Method for solving signal drift of oxygen sensor in high-temperature and high-pressure environment

Abstract

The invention provides a method for solving signal drift of an oxygen sensor in a high-temperature and high-pressure environment, which comprises the following steps: collecting acid-base gas concentration data and zirconium oxide material surface state information in the high-temperature and high-pressure environment, and classifying erosion sensitivity degrees by adopting a support vector machine algorithm to obtain an initial response characteristic deviation value; a neural network model is adopted to process the determined potential signal drift trend and oxygen content detection historical data, a simulation response characteristic curve is generated, and a quantitative mapping relation of the pH value change to erosion sensitivity is obtained; refining the correlation between the determined long-term monitoring reliability index and the pH value change by adopting a neural network model, generating a dynamic response characteristic prediction sequence, and obtaining a signal drift suppression threshold set; and through fusion of the obtained signal drift suppression threshold set and the oxygen content detection real-time signal, if it is judged that erosion sensitivity is aggravated in a prediction sequence, a compensation model is activated to update and circulate, and optimized sensor stable output is obtained.

Description

Technical Field

[0001] This invention relates to the field of oxygen sensor technology, and in particular to a solution for oxygen sensor signal drift under high temperature and high pressure conditions. Background Technology

[0002] Accurate oxygen content detection in extreme high-temperature and high-pressure environments is a core requirement for the safe operation and efficient control of critical equipment such as industrial furnaces, chemical reactors, and aero-engine combustion chambers. This field directly relates to energy efficiency, emission control, and equipment lifespan. Oxygen sensors need to operate stably for extended periods under harsh conditions of temperatures exceeding 1,000 degrees Celsius and pressures exceeding 10 MPa to provide real-time and reliable oxygen concentration signals, thereby supporting process optimization and fault early warning.

[0003] Current oxygen sensors primarily rely on the oxygen ion conductivity of materials such as zirconia ceramics at high temperatures for detection. However, in complex real-world operating conditions, sensors are often exposed to flue gas or reaction media containing various corrosive gases. These methods perform reasonably well in relatively clean or pH-stable environments, but once the pH of the medium changes, the sensor signal is prone to unpredictable shifts, leading to distorted monitoring results.

[0004] Material tolerance has become a fundamental bottleneck restricting sensor reliability. Zirconia ceramics and their electrodes are extremely sensitive to corrosion by acidic gases such as sulfur dioxide and nitrogen oxides, or alkaline gases such as ammonia, under extreme high temperature and pressure. These gases undergo chemical adsorption and reaction on the material surface or interface, gradually altering the ion conduction characteristics of the sensitive layer and the electrochemical activity of the electrodes. The corrosion process further exacerbates signal drift, causing the correspondence between the sensor output and the actual oxygen concentration to gradually deviate from its original pattern. This drift is particularly difficult to compensate for through conventional calibration under conditions of frequent fluctuations in pH.

[0005] Therefore, accurately understanding the changes in the response characteristics of oxygen sensor sensitive materials and electrodes after being corroded by acid and alkaline media in a corrosive environment with high temperature, high pressure and accompanying acid-base changes, and revealing the degree of influence of acid-base on signal drift, has become a key issue for achieving long-term stable monitoring of oxygen content. Summary of the Invention

[0006] This invention provides a solution to oxygen sensor signal drift under high temperature and high pressure conditions, mainly including: By collecting acid and alkali gas concentration data and zirconium oxide material surface condition information in a high temperature and high pressure environment, the corrosion sensitivity is classified using a support vector machine algorithm to obtain the initial response characteristic deviation value. Based on the obtained initial response characteristic deviation value, the corresponding pH change record is obtained, and the signal drift feature vector is extracted from it. If the feature vector exceeds the preset threshold, it is marked as a high-corrosion environment influence area to determine the potential signal drift trend. A neural network model was used to process the identified potential signal drift trend and historical oxygen content detection data to generate simulated response characteristic curves and obtain a quantitative mapping relationship between pH changes and erosion sensitivity. By analyzing the obtained quantitative mapping relationship, and integrating the real-time monitoring input of acid and alkali gases to meet the sensor stability requirements, the support vector machine algorithm is used to optimize the response characteristic calibration parameters, resulting in an adjusted signal drift compensation model. Based on the adjusted signal drift compensation model, experimental sample data of zirconia materials under extreme conditions are obtained. If the matching degree between the compensation model output and the actual response characteristics is higher than the threshold, the corrosion environment adaptability is confirmed and the long-term monitoring reliability index is determined. A neural network model was used to refine the correlation between the long-term monitoring reliability indicators and pH changes, generate a dynamic response characteristic prediction sequence, and obtain a set of signal drift suppression thresholds. By fusing the obtained signal drift suppression threshold set with the real-time oxygen content detection signal, if the predicted sequence shows increased erosion sensitivity, the compensation model update cycle is activated to obtain the optimized stable sensor output.

[0007] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a comprehensive solution to the signal drift problem of oxygen sensors under high temperature and high pressure environments, aiming to address the core business scenario issue of the impact of changes in acid and alkaline gas concentrations on sensor corrosion sensitivity and long-term monitoring reliability. This invention collects environmental data and material surface conditions, uses a support vector machine algorithm to classify corrosion sensitivity, combines a neural network model to analyze the quantitative mapping relationship between signal drift trends and pH changes, optimizes response characteristic calibration parameters, and establishes a signal drift compensation model. After verifying the model's matching degree under extreme conditions, reliability indicators are refined, a dynamic prediction sequence is generated, and the compensation model update cycle is activated based on the intensification of corrosion sensitivity, ultimately achieving stable sensor output. The most prominent technical effect of this invention lies in its effective suppression of signal drift through multi-algorithm fusion and dynamic adjustment mechanisms, improving the sensor's adaptability and long-term stability in corrosive environments, and providing a reliable guarantee for accurate monitoring under high temperature and high pressure environments. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a solution to oxygen sensor signal drift under high temperature and high pressure conditions according to the present invention.

