Gas analyzer fault early warning method and system based on battery state data

Abstract

The invention discloses a gas analyzer fault early warning method and system based on battery state data, and relates to the technical field of data processing, and the method comprises the steps: collecting necessary parameters such as real-time terminal voltage and open-circuit voltage of an oxygen battery and auxiliary parameters such as an analyzer working mode; preprocessing the necessary sampling parameters and the auxiliary parameters, constructing an initial library, and marking interpolation nodes and unit operation intervals; obtaining and screening a parameter set based on the initial library and the interpolation node to obtain a third initial parameter set; constructing a discharge fault inflection point prediction model in combination with the discharge characteristics of the oxygen battery; a prediction parameter set is generated and imported into the model to obtain an undervoltage fault prediction inflection point; and generating alarm information based on the inflection point. The system comprises an acquisition module, a data processing module, a communication module, an early warning module and subordinate units, and the method is realized. The method has the advantages that fault risks can be recognized in advance, non-planned shutdown is reduced, accurate operation of the analyzer is guaranteed, and the method is suitable for environment protection monitoring and other scenes.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for early warning of faults in a gas analyzer based on battery status data. Background Technology

[0002] Gas analyzers are key equipment in environmental monitoring and industrial production, and their detection accuracy and operational stability directly determine the accuracy of environmental data, production safety, and the scientific nature of decision-making. Among these components, the oxygen cell, as the core power supply unit, is highly susceptible to damage from abnormal voltage, temperature, and other parameters, which can easily lead to sampling deviations, data interruptions, or even shutdowns. Environmental monitoring (such as continuous monitoring of exhaust emissions from power plants) places extremely high demands on the continuous operation of equipment. Therefore, ensuring the reliable operation of gas analyzers is of significant practical importance in preventing environmental data distortion and reducing production risks.

[0003] Existing gas analyzer fault early warning methods and systems based on battery status data mostly rely on periodic manual maintenance for risk identification, resulting in early warning lag and an inability to capture dynamic fault risks such as battery voltage decay and abnormal temperature in real time. Furthermore, they lack targeted optimization of parameter analysis logic based on unit operating status (e.g., normal operation / shutdown), making them prone to missed or incorrect diagnoses due to data interference. They also lack a linkage mechanism with maintenance plans (e.g., shutdown for replacement), leading to high maintenance costs for unplanned downtime and failing to meet the requirements for continuous and accurate equipment operation in environmental monitoring scenarios. Therefore, a gas analyzer fault early warning method and system based on battery status data is needed to address the aforementioned problems. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a method and system for early warning of gas analyzer faults based on battery status data. This solution resolves the problems of existing methods and systems for early warning of gas analyzer faults based on battery status data, which rely on periodic manual maintenance leading to delayed warnings, are unable to capture dynamic fault hazards in real time, lack optimization of parameter analysis logic based on unit operating status leading to missed or misjudged faults, lack of linkage mechanism with operation and maintenance plans, and high maintenance costs due to unplanned downtime. These issues make it difficult to meet the requirements for continuous and accurate operation of equipment in environmental monitoring scenarios.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A fault early warning method for a gas analyzer based on battery state data includes:

[0007] S100. Collects real-time terminal voltage, open-circuit voltage, temperature and cumulative charge-discharge cycle count of oxygen cell to obtain essential battery parameters. Simultaneously collects auxiliary analyzer parameters, including the working mode of the gas analyzer, the oxygen measurement value and the standard gas calibration deviation, and the sensor output voltage signal.

[0008] S200. Perform data preprocessing on the mandatory battery parameters and auxiliary analyzer parameters respectively, construct the initial parameter library to be analyzed, synchronously obtain the time nodes corresponding to the outliers and missing values ​​in the data preprocessing process as interpolation nodes, and mark the time interval corresponding to the unit's operating status.

[0009] S300. Based on the initial parameter library to be analyzed and the interpolation nodes, obtain the first initial parameter set and the second initial parameter set corresponding to the interpolation nodes. Simultaneously, according to a preset time window, determine whether there are any additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, remove the first initial parameter set and / or the second initial parameter set with additional interpolation nodes. If not, truncate the first initial parameter set and / or the second initial parameter set according to the preset time window to obtain the third initial parameter set.

[0010] S400. Based on the first initial parameter set, the second initial parameter set, and the third initial parameter set, combined with the oxygen cell discharge characteristics of the gas analyzer, obtain the discharge fault inflection point prediction dataset, and construct the discharge fault inflection point prediction model. The model input features include the oxygen cell voltage decay rate, the oxygen measurement value deviation growth rate, and the temperature influence coefficient.

[0011] S500. Based on the initial parameter library to be analyzed, generate the first prediction parameter set, import the first prediction parameter set into the discharge fault inflection point prediction model, and obtain the oxygen battery voltage insufficient fault prediction inflection point.

[0012] S600 generates alarm information based on the inflection point of oxygen cell voltage deficiency fault prediction.

[0013] In an optional embodiment, step S100 specifically includes:

[0014] S101. Deploy the voltage acquisition module and connect it to the positive and negative terminals of the oxygen cell. Set the acquisition frequency to 1Hz and record the oxygen cell terminal voltage data U(t) in real time, where t is the acquisition timestamp in seconds.

[0015] S102. When the unit is shut down, disconnect the oxygen cell from the analyzer load by switching the switch, collect the oxygen cell open circuit voltage U0 for 5 minutes, and take the average voltage during this period as the open circuit voltage reference value under the shutdown state.

[0016] S103. Attach a temperature sensor to the surface of the oxygen battery casing, set the acquisition frequency to 0.1Hz, record the battery temperature data T(t), and synchronously associate it with the corresponding voltage acquisition timestamp;

[0017] S104. Record the cumulative working time t_total of the oxygen battery from installation and activation to the current moment through the counter, and calculate the cumulative number of charge and discharge cycles N by combining the working mode conversion records of the analyzer. A single cycle is defined as a complete working cycle from the start-up of the analyzer to the shutdown.

