorbance by combining ultraviolet and visible spectra to extract the corrosion characteristics of the copper conductor surface and the oil absorbance value. Step S4: Based on the adsorption energy threshold and electron density of states distribution in the microscopic parameter database and the temperature-time product factor in the multi-factor coupling experimental dataset, train a corrosion rate prediction model through a random forest algorithm. The corrosion rate prediction model takes the adsorption energy threshold, Fermi level shift, and environmental factor as input variables and outputs the corrosion risk level. Step S5: Real-time collect the sulfide concentration and temperature data in the transformer oil, and input them into the corrosion rate prediction model to generate a dynamic corrosion risk level and a cuprous sulfide generation rate curve. Step S6: Trigger the optimization of the protection strategy according to the corrosion risk level, and regularly detect the coverage of cuprous sulfide on the copper conductor surface by scanning electron microscopy, and calculate the error with the predicted value of the corrosion rate prediction model. If the error exceeds the preset range, expand the multi-factor coupling experimental data and update the parameter weights and microscopic parameter database of the model.

2. The intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to claim 1, wherein The step S1 includes: Simulate the adsorption configuration of sulfide molecules on the copper surface by density functional theory to calculate the initial adsorption energy and electron density of states distribution. Based on molecular dynamics, simulate the diffusion path and solvation effect of sulfide in transformer oil, and correct the environmental influence parameters of the initial adsorption energy according to the energy barrier distribution and solvation free energy change of the diffusion path. Integrate the corrected adsorption energy, electron density of states distribution, and sulfide molecule bond-breaking characteristics into structured parameters and associate them with sulfide types to form a dataset including molecular oxidation state, bond-breaking mode, and copper surface interaction strength.

3. The intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to claim 1, characterized in that The step S2 includes: According to the adsorption energy threshold and bond-breaking characteristics stored in the microscopic parameter database, set the sulfide concentration gradient, temperature gradient, and aging time gradient in the orthogonal experimental matrix. Observe the coverage of cuprous sulfide on the copper surface by scanning electron microscopy, and analyze the change in oil absorbance by ultraviolet and visible spectra, and quantify the corrosion intensity according to the linear relationship between the coverage and the absorbance change rate. Map the product factor of the temperature gradient and aging time gradient in the experimental conditions to the adsorption energy threshold and electron density of states distribution in the microscopic parameter database to generate a training sample including the temperature-time product factor and the corrosion rate.

4. The intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to claim 1, characterized in that, The step S4 includes: Concatenate the adsorption energy threshold in the microscopic parameter database and the Fermi level offset of the electron density of states distribution with the temperature-time product factor in the multi-factor coupling experimental dataset as the input features of the random forest algorithm; Conduct feature importance analysis by calculating the Gini coefficient, identify the non-linear correlation between the adsorption energy and the corrosion rate, and establish a corrosion risk level classification rule based on the decision tree splitting rule; Deploy the corrosion risk level classification rule as a dynamic prediction module of the corrosion rate prediction model, and output a prediction result including the corrosion risk level and the copper sulfide generation rate curve.

5. The intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to claim 1, characterized in that, The step S5 includes: Monitor the sulfide concentration through an optical fiber spectral sensor and achieve real-time synchronization with the temperature sensor data through timestamp alignment; Input the synchronized real-time data into the corrosion rate prediction model to generate the corrosion risk level and protection strategy suggestions; When the corrosion risk level exceeds the preset threshold, adjust the oil parameters by adjusting the oil circulation rate or adding corrosion inhibitors, and trigger a warning signal.

6. The intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to claim 1, wherein The step S6 includes: Regularly detect the copper sulfide coverage on the surface of the copper conductor through a scanning electron microscope, and calculate the root mean square error between the measured coverage and the predicted value of the corrosion rate prediction model; When the error exceeds the preset range, increase the experimental data samples under different solvation environments, and optimize the feature weights by retraining the decision tree node splitting threshold in the random forest model; Supplement the adsorption energy correction parameters and bond-breaking characteristics corresponding to the new sulfide type to the microscopic parameter database, and update the molecular oxidation state classification index.

7. The intelligent analysis method for the corrosivity of sulfide molecules in transformer oil according to claim 6, characterized in that, It also includes: Migrate the electron density of states distribution characteristics of known sulfides to the corrosion rate prediction model through feature space mapping for predicting the corrosion rate of unknown sulfide types; Align the timestamp sequence of multi-source sensor data with the adsorption energy correction parameters obtained from molecular dynamics simulation in the time dimension; Establish a feedback mechanism between the prediction result and the experimental design. According to the residual analysis result of the corrosion rate prediction model, dynamically adjust the temperature gradient setting of the subsequent multi-factor coupling experiment, and update the new experimental data back to the microscopic parameter database.

Images