[0009] Figure 2This is a schematic diagram of a solution to the oxygen sensor signal drift problem under high temperature and high pressure conditions according to the present invention.

[0010] Figure 3 This is another schematic diagram of a solution to the oxygen sensor signal drift problem under high temperature and high pressure conditions according to the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0012] like Figures 1-3 This embodiment of a solution to oxygen sensor signal drift under high temperature and high pressure conditions may specifically include: Step S101: By collecting acid and alkali gas concentration data and zirconium oxide material surface state information in a high temperature and high pressure environment, the corrosion sensitivity is classified using a support vector machine algorithm to obtain the initial response characteristic deviation value.

[0013] Real-time data on acid and alkali gas concentrations and the surface condition of zirconia materials are collected using sensors in a high-temperature, high-pressure environment. Data preprocessing is performed on the collected acid and alkali gas concentration data and surface condition information to remove outliers and standardize the data. A support vector machine algorithm is then used to classify the preprocessed concentration data and surface condition information to determine the corrosion sensitivity level; the specific calculation formula is as follows: ; in, This represents the erosion sensitivity level of the i-th data point. This represents the weight coefficient of the j-th feature in the support vector machine algorithm. The j-th feature value of the i-th data point includes acid and alkali gas concentration and surface condition information, b represents the bias term, and m represents the total number of features; this formula calculates the erosion sensitivity of each sampling point using a support vector machine classification algorithm.

[0014] If the corrosion sensitivity level is higher than a preset threshold, the corresponding data point is marked as a highly sensitive area. The deviation value of the response characteristics of each region on the zirconia material surface is calculated based on the results of marking these highly sensitive areas; the specific calculation formula is as follows: ; in, This represents the deviation value of the response characteristics of the k-th region on the surface of the zirconia material. This represents the actual response value of the i-th measurement point in the k-th region. This represents the standard response value of the k-th region, and n represents the total number of highly sensitive markers in that region. This formula is used to quantify the degree of response deviation of each region on the material surface relative to the standard state.

[0015] Based on the distribution of response characteristic deviation values, the surface state changes of the material corresponding to the concentrated areas of deviation values ​​are identified. A linear regression algorithm is used to analyze the correlation between the surface state changes in the concentrated areas of deviation values ​​and the concentration of acid and alkali gases, obtaining the deviation prediction coefficient; the specific calculation formula is as follows: ; Indicates the deviation prediction coefficient. This represents the acid / base gas concentration value corresponding to the i-th deviation concentration region. This represents the average concentration of acid and alkali gases across all regions with concentrated deviations. This represents the change in surface state of the i-th deviation concentration region. denoted by , represents the average value of surface state changes across all areas of concentrated deviation, where p represents the total number of such areas. This formula establishes the correlation between surface state changes and gas concentration through linear regression.

[0016] The process of collecting acid and alkali gas concentration data and zirconia material surface state information in a high-temperature and high-pressure environment, and then using a support vector machine algorithm to classify the corrosion sensitivity to obtain initial response characteristic deviation values, can be implemented through the following specific method. First, a high-precision gas sensor is used to monitor the acid and alkali gas concentrations in real time at an environment of 800 degrees Celsius and 10 MPa. For example, the hydrogen sulfide concentration is measured to be 50 ppm and the ammonia concentration to be 30 ppm. These values ​​are stored in a database at a frequency of once per minute using a data acquisition system. Simultaneously, high-resolution microscopy imaging technology combined with image processing algorithms is used to extract the microscopic morphological features of the zirconia material surface, such as a surface roughness of 2.5 micrometers and a crack density of 0.3 cracks per square millimeter. These data are then associated with the gas concentration data to form a multidimensional dataset. Next, the collected data is preprocessed. A normalization method is used to map the gas concentration and surface state parameters to a range of 0 to 1. For example, a hydrogen sulfide concentration of 50 ppm is normalized to 0.5. Principal component analysis is then used to reduce dimensionality and extract key features to reduce computational complexity and ensure classification accuracy. Subsequently, a classification model was constructed based on the Support Vector Machine (SVM) algorithm, using a radial basis function kernel. The penalty parameter C was set to 10, and the kernel parameter γ to 0.1. The model parameters were optimized using cross-validation. The data was divided into training and test sets in an 8:2 ratio. After training, the model achieved a classification accuracy of 92.5% for erosion sensitivity, categorized into high, medium, and low sensitivity with corresponding bias ranges of ±5%, ±3%, and ±1%, respectively. Finally, the initial response characteristic bias was calculated based on the classification results. For example, a sample classified as high sensitivity had a bias of 4.8%. Comparison with historical data revealed a positive correlation between the bias and hydrogen sulfide concentration, with a correlation coefficient of 0.85, providing data support for subsequent material optimization. Through this method, data processing and algorithm application at each stage form a closed-loop logic, ensuring automation and scientific rigor throughout the entire process from data acquisition to bias calculation.