[0018] S105. Read the operating mode signal M(t) from the control system of the gas analyzer, where M(t) = 1 indicates continuous monitoring mode and M(t) = 0 indicates intermittent calibration mode;

[0019] S106. Acquire the oxygen measurement value C(t) output by the analyzer, and simultaneously obtain the standard gas concentration value C_std introduced at the same time, and calculate the real-time calibration deviation ΔC(t)=|C(t)-C_std| / C_std×100%;

[0020] S107. Acquire the raw voltage signal V_s(t) output by the sensor. This signal is the raw response voltage of the oxygen cell when it participates in the electrochemical reaction. It forms a correlation and comparison data with the oxygen cell terminal voltage U(t).

[0021] S108. Align the collected U(t), U0, T(t), N, M(t), ΔC(t), and V_s(t) in steps S101-S107 according to the timestamps, and store them in the real-time database to form the original record of parameter collection.

[0022] In an optional embodiment, step S200 specifically includes:

[0023] S201. Noise filtering is applied to the real-time terminal voltage U(t) of the oxygen cell in the mandatory battery parameters. The sliding window averaging method is used, with a window size of 5 sampling points. The filtered voltage value is then calculated.

[0024] U_f(t)=[U(t-4)+U(t-3)+U(t-2)+U(t-1)+U(t)] / 5, where t≥4;

[0025] S202. Perform outlier detection on battery temperature data T(t), set a normal temperature range, and mark T(t) as an outlier when it exceeds the normal temperature range, and record the corresponding time node t_err;

[0026] S203. The real-time calibration deviation ΔC(t) in the auxiliary analyzer parameters is smoothed by using the exponential weighted average method to calculate the smoothed deviation value ΔC_f(t)=α×ΔC(t)+(1-α)×ΔC_f(t-1), where α is the smoothing coefficient;

[0027] S204. Set a missing duration threshold Δt_std. For all missing values ​​in the mandatory battery parameters and auxiliary analyzer parameters, determine the missing duration Δt. When Δt ≤ Δt_std, use linear interpolation to fill the missing value. That is, for the missing point t_m, the filled value is:

[0028] X(t_m)=X(t_m-1)+[X(t_m+1)-X(t_m-1)]×(t_m-t_m-1) / (t_m+1-t_m-1);

[0029] S205. When the missing duration Δt > Δt_std, mark the time period as a data interruption interval and record the start time t_start and end time t_end of the interval as interpolation nodes;

[0030] S206. Integrate the filtered, smoothed, and interpolated parameter data to construct the initial parameter library DB to be analyzed. Each record in the library contains a timestamp t and the corresponding U_f(t), T_f(t), N, M(t), ΔC_f(t), and V_s_f(t).

[0031] S207. Obtain the unit operating status signal S(t) from the unit control system, where S(t) = 1 indicates normal operation and S(t) = 0 indicates shutdown. Associate S(t) with the timestamp in the parameter library DB to mark the time interval of the unit operating status corresponding to each parameter.

[0032] S208. Statistically calculate the number of outliers N_err and the number of missing value intervals N_miss generated during the preprocessing process.

[0033] In an optional embodiment, step S300 specifically includes:

[0034] S301. Extract parameter data corresponding to all interpolation nodes from the initial parameter library DB to be analyzed, forming the first initial parameter set P1, where each record in P1 contains the interpolation node time t_k and the corresponding parameter value set {U_f(t_k),T_f(t_k),ΔC_f(t_k)};

[0035] S302. Extract parameter data within a 30-second range before and after the interpolation node to form a second initial parameter set P2, where each record in P2 contains a time series [t_k-30, t_k+30] and the corresponding continuous parameter values;

[0036] S303. Set a preset time window W, where W = 3600 seconds when the unit is in normal operation and W = 300 seconds when it is in shutdown state;

[0037] S304. Traverse each record in P1 and check if there are other interpolation nodes in its corresponding time window. If there are, mark the record as invalid data and remove it from P1.

[0038] S305. Perform the same validity check on P2, remove time series containing additional interpolation nodes, and retain the valid subset of P2;

[0039] S306. Cut the valid subsets of P1 and P2 according to the time window W, ensuring that the time length of each cut segment is W and there are no interpolation nodes in the segment, to form the third initial parameter set P3;

[0040] S307. Select data segments from P3 under the unit shutdown state, extract the oxygen cell open circuit voltage U0 data, calculate the average value U0_avg of U0 within the same shutdown cycle, and use it as the model calibration benchmark value for that period.

[0041] S308. Establish a parameter set index table to record the time range, parameter types, and corresponding unit operating status of each parameter set in P1, P2, and P3.

[0042] In an optional embodiment, step S400 specifically includes:

[0043] S401. Extract oxygen cell voltage data U_f(t) from the first initial parameter set P1, calculate the voltage difference ΔU = U_f(t+1) - U_f(t) between adjacent time points, and calculate the oxygen cell voltage decay rate v_U = ΣΔU / 100 based on the cumulative value of ΔU over 100 consecutive sampling points, in units of V / second;

[0044] S402. Extract calibration deviation data ΔC_f(t) from the second initial parameter set P2, and calculate the oxygen measurement deviation growth rate v_ΔC=(ΔC_f(t_end)-ΔC_f(t_start)) / (t_end-t_start), in % / second, where t_start and t_end are the start and end times of the time series in P2;

[0045] S403. Extract temperature data T_f(t) from the third initial parameter set P3, combine it with oxygen cell voltage data U_f(t), and calculate the temperature influence coefficient k_T using a multiple linear regression method:

[0046] U_f(t)=a×T_f(t)+b×v_U+c, where k_T=a is the coefficient of temperature on voltage;

[0047] S404. Integrate oxygen cell voltage decay rate v_U, oxygen measurement deviation growth rate v_ΔC, and temperature influence coefficient k_T as input features to construct a discharge fault inflection point prediction dataset D, where each sample contains a feature vector [v_U,v_ΔC,k_T] and the corresponding actual fault occurrence time label t_fail;

[0048] S405. An LSTM neural network is used to construct a discharge fault inflection point prediction model, and the output value is the predicted fault inflection point time t_pred;

[0049] S406. Set the oxygen cell open-circuit voltage error threshold U0_avg_std, use the oxygen cell open-circuit voltage U0_avg during shutdown as a benchmark, calibrate the model, and adjust the model parameters so that the error of the predicted voltage curve at U0_avg is ≤U0_avg_std.

[0050] S407. Use 70% of dataset D as the training set and 30% as the test set. Train the model using the backpropagation algorithm until the model's prediction error MAE on the test set is ≤24 hours.

[0051] S408. Save the parameters of the trained model, establish a model version management mechanism, and record the model training time, the range of the dataset used, and the calibration benchmark value.

[0052] In an optional embodiment, step S500 specifically includes:

[0053] S501. Extract parameter data from the initial parameter library DB to be analyzed for the most recent 72 hours, and generate a first prediction parameter set P_pred that includes the new oxygen cell voltage decay rate v_U_new, the new oxygen measurement deviation growth rate v_ΔC_new, and the new temperature influence coefficient k_T_new according to the feature calculation method in step S400.

[0054] S502. Input P_pred into the trained discharge fault inflection point prediction model to obtain the oxygen battery voltage insufficient fault prediction inflection point t_pred. The oxygen battery voltage insufficient fault prediction inflection point is the time point when the oxygen battery voltage drops to the normal operating threshold U_min of the analyzer.

[0055] In an optional embodiment, step S600 specifically includes:

[0056] S601. Obtain the next planned shutdown time t_stop from the unit operation and maintenance system, and calculate the difference between the predicted inflection point and the shutdown time Δt_stop=t_stop-t_pred;

[0057] S602. When Δt_stop<0, that is, the predicted inflection point is earlier than the planned shutdown time, a level one warning is triggered. Among them, when the difference between t_pred and the current time t_now, Δt_now=t_pred-t_now<7 days, the warning level is level one.

[0058] S603. When 7 days ≤ Δt_now ≤ 30 days, the warning level is Level II, indicating that the oxygen cell should be replaced during the next shutdown.

[0059] Furthermore, a gas analyzer fault early warning system based on battery state data is proposed to implement the early warning method described above, including:

[0060] The acquisition module is used to acquire the real-time terminal voltage, open-circuit voltage, temperature and cumulative charge-discharge cycle number of the oxygen cell to obtain the essential battery parameters. It is also used to acquire auxiliary analyzer parameters, including the working mode of the gas analyzer, the oxygen measurement value and the calibration deviation of the standard gas, and the sensor output voltage signal.

[0061] The data processing module is used to preprocess the mandatory battery parameters and auxiliary analyzer parameters respectively, construct an initial parameter library to be analyzed, obtain the time nodes corresponding to outliers and missing values ​​in the data preprocessing process as interpolation nodes, and mark the time intervals corresponding to the unit operating status. Based on the initial parameter library to be analyzed and the interpolation nodes, it obtains the first initial parameter set and the second initial parameter set corresponding to the interpolation nodes, and simultaneously determines whether there are additional interpolation nodes in the first initial parameter set and / or the second initial parameter set according to a preset time window. If so, the first initial parameter set and / or the second initial parameter set with additional interpolation nodes are then processed. If the parameter set is removed, the first initial parameter set and / or the second initial parameter set are truncated according to a preset time window to obtain the third initial parameter set. It is also used to obtain the discharge fault inflection point prediction dataset based on the first initial parameter set, the second initial parameter set and the third initial parameter set, combined with the oxygen battery discharge characteristics of the gas analyzer, and to construct the discharge fault inflection point prediction model. The model input features include the oxygen battery voltage decay rate, the oxygen measurement value deviation growth rate and the temperature influence coefficient. It is used to generate the first prediction parameter set according to the initial parameter library to be analyzed, and import the first prediction parameter set into the discharge fault inflection point prediction model to obtain the oxygen battery insufficient voltage fault prediction inflection point.

[0062] The communication module is used to generate alarm information based on the predicted inflection point of insufficient oxygen cell voltage fault, and send alarm information including the predicted inflection point time and the current voltage value to the remote operation and maintenance platform.

[0063] The early warning module is used to issue early warning signals based on alarm information.

[0064] In an optional embodiment, the acquisition module includes:

[0065] The first acquisition unit is used to acquire the real-time terminal voltage, open-circuit voltage, battery temperature and cumulative charge-discharge cycle number of the oxygen battery to obtain the essential battery parameters.

[0066] The second acquisition unit is used to acquire auxiliary analyzer parameters, including the gas analyzer's operating mode, oxygen measurement value and standard gas calibration deviation, and sensor output voltage signal.

[0067] In an optional embodiment, the data processing module includes:

[0068] The preprocessing unit is used to preprocess the parameters of the required battery and the parameters of the auxiliary analyzer respectively, and to build an initial parameter library to be analyzed;

[0069] The marking unit is used to obtain the time nodes corresponding to outliers and missing values ​​in the data preprocessing process as interpolation nodes, and to mark the time intervals corresponding to the unit's operating status.

[0070] The filtering unit is used to obtain the first initial parameter set and the second initial parameter set corresponding to the interpolation node based on the initial parameter library to be analyzed and the interpolation node. Simultaneously, according to a preset time window, it determines whether there are any additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, the first initial parameter set and / or the second initial parameter set with additional interpolation nodes are removed. If not, the first initial parameter set and / or the second initial parameter set are truncated according to the preset time window to obtain the third initial parameter set.

[0071] The model building unit is used to obtain a discharge fault inflection point prediction dataset based on the first initial parameter set, the second initial parameter set, and the third initial parameter set, combined with the oxygen cell discharge characteristics of the gas analyzer, and to build a discharge fault inflection point prediction model. The model input features include the oxygen cell voltage decay rate, the oxygen measurement value deviation growth rate, and the temperature influence coefficient.

[0072] The fault prediction unit is used to generate a first prediction parameter set based on the initial parameter library to be analyzed, and to import the first prediction parameter set into the discharge fault inflection point prediction model to obtain the oxygen battery voltage insufficient fault prediction inflection point.

[0073] Compared with the prior art, the beneficial effects of the present invention are:

[0074] This solution proposes a fault early warning method for gas analyzers based on battery status data. By real-time acquisition of essential battery parameters such as oxygen battery terminal voltage, oxygen battery open-circuit voltage, battery temperature, and cumulative charge-discharge cycle count, and simultaneously acquiring auxiliary analyzer parameters such as gas analyzer operating mode, oxygen measurement value and standard gas calibration deviation, and sensor output voltage signal, it achieves comprehensive and dynamic monitoring of the oxygen battery status and related parameters of the gas analyzer. This avoids problems such as delayed early warning, inability to capture dynamic fault hazards such as battery voltage decay and abnormal temperature in real time, and ensures the real-time and completeness of parameter monitoring.

[0075] This solution proposes a gas analyzer fault early warning method based on battery status data. By performing data preprocessing such as noise filtering, outlier detection, smoothing, and missing value imputation on the mandatory battery parameters and auxiliary analyzer parameters respectively, an initial parameter library to be analyzed is constructed. The time interval corresponding to the unit's operating status is marked simultaneously. The parameter set extraction logic is optimized by combining the unit's normal operation and shutdown status, which achieves accurate adaptation between parameter analysis and unit operating status. The parameter analysis logic is optimized in a targeted manner based on the unit's operating status, which reduces the problem of missed or misjudged judgments due to data interference, and improves the accuracy and adaptability of parameter analysis.

[0076] This solution proposes a fault early warning method for gas analyzers based on battery status data. By constructing a discharge fault inflection point prediction model based on the first, second, and third initial parameter sets and combined with the oxygen battery discharge characteristics, the model uses the oxygen battery voltage decay rate, the oxygen measurement deviation growth rate, and the temperature influence coefficient as input features. This generates a fault prediction inflection point for insufficient oxygen battery voltage. Then, combined with the next planned shutdown time of the unit, a graded early warning is generated (level one early warning: flashing red light + remote alarm; level two early warning: yellow light indicating shutdown and replacement). This achieves deep linkage between fault early warning and operation and maintenance plan, meeting the needs of continuous and accurate equipment operation in environmental monitoring scenarios. Attached Figure Description

[0077] Figure 1 This is a flowchart of a gas analyzer fault early warning method based on battery status data proposed in this invention;

[0078] Figure 2 This is a flowchart of the data preprocessing for the required battery parameters and auxiliary analyzer parameters in this invention;

[0079] Figure 3 This is a flowchart illustrating the acquisition process of the first initial parameter set, the second initial parameter set, and the third initial parameter set in this invention.

[0080] Figure 4 This is a system framework diagram of a gas analyzer fault early warning system based on battery status data proposed in this invention. Detailed Implementation

[0081] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0082] Reference Figure 1 - Figure 4 As shown, a fault early warning method for a gas analyzer based on battery status data includes:

[0083] S100. Collects real-time terminal voltage, open-circuit voltage, temperature and cumulative charge-discharge cycle count of oxygen cell to obtain essential battery parameters. Simultaneously collects auxiliary analyzer parameters, including the working mode of the gas analyzer, the oxygen measurement value and the standard gas calibration deviation, and the sensor output voltage signal.

[0084] S200. Perform data preprocessing on the mandatory battery parameters and auxiliary analyzer parameters respectively, construct the initial parameter library to be analyzed, synchronously obtain the time nodes corresponding to the outliers and missing values ​​in the data preprocessing process as interpolation nodes, and mark the time interval corresponding to the unit's operating status.

[0085] S300. Based on the initial parameter library to be analyzed and the interpolation nodes, obtain the first initial parameter set and the second initial parameter set corresponding to the interpolation nodes. Simultaneously, according to a preset time window, determine whether there are any additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, remove the first initial parameter set and / or the second initial parameter set with additional interpolation nodes. If not, truncate the first initial parameter set and / or the second initial parameter set according to the preset time window to obtain the third initial parameter set.

[0086] Among them, the oxygen cell open-circuit voltage data during the shutdown period is retained as the model calibration benchmark.

[0087] S400. Based on the first initial parameter set, the second initial parameter set, and the third initial parameter set, combined with the oxygen cell discharge characteristics of the gas analyzer, obtain the discharge fault inflection point prediction dataset, and construct the discharge fault inflection point prediction model. The model input features include the oxygen cell voltage decay rate, the oxygen measurement value deviation growth rate, and the temperature influence coefficient.

[0088] S500. Based on the initial parameter library to be analyzed, generate the first prediction parameter set, import the first prediction parameter set into the discharge fault inflection point prediction model, and obtain the oxygen battery voltage insufficient fault prediction inflection point.

[0089] S600 generates alarm information based on the inflection point of oxygen cell voltage deficiency fault prediction.