[0017] Step S102: Based on the obtained initial response characteristic deviation value, acquire the corresponding pH change record, extract the signal drift feature vector from it, and determine if the feature vector exceeds a preset threshold. If so, mark it as a high-corrosion environment influence zone to determine the potential signal drift trend. The specific calculation formula is as follows: ; This is the formula for calculating the signal drift eigenvector; This represents the signal drift characteristic vector value within the time period t, where T represents the total number of sampling time points. This represents the pH response signal value at time point i. This represents the initial reference response value. Indicates the standard deviation of the signal. This represents the weight coefficient at the i-th time point; when When the threshold is exceeded, it is marked as a high-corrosion environment influence zone.

[0018] Based on the signal drift trend direction, corresponding environmental parameter fluctuation data are acquired. The fluctuation data is then processed in time series segments to obtain the fluctuation amplitude distribution within each time period. Using the fluctuation amplitude distribution, a support vector machine algorithm is employed to classify the environmental parameter fluctuation data within each time period, identifying high-risk time intervals. Based on these high-risk time intervals, corresponding acid-base change details are extracted from the data records, and the detailed data is smoothed to obtain a stable trend curve. Through this stable trend curve, environmental impact factors related to highly corrosive areas are acquired. If these environmental impact factors exceed a preset threshold, the area is marked as a key monitoring target. Based on these key monitoring targets, historical records of their response deviations are acquired, and stratified analysis is performed on these historical records to determine potential patterns of deviation accumulation. Through these potential patterns of deviation accumulation, long-term change monitoring data related to initial characteristics is acquired. If the long-term change monitoring data deviates from a preset range, the area is marked as an abnormal state region.

[0019] Based on an initial response characteristic deviation of 4.8%, the system automatically retrieves pH change records for the corresponding time period from the sensor's historical database. For example, it extracts a continuous monitoring sequence showing pH fluctuations from 6.2 to 4.7 over the past 72 hours, and simultaneously loads environmental parameters at a temperature of 900 degrees Celsius and a pressure of 12 MPa to form a time-series dataset. Subsequently, the pH signal is decomposed into multiple scales using a wavelet transform algorithm. A Daubechies wavelet basis function of order 4 is selected, and the decomposition level is set to 6 levels. Low-frequency approximation coefficients and high-frequency detail coefficients are extracted as signal drift feature vectors. For example, the low-frequency component vector with an energy proportion of 68% has a dimension of 128, and the standard deviation of the high-frequency noise component is 0.42. Then, the Euclidean norm of the eigenvector was calculated to be 3.65 and compared with a preset threshold of 3.2. Since the norm exceeded the threshold, the algorithm automatically marked this environmental segment as a high-corrosion environment influence zone. Simultaneously, a linear regression model was used to fit the drift trend. Inputting the eigenvector and timestamp, the slope coefficient was 0.027 per hour, the intercept was 1.15, and the coefficient of determination R² was 0.91, indicating a significant positive acceleration trend in signal drift. Further verification was performed using historical high-corrosion zone records and a K-means algorithm with a cluster size of 3. It was confirmed that the sample fell within 0.18 of the high-influence cluster center, thus determining that the potential signal drift could increase to 6.2% in the following 24 hours. This provides accurate trend prediction data for the sensor calibration module, supporting fully automated risk assessment and early warning.

[0020] Step S103: A neural network model is used to process the determined potential signal drift trend and historical oxygen content detection data to generate a simulated response characteristic curve, obtaining a quantitative mapping relationship between pH changes and erosion sensitivity. The calculation method is as follows: ; Where R(pH) represents the simulated response characteristic curve value under specific acidity / alkalinity conditions, and K represents the number of feature layers extracted by the neural network. This represents the weight coefficient of the feature at layer k. Describes the feature function of the k-th layer. This represents oxygen content detection data, where ΔS represents the signal drift trend. This parameter represents the sensitivity of the k-th layer feature to changes in pH, where pH represents the current pH value. This indicates the reference pH value.

[0021] A neural network model was used to jointly process the potential signal drift trend and historical oxygen content detection data to obtain an initial simulated response characteristic curve. Multi-layer feature vectors were extracted from the initial simulated response characteristic curve to determine the response distribution of the feature vectors under different pH conditions. The correspondence coefficient between pH changes and feature vectors was calculated using the response distribution to obtain a preliminary quantitative correspondence. Curve fitting was performed on the preliminary quantitative correspondence to determine a smoothed simulated response characteristic curve. The influence weight of oxygen content fluctuations on corrosion sensitivity was obtained from the smoothed simulated response characteristic curve. If the influence weight exceeded a preset threshold, the corresponding pH range was marked as a high-sensitivity region. For the high-sensitivity regions, associated time-series segments were extracted from historical oxygen content detection data to obtain drift increment sequences within the time-series segments. Regression analysis was performed between the drift increment sequences and pH changes to determine the final quantitative correspondence of corrosion sensitivity.