[0090] Furthermore, step S100 specifically includes:

[0091] S101. Deploy the voltage acquisition module and connect it to the positive and negative terminals of the oxygen cell. Set the acquisition frequency to 1Hz and record the oxygen cell terminal voltage data U(t) in real time, where t is the acquisition timestamp in seconds.

[0092] S102. When the unit is shut down, disconnect the oxygen cell from the analyzer load by switching the switch, collect the oxygen cell open circuit voltage U0 for 5 minutes, and take the average voltage during this period as the open circuit voltage reference value under the shutdown state.

[0093] S103. Attach a temperature sensor to the surface of the oxygen battery casing, set the acquisition frequency to 0.1Hz, record the battery temperature data T(t), and synchronously associate it with the corresponding voltage acquisition timestamp;

[0094] S104. Record the cumulative working time t_total of the oxygen battery from installation and activation to the current moment through the counter, and calculate the cumulative number of charge and discharge cycles N by combining the working mode conversion records of the analyzer. A single cycle is defined as a complete working cycle from the start-up of the analyzer to the shutdown.

[0095] S105. Read the operating mode signal M(t) from the control system of the gas analyzer, where M(t) = 1 indicates continuous monitoring mode and M(t) = 0 indicates intermittent calibration mode;

[0096] S106. Acquire the oxygen measurement value C(t) output by the analyzer, and simultaneously obtain the standard gas concentration value C_std introduced at the same time, and calculate the real-time calibration deviation ΔC(t)=|C(t)-C_std| / C_std×100%;

[0097] S107. Acquire the raw voltage signal V_s(t) output by the sensor. This signal is the raw response voltage of the oxygen cell when it participates in the electrochemical reaction. It forms a correlation and comparison data with the oxygen cell terminal voltage U(t).

[0098] S108. Align the U(t), U0, T(t), N, M(t), ΔC(t), and V_s(t) collected in steps S101-S107 according to the timestamps, and store them in the real-time database to form the original record of parameter collection.

[0099] Specifically, steps S101-S108 involve deploying a voltage acquisition module to acquire the real-time terminal voltage U(t) of the oxygen battery at a frequency of 1Hz, acquiring and calculating the open-circuit voltage reference value U0 of the oxygen battery when the unit is shut down, acquiring the battery temperature T(t) at a frequency of 0.1Hz and associating it with a timestamp, recording the cumulative working time and calculating the number of charge-discharge cycles N, reading the analyzer's working mode signal M(t), calculating the calibration deviation ΔC(t) between the oxygen measurement value and the standard gas, and acquiring the sensor's raw voltage V_s(t). All parameters are then stored aligned with the timestamps. This achieves comprehensive, synchronous, and high-precision acquisition of the core state of the oxygen battery (voltage, temperature, cycle life) and the gas analyzer's operational parameters (working mode, detection accuracy, sensor response). This provides complete and reliable raw data support for subsequent data preprocessing and fault warning model construction, avoiding warning deviations caused by incomplete or asynchronous parameter acquisition.

[0100] The collected parameters cover two dimensions: "battery status" and "analyzer performance." The oxygen cell open-circuit voltage U0 serves as a baseline value under shutdown conditions and can be used for subsequent model calibration. The calibration deviation ΔC(t) directly reflects the analyzer's detection accuracy and correlates battery status with the validity of the detection results. A comparison between the sensor's raw voltage V_s(t) and terminal voltage U(t) helps pinpoint the root cause of the fault (battery or sensor issue). Simultaneously, a 1Hz voltage acquisition frequency dynamically captures voltage fluctuations, while a 0.1Hz temperature acquisition frequency balances accuracy and power consumption. Timestamp alignment ensures parameter timing consistency. These designs not only meet the real-time and complete data requirements of fault early warning but also adapt to the actual operating power consumption and data processing capabilities of gas analyzers (such as those used in environmental monitoring scenarios).

[0101] Furthermore, step S200 specifically includes:

[0102] S201. Noise filtering is applied to the real-time terminal voltage U(t) of the oxygen cell in the mandatory battery parameters. The sliding window averaging method is used, with a window size of 5 sampling points. The filtered voltage value is then calculated.

[0103] U_f(t)=[U(t-4)+U(t-3)+U(t-2)+U(t-1)+U(t)] / 5, where t≥4;

[0104] S202. Perform outlier detection on battery temperature data T(t), set a normal temperature range, and mark T(t) as an outlier when it exceeds the normal temperature range, and record the corresponding time node t_err;

[0105] S203. The real-time calibration deviation ΔC(t) in the auxiliary analyzer parameters is smoothed by using the exponential weighted average method to calculate the smoothed deviation value ΔC_f(t)=α×ΔC(t)+(1-α)×ΔC_f(t-1), where α is the smoothing coefficient;

[0106] S204. Set a missing duration threshold Δt_std. For all missing values ​​in the mandatory battery parameters and auxiliary analyzer parameters, determine the missing duration Δt. When Δt ≤ Δt_std, use linear interpolation to fill the missing value. That is, for the missing point t_m, the filled value is:

[0107] X(t_m)=X(t_m-1)+[X(t_m+1)-X(t_m-1)]×(t_m-t_m-1) / (t_m+1-t_m-1);

[0108] S205. When the missing duration Δt > Δt_std, mark the time period as a data interruption interval and record the start time t_start and end time t_end of the interval as interpolation nodes;

[0109] S206. Integrate the filtered, smoothed, and interpolated parameter data to construct an initial parameter library DB to be analyzed. Each record in the library contains a timestamp t and the corresponding U_f(t), T_f(t), N, M(t), ΔC_f(t), and V_s_f(t). Among them, U_f(t) is the real-time terminal voltage of the oxygen cell after filtering, T_f(t) is the temperature data of the oxygen cell after processing, N is the cumulative number of charge-discharge cycles of the oxygen cell, M(t) is the working mode signal of the gas analyzer, ΔC_f(t) is the deviation between the smoothed oxygen measurement value and the standard gas calibration, and V_s_f(t) is the original output voltage signal of the sensor after processing.