[0022] The system first processes potential signal drift trend data using a neural network model, combining it with historical oxygen content detection data to construct a simulated response characteristic curve, quantifying the mapping relationship between pH changes and erosion sensitivity. In specific implementation, the system automatically extracts oxygen content data from the database over the past 48 hours, such as the time-series change in oxygen concentration from 21.5% to 18.3%, and loads corresponding pH records, such as a dataset showing pH fluctuations between 5.8 and 4.3. Next, a Long Short-Term Memory (LSTM) network algorithm is used to model the time-series data, setting the number of hidden layer units to 64, the time step to 12 hours, the number of training iterations to 200, the Adam optimizer to be used, and the learning rate to be 0.001. The resulting model-predicted response curve has a mean squared error of 0.15. Subsequently, the system analyzed the relationship between pH changes and erosion sensitivity using curve analysis. It calculated that for every 0.5 unit decrease in pH, the erosion sensitivity index increased by approximately 7.2%, generating a mapping table. For example, the sensitivity index was 35.6 at pH 4.5, while it decreased to 28.4 at pH 5.0. To further verify the reliability of the mapping relationship, the system automatically retrieved historical environmental parameters, such as erosion records at 75% humidity and 85 degrees Celsius, and performed regression analysis using a Support Vector Machine (SVM) algorithm. The radial basis function (RBF) kernel was selected, and the penalty parameter C was set to 10. The prediction error was 0.09, and the correlation coefficient was 0.88, indicating high accuracy in the mapping relationship. Finally, the system stored the quantification results in the environmental impact assessment database, automatically updating the parameters of the erosion risk model, providing data support for subsequent environmental monitoring, and forming a complete logical chain from data extraction to relationship quantification.

[0023] Step S104: By analyzing the obtained quantization mapping relationship, and integrating real-time monitoring inputs of acid and alkali gases to meet sensor stability requirements, the response characteristic calibration parameters are optimized using a support vector machine algorithm to obtain the adjusted signal drift compensation model. The specific model is as follows: ; This formula represents the decision function of the support vector machine algorithm for signal drift calibration; This represents the output calibration value of the support vector machine, where M represents the number of support vectors. This represents the weight coefficient of the k-th support vector. The label represents the k-th support vector. Represents the kernel function. Let x represent the k-th support vector, x represent the input drift signal features, and b represent the bias parameter.

[0024] Real-time monitoring data of acid and alkali gases is used to perform preliminary processing of the monitoring input. Standardization tools are employed to unify the data format, resulting in a processed monitoring dataset. Based on this dataset, sensor response changes are analyzed, key fluctuation segments in the response signal are extracted, and the corresponding time windows are determined. For each time window corresponding to a fluctuation segment, the signal drift distribution range is obtained, and any range exceeding a preset threshold is marked as an abnormal drift range. Based on these abnormal drift ranges, a support vector machine algorithm is used to calibrate the signal drift, constructing a compensation model to obtain calibrated signal compensation parameters. For these calibrated parameters, and considering stability requirements, the sensor response configuration is adjusted to determine the adjusted response baseline value. Using this baseline value, a drift correction scheme is generated. This scheme is then matched with an optimization framework to obtain the final correction execution strategy. Based on this strategy, the acid and alkali gas monitoring input is continuously tracked to obtain real-time calibrated sensor output data.

[0025] The system first automatically extracts real-time monitoring data of acid and alkaline gases from the environmental monitoring database based on the obtained quantitative mapping relationship. For example, it records the change in acid gas concentration from 12.3 ppm to 15.7 ppm in the past 24 hours. Combining this with sensor stability requirements, it loads corresponding signal output deviation data, such as a dataset of voltage offsets fluctuating between 0.02V and 0.05V. Next, the system uses a support vector machine algorithm to model and analyze the deviation between the real-time monitoring input and the signal. The kernel function is set to a linear kernel, the penalty parameter C is 5, and the tolerance parameter epsilon is 0.01. Through regression processing of the data, the system calculates the response characteristic calibration parameters. For example, the calibration coefficient is adjusted from 1.08 to 1.12, and the deviation correction value is reduced from 0.03V to 0.01V. Subsequently, the system applies the calibration parameters to the signal drift compensation model, automatically generating adjusted model parameters. For example, the drift compensation coefficient is updated to 0.95, and the signal smoothing window size is set to 8 hours. Based on historical trends in acid and alkaline gas concentration changes, the analysis shows that the compensation model's response error for concentration fluctuations less than 2 ppm is controlled within 0.005V. To ensure model adaptability, the system further utilizes environmental humidity data, such as records of humidity changes from 60% to 72%, to analyze the potential impact of humidity on signal drift. It calculates that for every 5% increase in humidity, the signal offset increases by approximately 0.002V, and this influencing factor is incorporated into the compensation model, forming a complete parameter optimization logic. Finally, the system stores the adjusted signal drift compensation model in the sensor calibration database, automatically updating the calibration rules of the real-time monitoring system to provide stable support for subsequent acid and alkali gas monitoring, ensuring a closed-loop process from data extraction to model optimization.

[0026] Step S105: Based on the obtained adjusted signal drift compensation model, acquire experimental sample data of zirconia material under extreme conditions, and determine if the matching degree between the compensation model output and the actual response characteristics is higher than the threshold. Then, confirm the adaptability to the corrosion environment and determine the long-term monitoring reliability index.

[0027] This study addresses signal drift issues by collecting experimental sample data of zirconia materials under extreme conditions. Standardized tools are used to preprocess the collected data, resulting in a standardized sample dataset. Based on this standardized dataset, the response data of zirconia materials in a corrosive environment is analyzed, key fluctuation ranges are extracted, and corresponding environmental impact segments are identified. For each environmental impact segment, a preset threshold is used for comparison. If the response data in a fluctuation range exceeds the preset threshold, it is marked as an abnormal response range, resulting in anomaly marking results. Using the anomaly marking results, a support vector machine algorithm is employed to correct the abnormal response ranges, constructing a corrected response data set and determining the corrected data range. Based on the corrected data range, the adaptability of zirconia materials in long-term monitoring is analyzed, stability assessment parameters are obtained, and stability assessment results are obtained. Based on the stability assessment results and the data characteristics of the corrosive environment, a targeted monitoring adjustment strategy is generated, determining the final monitoring configuration scheme. Using the final monitoring configuration scheme, the response data of zirconia materials under extreme conditions is continuously tracked, and real-time adjusted monitoring output data is obtained.