[0110] S207. Obtain the unit operating status signal S(t) from the unit control system, where S(t) = 1 indicates normal operation and S(t) = 0 indicates shutdown. Associate S(t) with the timestamp in the parameter library DB to mark the time interval of the unit operating status corresponding to each parameter.

[0111] S208. Count the number of outliers N_err and the number of missing value intervals N_miss generated during the statistical preprocessing. When N_err accounts for more than 5% of the total data volume or N_miss exceeds 3, trigger a data acquisition quality warning and prompt to check the acquisition equipment.

[0112] Specifically, steps S201-S208 involve: calculating the filtered voltage U_f(t) from the real-time terminal voltage U(t) of the oxygen cell in the mandatory battery parameters using the sliding window averaging method; detecting outliers and marking abnormal time nodes t_err from the battery temperature data T(t); obtaining the smoothed deviation ΔC_f(t) from the calibration deviation ΔC(t) in the auxiliary analyzer parameters using the exponential weighted averaging method; filling short-duration missing values ​​with linear interpolation based on the missing duration threshold Δt_std and marking long-duration missing intervals as interpolation nodes; constructing the initial parameter library DB to be analyzed from the integrated processed parameters; and associating the unit operating status signals. S(t) marks the operating range corresponding to each parameter, the number of outliers N_err and the number of missing value ranges N_miss during the statistical preprocessing process, realizing the systematic optimization of the raw parameters (mandatory battery parameters and auxiliary analyzer parameters) collected by S101-S108, eliminating the interference of instantaneous noise, invalid outliers and missing data in the raw data on subsequent analysis, and completing the structured storage of parameters and the labeling of operating scenarios. This provides a high-quality and scenario-based data foundation for extracting effective parameter sets in S300 and building fault prediction models in S400, avoiding misjudgment or omission of fault warnings due to defects in the quality of raw data.

[0113] The sliding window averaging method (window size of 5 sampling points) can accurately filter high-frequency interference (such as circuit fluctuations) in voltage data, ensuring the stability of U_f(t) and providing a reliable data source for subsequent calculation of voltage decay rate (S401). The exponential weighted averaging method (smoothing coefficient α ranges from 0.3 to 0.5) retains the real-time trend of calibration deviation ΔC(t) while weakening the impact of instantaneous fluctuations, enabling ΔC_f(t) to accurately reflect the true changes in the analyzer's detection accuracy. Missing values ​​are processed differently according to their duration, ensuring the continuity of short-duration data through linear interpolation and avoiding data distortion caused by blindly filling in long-duration missing intervals. Corresponding to the unit's operating status allows subsequent parameter analysis to adapt to the differences in parameter characteristics under different operating conditions (normal operation / shutdown), improving the analysis's relevance. Furthermore, statistically analyzing the number of outliers and missing values ​​and triggering acquisition quality warnings can promptly detect acquisition equipment faults (such as poor sensor contact), ensuring data reliability from the source and fully meeting the core requirements of gas analyzers for the accuracy and completeness of fault warning data in environmental monitoring and other scenarios.

[0114] Furthermore, step S300 specifically includes:

[0115] S301. Extract the parameter data corresponding to all interpolation nodes from the initial parameter library DB to be analyzed, forming the first initial parameter set P1. Each record in P1 contains the interpolation node time t_k and the corresponding parameter value set {U_f(t_k), T_f(t_k), ΔC_f(t_k)}, where U_f(t_k) is the real-time terminal voltage of the oxygen cell after filtering at the interpolation node t_k, T_f(t_k) is the processed oxygen cell temperature data at the interpolation node t_k, and ΔC_f(t_k) is the smoothed oxygen measurement calibration deviation at the interpolation node t_k.

[0116] S302. Extract parameter data within a 30-second range before and after the interpolation node to form a second initial parameter set P2, where each record in P2 contains a time series [t_k-30, t_k+30] and the corresponding continuous parameter values;

[0117] S303. Set a preset time window W, where W = 3600 seconds (1 hour) when the unit is in normal operation and W = 300 seconds (5 minutes) when the unit is in shutdown state;

[0118] S304. Traverse each record in P1 and check if there are other interpolation nodes in its corresponding time window. If there are, mark the record as invalid data and remove it from P1.

[0119] S305. Perform the same validity check on P2, remove time series containing additional interpolation nodes, and retain the valid subset of P2;

[0120] S306. Cut the valid subsets of P1 and P2 according to the time window W, ensuring that the time length of each cut segment is W and there are no interpolation nodes in the segment, to form the third initial parameter set P3;

[0121] S307. Select data segments from P3 under the unit shutdown state, extract the oxygen cell open circuit voltage U0 data, calculate the average value U0_avg of U0 within the same shutdown cycle, and use it as the model calibration benchmark value for that period.

[0122] S308. Establish a parameter set index table to record the time range, parameter types, and corresponding unit operating status of each parameter set in P1, P2, and P3.

[0123] Specifically, to prioritize retaining the oxygen cell open-circuit voltage data during the shutdown period as a model calibration benchmark, and to ensure the accuracy of the discharge fault inflection point prediction model, based on the initial parameter library to be analyzed and the interpolation nodes, a first initial parameter set and a second initial parameter set corresponding to the interpolation nodes are obtained. Alternatively, the first initial parameter set is the parameter data extracted within the first 60 seconds of the interpolation nodes, forming another first initial parameter set P1^new, where each record of P1^new contains a time series [t_k-60, t_k] and the corresponding continuous parameter value. The second initial parameter set is the parameter data extracted within the last 60 seconds of the interpolation nodes, forming another second initial parameter set P2^new, where each record of P2^new contains a time series [t_k, t_k+60] and the corresponding continuous parameter value.