[0028] Based on an adjusted signal drift compensation model, the system automatically extracts sample data of zirconia sensors under high-temperature and high-corrosion environments from an extreme condition experimental database. For example, in a continuous 168-hour exposure test record at 95℃, SO2 concentration of 450ppm, and relative humidity of 85%, the original signal output decayed from an initial 4.72V to 4.31V, accompanied by a response time increase from 18 seconds to 37 seconds. The system first applies the compensation model to this dataset to calculate the compensated signal value. For example, after applying a 0.95 coefficient correction to the hourly drift, the recovered signal range is 4.68V to 4.75V, and the response time is recalculated to be 19 seconds to 22 seconds. Subsequently, the system uses a cosine similarity algorithm to evaluate the matching degree between the compensated output and the actual reference response characteristics. The compensated signal sequence is compared with the standard oxygen partial pressure conversion curve, yielding a similarity score of 0.973, which is much higher than the preset threshold of 0.92. Next, the system further calculated the standard deviation of the matching degree deviation distribution to be 0.007V, and verified it through 1000 random perturbations using Monte Carlo simulation. Within a concentration fluctuation range of ±50ppm, the compensation error remained within 0.012V, thus confirming the adaptability of zirconia materials in highly corrosive environments. Based on the above analysis, the system automatically determined long-term monitoring reliability indicators, such as an expected fault-free operating time of over 12,000 hours and an annual drift rate controlled below 1.8%. These indicators, along with the verification data, were written into the sensor performance evaluation library, achieving a fully automated closed-loop processing from extreme condition data acquisition to reliability determination.

[0029] Step S106: A neural network model is used to refine the correlation between the determined long-term monitoring reliability indicators and pH changes, generating a dynamic response characteristic prediction sequence and obtaining a set of signal drift suppression thresholds. The formula is as follows: ; in, This represents the suppression threshold corresponding to the i-th drift point, and β represents the drift amplitude adjustment coefficient. This represents the value of the i-th detected signal drift key point. The baseline response value is represented by γ, the fluctuation suppression coefficient is represented by L, and the calculation window length is represented by L. This represents the value of the j-th drift point within the window. This represents the average value of the drift points within the window.

[0030] A neural network model was trained on long-term monitoring data of zirconia materials to obtain a refined reliability index sequence. This refined reliability index sequence was then matched with real-time acquired pH change sequences to generate a dynamic response characteristic prediction sequence. The fluctuation amplitude over continuous time periods was analyzed using the dynamic response characteristic prediction sequence to extract a set of key signal drift points. For each key point in the extracted set, a preset comparison rule was used to determine whether it exceeded a preset range; if so, it was marked as a drift point, resulting in a list of marked drift points. The suppression amplitude value corresponding to each drift point was calculated based on the list of marked drift points, determining a set of signal drift suppression thresholds. Subsequent acquired response data was compared point-by-point using the signal drift suppression threshold set. If the current response data exceeded the corresponding threshold, a suppression adjustment command was directly output, resulting in an adjusted monitoring output sequence. The neural network model parameters were updated based on the adjusted monitoring output sequence, completing the regeneration of the dynamic response characteristic prediction sequence for the current period.

[0031] The system employs a neural network model to refine the correlation between long-term monitoring reliability indicators and pH changes, generating a dynamic response characteristic prediction sequence and ultimately obtaining a set of signal drift suppression thresholds. First, the system extracts test data from a historical database of a zirconia sensor within a pH range of 2.5 to 11.5, including 240 consecutive hours of signal recordings. The initial signal strength was 5.12V, which attenuated to a minimum of 4.88V under pH fluctuations. Next, the system constructs a neural network model based on Long Short-Term Memory (LSTM), using pH changes as input features to train the model to predict signal attenuation trends. A mean squared error loss function is used during training, and the optimized model's prediction error is controlled within 0.015V. Then, the system uses the trained model to generate a dynamic response characteristic prediction sequence, predicting signal strength changes for every 0.5 unit change in pH (e.g., a 0.03V decrease in signal strength when pH rises from 5.0 to 5.5), and records the trend curve of response delay increasing from 12 seconds to 15 seconds. Furthermore, the system analyzes the deviation between the predicted sequence and the actual data to calculate a set of signal drift suppression thresholds. Using a K-means clustering algorithm, the drift values ​​are divided into five intervals, resulting in a threshold set ranging from 0.02V to 0.08V, with a maximum threshold of 0.075V for drastic pH changes. Finally, the system binds the threshold set to the pH change intervals to form a dynamic adjustment strategy. For example, when the pH value is below 3.0, 0.06V is automatically selected as the suppression threshold to ensure signal stability. The results are stored in an environmental adaptability analysis library to provide data support for subsequent sensor optimization. This process is implemented through automated algorithms, forming a complete logical chain from data extraction to threshold determination.

[0032] Step S107: By fusing the obtained signal drift suppression threshold set with the real-time oxygen content detection signal, if the predicted sequence shows increased erosion sensitivity, the compensation model update loop is activated to obtain the optimized stable sensor output; the compensation model is: ; Let σ represent the compensation coefficient after optimization (k-th iteration), and let σ represent the sigmoid activation function. This represents the forget gate weight matrix. This indicates the hidden state at the previous moment. This represents the input weight matrix. This represents the k-th historical oxygen content signal sequence. denoted as the bias vector; this formula calculates the optimized compensation coefficients based on the long short-term memory network structure.