[0124] Understandably, steps S301-S308 involve extracting parameter data corresponding to interpolation nodes from the initial parameter library DB to form a first initial parameter set P1, extracting data 30 seconds before and after the interpolation nodes to form a second initial parameter set P2, setting different preset time windows W (1 hour / 5 minutes) according to the unit's operating status (normal operation / shutdown), removing invalid data containing extra interpolation nodes from P1 and P2, extracting valid data according to window W to form a third initial parameter set P3, and calculating the average open-circuit voltage U0_avg as the model calibration benchmark by filtering data from the shutdown period. A parameter set index table was established to record the time range and operating status of each parameter set. This enabled the accurate extraction of effective parameter sets strongly correlated with fault analysis from the preprocessed parameter library, eliminating invalid information containing data interruptions. At the same time, the analysis window was set according to the differences in unit operating scenarios, clarifying the operating background corresponding to each parameter set. It also provided a voltage reference under shutdown conditions for subsequent model calibration. This provided a core dataset with scenario adaptation, data purity, and clear reference for building a discharge fault inflection point prediction model for S400, avoiding model prediction deviations caused by invalid data or insufficient scenario adaptation.

[0125] Extracting relevant data from interpolation nodes to form P1 and P2 allows for focusing on key nodes of data anomalies (such as before and after missing intervals), accurately capturing potential fault correlation characteristics. Time windows are set according to unit status differences: during normal operation, a 1-hour window covers sufficient data to reflect parameter trends; during shutdown, a 5-minute short window focuses on shutdown-specific parameters (such as open-circuit voltage), improving the targeting of parameter extraction. Invalid data containing additional interpolation nodes is removed, ensuring the continuity and reliability of P1, P2, and P3 data, avoiding data interruptions that could interfere with trend analysis. Using U0_avg at shutdown as the model calibration benchmark eliminates the influence of operating load on voltage, improving the model's accuracy in judging the true state of the battery. Establishing a parameter set index table facilitates rapid retrieval of parameter data under different operating conditions and time periods, improving the efficiency of model construction and fault analysis. These designs are fully adaptable to the differences in unit operation and shutdown conditions in gas analyzers (such as environmental monitoring scenarios), balancing data validity, analytical targeting, and operational convenience, providing crucial support for accurate fault early warning.

[0126] Furthermore, step S400 specifically includes:

[0127] S401. Extract oxygen cell voltage data U_f(t) from the first initial parameter set P1, calculate the voltage difference ΔU = U_f(t+1) - U_f(t) between adjacent time points, and calculate the oxygen cell voltage decay rate v_U = ΣΔU / 100 based on the cumulative value of ΔU over 100 consecutive sampling points, in units of V / second;

[0128] S402. Extract calibration deviation data ΔC_f(t) from the second initial parameter set P2, and calculate the oxygen measurement deviation growth rate v_ΔC=(ΔC_f(t_end)-ΔC_f(t_start)) / (t_end-t_start), in % / second, where t_start and t_end are the start and end times of the time series in P2;

[0129] S403. Extract temperature data T_f(t) from the third initial parameter set P3, combine it with oxygen cell voltage data U_f(t), and calculate the temperature influence coefficient k_T using a multiple linear regression method:

[0130] U_f(t)=a×T_f(t)+b×v_U+c, where k_T=a is the coefficient of temperature on voltage;

[0131] S404. Integrate oxygen cell voltage decay rate v_U, oxygen measurement deviation growth rate v_ΔC, and temperature influence coefficient k_T as input features to construct a discharge fault inflection point prediction dataset D, where each sample contains a feature vector [v_U,v_ΔC,k_T] and the corresponding actual fault occurrence time label t_fail;

[0132] S405. An LSTM neural network is used to construct a discharge fault inflection point prediction model. The model has an input layer dimension of 3, a hidden layer containing 2 layers of 64 neurons each, and an output layer of 1 neuron. The output value is the predicted fault inflection point time t_pred.

[0133] S406. Set the oxygen cell open-circuit voltage error threshold U0_avg_std, and use the oxygen cell open-circuit voltage U0_avg during the shutdown period as a reference to calibrate the model. Adjust the model parameters so that the error of the predicted voltage curve at U0_avg is ≤ U0_avg_std (U0_avg_std=0.05V).

[0134] S407. Use 70% of dataset D as the training set and 30% as the test set. Train the model using the backpropagation algorithm until the model's prediction error MAE on the test set is ≤24 hours.

[0135] S408. Save the parameters of the trained model, establish a model version management mechanism, and record the model training time, the range of datasets used, and the calibration benchmark values.

[0136] Specifically, steps S401-S408 involve extracting oxygen cell voltage data from the first initial parameter set P1 to calculate the voltage decay rate v_U, extracting calibration deviation data from the second initial parameter set P2 to calculate the deviation growth rate v_ΔC, extracting temperature and voltage data from the third initial parameter set P3 and using multiple linear regression to calculate the temperature influence coefficient k_T, integrating the three to construct a discharge fault inflection point prediction dataset D, using an LSTM neural network to construct a prediction model and calibrating the model with the average open-circuit voltage U0_avg during shutdown, dividing the training set and test set in a 7:3 ratio to train the model until the prediction error MAE ≤ 24 hours, saving the model parameters and establishing a version management mechanism. This process transforms the preprocessed parameters into core features that reflect the fault trend, constructs a fault inflection point prediction model adapted to the discharge characteristics of oxygen cells, and ensures the model's prediction accuracy through benchmark calibration and error control. This provides scientific and reliable model support for the subsequent accurate identification of oxygen cell voltage deficiency fault inflection points, avoiding fault prediction delays or misjudgments caused by missing model features or insufficient accuracy.