[0033] A erosion sensitivity prediction sequence is generated by point-by-point matching of the real-time oxygen content detection signal sequence with a set of signal drift suppression thresholds. Based on the increase in the erosion sensitivity prediction sequence over a continuous time period, if the increase exceeds a preset range, a compensation model activation command is triggered. A Long Short-Term Memory (LSTM) network is used to train the activated compensation model with historical oxygen content signal sequences, resulting in an optimized compensation coefficient sequence. Point-by-point correction operations are performed on the real-time oxygen content detection signal sequence based on the optimized compensation coefficient sequence to determine the corrected sensor output sequence. The signal drift suppression threshold set is then re-matched using the corrected sensor output sequence to obtain an updated erosion sensitivity sequence. The fluctuation change within a continuous segment is calculated based on the updated erosion sensitivity sequence; if the fluctuation change is below a preset range, the stabilized sensor output sequence is confirmed. The original real-time oxygen content detection signal sequence is replaced with the stabilized sensor output sequence to obtain the optimized monitoring output sequence.

[0034] The system dynamically assesses erosion sensitivity by integrating a previously obtained set of signal drift suppression thresholds with real-time oxygen content detection signals. Based on this, it activates a compensation model update loop, ultimately obtaining an optimized, stable sensor output. The system first acquires the current oxygen content signal stream from the real-time acquisition module at a sampling frequency of 10Hz. The average oxygen partial pressure signal over the last 300 seconds is 18.7mV. A signal drift amplitude of 0.11V is observed within the oxygen concentration fluctuation range of 15.2% to 21.8%. Next, the system compares the 0.03V to 0.09V range from the signal drift suppression threshold set with the real-time signal point by point, calculating the moving average of the drift deviation within a sliding window. When the deviation exceeds 0.07V for 50 consecutive sampling points, the probability of increased erosion sensitivity is determined to be above 85%. Subsequently, the system immediately activates a compensation model update loop based on a convolutional neural network (CNN), using the real-time oxygen content sequence and temperature auxiliary features as dual-channel inputs. Online fine-tuning is performed using the Adam optimizer with a learning rate of 0.0005. The update process is executed every 100 new sampling points, with the compensation coefficient adjustment range controlled within 0.012. Furthermore, the updated model performs forward compensation on subsequent signals, outputting the corrected oxygen partial pressure value in real time. For example, when the oxygen concentration suddenly drops to 14.9%, the original signal attenuates by 0.14V, but after compensation, it stabilizes at around 18.5mV, improving the drift suppression rate to 92%. Finally, the system stores the optimized stable output sequence along with the compensation parameters in the sensor adaptive database, forming a closed-loop feedback mechanism. This provides a continuously optimized data foundation for long-term operation under erosion environments. The entire process is automatically executed by the embedded processor, ensuring a complete real-time logic chain from judgment to compensation.

[0035] The above description is merely a specific implementation of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A method for solving the signal drift of an oxygen sensor in a high-temperature high-pressure environment, characterized in that, The method comprises: By collecting the acid-base gas concentration data and the surface state information of the zirconia material in a high-temperature and high-pressure environment, the support vector machine algorithm is used to classify the erosion sensitivity, and an initial response characteristic deviation value is obtained; According to the obtained initial response characteristic deviation value, the corresponding acid-base degree change record is obtained, the signal drift feature vector is extracted therefrom, and if the feature vector exceeds a preset threshold, it is marked as a high-corrosion environment influence area, and a potential signal drift trend is determined; The neural network model is used to process the determined potential signal drift trend and the oxygen content detection historical data, to generate a simulated response characteristic curve, and to obtain a quantitative mapping relationship of the acid-base degree change on the erosion sensitivity; By analyzing the obtained quantitative mapping relationship, the real-time monitoring input of the acid-base gas is integrated according to the sensor stability requirement, the support vector machine algorithm is used to optimize the response characteristic calibration parameter, and an adjusted signal drift compensation model is obtained; According to the obtained adjusted signal drift compensation model, the zirconia material experimental sample data under extreme conditions is obtained, and if the matching degree of the compensation model output and the actual response characteristic is higher than a threshold, the corrosion environment adaptability is confirmed, and a long-term monitoring reliability index is determined; The neural network model is used to refine the determined long-term monitoring reliability index and the acid-base degree change correlation, to generate a dynamic response characteristic prediction sequence, and to obtain a signal drift suppression threshold set; By fusing the obtained signal drift suppression threshold set and the oxygen content detection real-time signal, if the prediction sequence shows that the erosion sensitivity is intensified, the compensation model update cycle is activated, and an optimized sensor stability output is obtained.

2. The method according to claim 1, wherein the method is characterized by: The method comprises: By collecting the acid-base gas concentration data and the surface state information of the zirconia material in a high-temperature and high-pressure environment, the support vector machine algorithm is used to classify the erosion sensitivity, and an initial response characteristic deviation value is obtained; The sensor is used to collect the acid-base gas concentration data and the surface state information of the zirconia material in a high-temperature and high-pressure environment; The collected acid-base gas concentration data and surface state information are preprocessed to remove outliers and standardize the processing; ； wherein, represents the erosion sensitivity level of the i-th data point, represents the weight coefficient of the j-th feature in the support vector machine algorithm, represents the j-th feature value of the i-th data point, including acid-base gas concentration and surface state information, b represents the bias term, and m represents the total number of features; the formula calculates the erosion sensitivity of each sampling point by the support vector machine classification algorithm; The support vector machine algorithm is used to classify the preprocessed concentration data and surface state information to determine the erosion sensitivity level; the specific calculation formula is as follows: If the erosion sensitivity level is higher than a preset threshold, the corresponding data point is marked as a high-sensitivity area; ； wherein, represents the response characteristic deviation value of the kth region on the surface of the zirconia material, represents the actual response value of the ith measuring point in the kth region, represents the standard response value of the kth region, and n represents the total number of high-sensitivity marker points in the region; the formula is used to quantify the response deviation degree of each region on the surface of the material relative to the standard state; The response characteristic deviation value of each area of the zirconia material surface is calculated according to the high-sensitivity area marking result; the specific calculation formula is as follows: According to the response characteristic deviation value distribution, the material surface state change corresponding to the deviation value concentrated area is identified; ； represents a deviation prediction coefficient, represents the acid-base gas concentration value corresponding to the i-th deviation concentration region, represents the average value of the acid-base gas concentrations of all the deviation concentration regions, represents the surface state change amount of the i-th deviation concentration region, represents the average value of the surface state change amounts of all the deviation concentration regions, p represents the total number of the deviation concentration regions; the formula establishes the correlation between the surface state change and the gas concentration through linear regression.

3. The method of claim 1, wherein the method further comprises: The linear regression algorithm is used to analyze the correlation between the surface state change of the deviation value concentrated area and the acid-base gas concentration, and the deviation prediction coefficient is obtained; the specific calculation formula is as follows: ； This is the signal drift eigenvector calculation formula; The signal drift eigenvector value in the time period t is represented by T, and the total number of sampling time points is represented by T, The acid-base response signal value at the i-th time point is represented by The initial reference response value is represented by The signal standard deviation is represented by The weight coefficient at the i-th time point is represented by When When it exceeds the preset threshold, it is marked as a high corrosion environment affected area; According to the obtained initial response characteristic deviation value, the corresponding acid-base degree change record is obtained, the signal drift feature vector is extracted therefrom, and if the feature vector exceeds a preset threshold, it is marked as a high-corrosion environment influence area, and a potential signal drift trend is determined; the specific calculation formula is as follows: According to the signal drift trend direction, the corresponding environmental parameter fluctuation data is obtained, the time sequence segmentation processing is performed on the fluctuation data, and the fluctuation amplitude distribution in each period is obtained; The support vector machine algorithm is used to classify the environmental parameter fluctuation data in each period through the fluctuation amplitude distribution, and the high-risk period interval is determined; According to the high-risk period interval, the corresponding acid-base change details are extracted from the data record, the detail data is smoothed to obtain a stable change trend curve, and the environmental influence factor related to the high corrosion area is obtained through the stable change trend curve. If the environmental influence factor exceeds the preset threshold, it is marked as a key monitoring object; according to the key monitoring object, the historical record of its response deviation is obtained, and the historical record is analyzed in layers to determine the potential mode of deviation accumulation; through the potential mode of deviation accumulation, the long-term change monitoring data related to the initial characteristics is obtained, and if the long-term change monitoring data deviates from the preset range, it is marked as an abnormal state area.

4. The method of claim 1, wherein the method further comprises: The neural network model is used to process the determined potential signal drift trend and oxygen content detection historical data to generate a simulated response characteristic curve, obtain a quantitative mapping relationship of the pH change sensitive to corrosion, and the calculation method is as follows: ； wherein R(pH) represents the simulation response characteristic curve value under a specific pH condition, K represents the number of features extracted by the neural network, represents the weight coefficient of the kth feature, represents the kth feature function, represents the oxygen content detection data, and ΔS represents the signal drift trend, represents the sensitivity parameter of the kth feature to the change in pH, and pH represents the current pH value, represents the reference pH reference value; Specifically: the neural network model is used to jointly process the potential signal drift trend and the oxygen content historical detection data to obtain an initial simulated response characteristic curve; According to the initial simulated response characteristic curve, a multi-layer feature vector is extracted, and the response distribution of the feature vector under different pH conditions is determined; Through the response distribution, the corresponding coefficient of the pH change and the feature vector is calculated to obtain a preliminary quantitative corresponding relationship; The preliminary quantitative corresponding relationship is subjected to curve fitting processing to determine a smoothed simulated response characteristic curve; According to the smoothed simulated response characteristic curve, the influence weight of oxygen content fluctuation on corrosion sensitivity is obtained, and if the influence weight exceeds the preset threshold, the corresponding pH interval is marked as a high sensitivity area; For the high sensitivity area, the associated time sequence segment is extracted from the oxygen content historical detection data to obtain a drift increment sequence in the time sequence segment; Through regression analysis of the drift increment sequence and the pH change, the quantitative corresponding relationship of the final corrosion sensitivity is determined.

5. The method of claim 1, wherein the method further comprises: The quantitative mapping relationship obtained by analysis is used to integrate the real-time monitoring input of the acid-base gas according to the sensor stability requirement, the support vector machine algorithm is used to optimize the response characteristic calibration parameter, and an adjusted signal drift compensation model is obtained; The specific model is: ； The formula represents a decision function of the support vector machine algorithm used for signal drift calibration; M represents the number of support vectors, represents the weight coefficient of the kth support vector, represents the label of the kth support vector, represents the kernel function, represents the kth support vector, x represents the input drift signal feature, and b represents the bias parameter; Specifically, the monitoring input is preliminarily processed through the acid-base gas real-time monitoring data, the data is uniformly formatted by using a standardization tool, and a processed monitoring data set is obtained; According to the processed monitoring data set, the sensor response change is analyzed, the key fluctuation segment in the response signal is extracted, and the time window corresponding to the fluctuation segment is determined; For the time window corresponding to the fluctuation segment, combined with the data analysis result, the distribution interval of the signal drift is obtained, and if the distribution interval exceeds the preset threshold range, it is marked as an abnormal drift interval; According to the abnormal drift interval, the support vector machine algorithm is used to calibrate the signal drift to construct a compensation model and obtain calibrated signal compensation parameters; According to the calibrated signal compensation parameters, the sensor response configuration is adjusted according to the stability requirement to determine the adjusted response reference value; A drift correction scheme is generated based on the adjusted response reference value, and the drift correction scheme is matched with the optimization framework to obtain a final correction execution strategy; According to the final correction execution strategy, the acid-base gas monitoring input is continuously tracked to obtain real-time corrected sensor output data.

6. The method of claim 1, wherein the method is characterized by: According to the obtained adjusted signal drift compensation model, the experimental sample data of the zirconia material under extreme conditions is obtained, and if the compensation model output matches the actual response characteristic above a threshold, the corrosion environment adaptability is confirmed, and the long-term monitoring reliability index is determined, including: Through the experimental sample data of the zirconia material under extreme conditions, data collection is performed for the signal drift problem, and the collected data is preprocessed using a standardization tool to obtain a standardized sample data set; According to the standardized sample data set, the response data of the zirconia material in the corrosion environment is analyzed, the key fluctuation interval is extracted, and the corresponding environmental impact segment is determined; For the environmental impact segment, compare with the preset threshold, if the response data of the fluctuation interval exceeds the preset threshold, mark it as an abnormal response interval to obtain an abnormal marking result; Through the abnormal marking result, the support vector machine algorithm is used to correct the abnormal response interval, a corrected response data set is constructed, and the corrected data range is determined; According to the corrected data range, the adaptability of the zirconia material in long-term monitoring is analyzed, the stability evaluation parameter is obtained, and the stability evaluation result is obtained; According to the stability evaluation result, a targeted monitoring adjustment strategy is generated based on the data characteristics of the corrosion environment, and a final monitoring configuration scheme is determined; Through the final monitoring configuration scheme, the response data of the zirconia material under extreme conditions is continuously tracked, and real-time adjusted monitoring output data is obtained.

7. The method of claim 1, wherein the method further comprises: The long-term monitoring reliability index determined by the neural network model is associated with the pH change to generate a dynamic response characteristic prediction sequence, and a signal drift suppression threshold set is obtained; The formula is: ； wherein, represents the i-th detected signal drift key point value, represents the i-th detected signal drift key point value, represents the reference response value, γ represents the fluctuation suppression coefficient, and L represents the calculation window length, represents the j-th drift point value within the window, represents the average value of the drift points within the window; The specific way is as follows: the neural network model is used to train the collected long-term monitoring data of the zirconia material to obtain a refined reliability index sequence; According to the refined reliability index sequence and the real-time collected pH change sequence, a dynamic response characteristic prediction sequence is generated by input matching; Through the dynamic response characteristic prediction sequence, the fluctuation amplitude in the continuous time period is analyzed, and a signal drift key point set is extracted; For the extracted signal drift key point set, a preset comparison rule is used to determine whether each key point exceeds the preset range, and if it exceeds, it is marked as a drift point to obtain a marked drift point list; According to the marked drift point list, the suppression amplitude value corresponding to each drift point is calculated to determine the signal drift suppression threshold set; The signal drift suppression threshold set is used to compare the subsequent collected response data point by point, and if the current response data exceeds the corresponding threshold, an inhibition adjustment instruction is directly output to obtain an adjusted monitoring output sequence; According to the adjusted monitoring output sequence, the neural network model parameters are updated, and the dynamic response characteristic prediction sequence is regenerated for the current period.

8. The method of claim 1, wherein the method is characterized by: The signal drift inhibition threshold set obtained by fusion is matched with the oxygen content detection real-time signal, and if the predicted sequence shows that the erosion sensitivity is intensified, a compensation model update cycle is activated to obtain an optimized sensor stable output, and the compensation model is: ； denotes the kth optimized compensation coefficient, and σ denotes a sigmoid activation function, denotes a forget gate weight matrix, denotes a hidden state at a previous time, denotes an input weight matrix, denotes a kth historical oxygen content signal sequence, denotes a bias vector; The formula is based on the long short-term memory network structure to calculate the optimized compensation coefficient; Specifically, the oxygen content detection real-time signal sequence is matched point by point with the signal drift inhibition threshold set to generate an erosion sensitivity prediction sequence; According to the rising amplitude of the erosion sensitivity prediction sequence in a continuous time period, if the rising amplitude exceeds a preset range, a compensation model activation instruction is triggered; The long short-term memory network is used to load the historical oxygen content signal sequence to perform parameter training on the activated compensation model, and an optimized compensation coefficient sequence is obtained; According to the optimized compensation coefficient sequence, a point-by-point correction operation is performed on the oxygen content detection real-time signal sequence to determine a corrected sensor output sequence; The signal drift inhibition threshold set is matched with the corrected sensor output sequence to obtain an updated erosion sensitivity sequence; According to the updated erosion sensitivity sequence, the fluctuation change in a continuous period is calculated, and if the fluctuation change is lower than a preset range, a stable sensor output sequence is confirmed; The stable sensor output sequence is used to replace the original oxygen content detection real-time signal sequence to obtain an optimized monitoring output sequence.

Images