[0137] Voltage decay rate, deviation growth rate, and temperature influence coefficient were selected as model input features to accurately correlate oxygen cell state (voltage decay), analyzer performance (detection deviation), and environmental interference (temperature influence), comprehensively covering key factors in fault evolution. The use of an LSTM neural network effectively captures the temporal variation of parameters, adapting to the slow evolution of oxygen cell faults and providing more accurate predictions compared to traditional linear models. The model was calibrated using the average open-circuit voltage U0_avg during shutdown, eliminating the interference of operating load on voltage and ensuring the model predicts based on the actual health state of the battery. Strict control of the model's test set prediction error (MAE) to ≤24 hours ensures the time accuracy of fault inflection point prediction, allowing maintenance personnel sufficient preparation time for shutdown replacement. A model version management mechanism facilitates subsequent updates of model parameters based on new data, adapting to the characteristic differences of different batches of oxygen cells and extending the model's applicable lifespan. These designs not only meet the requirements of gas analyzers (such as those used in environmental monitoring scenarios) for accurate and timely fault warnings but also possess good scalability and adaptability.

[0138] Furthermore, step S500 specifically includes:

[0139] S501. Extract parameter data from the initial parameter library DB to be analyzed for the most recent 72 hours, and generate a first prediction parameter set P_pred that includes the new oxygen cell voltage decay rate v_U_new, the new oxygen measurement deviation growth rate v_ΔC_new, and the new temperature influence coefficient k_T_new according to the feature calculation method in step S400.

[0140] S502. Input P_pred into the trained discharge fault inflection point prediction model to obtain the oxygen battery voltage insufficient fault prediction inflection point t_pred. The oxygen battery voltage insufficient fault prediction inflection point is the time point when the oxygen battery voltage drops to the normal operating threshold U_min of the analyzer.

[0141] Specifically, the steps to input P_pred into the trained discharge fault inflection point prediction model to obtain the oxygen cell voltage deficiency fault prediction inflection point t_pred are as follows:

[0142] S502.1. From the first prediction parameter set P_pred, extract the three core input features that have been calculated: the rate of decay of the oxygen cell voltage v_U_new, the rate of change of the oxygen measurement deviation v_ΔC_new, and the coefficient of influence of the new temperature k_T_new, to form a feature vector matrix. The matrix format must be consistent with the input dimension during model training, that is, the feature vector of a single sample is [v_U_new, v_ΔC_new, k_T_new]. If P_pred contains multiple sets of time series data (such as 72 sets of features extracted at 1-hour intervals in the last 72 hours), then construct a 72×3 feature matrix (rows represent time series samples, and columns represent the three features);

[0143] S502.2. Call the feature standardization parameters stored during model training ("mean μ and standard deviation σ of training dataset features recorded by the version management mechanism in step S408") to standardize the feature vector of P_pred. The formula is as follows:

[0144] Standardized eigenvalue = (original eigenvalue - μ) / σ

[0145] For example, if during training, v_U has μ = 0.002 V / s and σ = 0.001 V / s, and a certain group in P_pred has v_U_new = 0.004 V / s, then the standardized value = (0.004 - 0.002) / 0.001 = 2, ensuring that the numerical range of the input features is consistent with the dataset during model training, and avoiding inference bias caused by differences in feature magnitude;

[0146] S502.3. If P_pred consists of multiple sets of time-series features (such as continuous features within 72 hours), the feature matrix must be arranged in the order of timestamps to ensure the temporal continuity of the features. Since the discharge fault inflection point prediction model is an LSTM neural network (step S405), its core logic depends on the correlation between the preceding and following time-series data. It is necessary to ensure that the row order of the input matrix is ​​consistent with the actual collection time order (such as t = 0 hours → t = 1 hour → … → t = 71 hours) to avoid the time sequence disorder affecting the model's judgment of the fault trend.

[0147] S502.4. From the model version management library (established in step S408), call the model file that matches the current analysis scenario (including the input layer, hidden layer, output layer weight parameters, bias terms, and error threshold after model calibration) to ensure that the loaded model is a valid model that has been calibrated by the furnace shutdown open circuit voltage U0_avg (step S406) and the test set prediction error MAE ≤ 24 hours (step S407), and avoid using uncalibrated or inaccurate models;

[0148] S502.5. Input the preprocessed feature matrix into the input layer of the LSTM model (3 dimensions, corresponding to 3 features), and the model performs inference according to the following logic:

[0149] The input layer passes the feature matrix to the first hidden layer (64 neurons), and calculates the output value of this layer through an activation function (such as ReLU) to capture the local correlation information of the features;

[0150] The output of the first hidden layer is passed to the second hidden layer (64 neurons). The gating mechanism of LSTM (input gate, forget gate, output gate) will filter and retain key temporal features (such as the continuous decay trend of v_U_new and the mutation features of v_ΔC_new) and forget irrelevant interference information.

[0151] The output of the second hidden layer is passed to the output layer (1 neuron), which outputs "the real-time voltage prediction sequence U_pred(t) of the oxygen cell in the future (e.g., the next 30 days)" through a linear activation function. Each time point in the sequence corresponds to a predicted voltage value (e.g., t = 1 day → U_pred = 1.2V, t = 2 days → U_pred = 1.18V...).

[0152] S502.6. Smooth the U_pred(t) sequence output by the model using the same exponential weighted average method as in step S203 (smoothing coefficient α = 0.3~0.5), the formula is as follows:

[0153] U_pred_smooth(t)=α×U_pred(t)+(1-α)×U_pred_smooth(t-1)

[0154] Eliminate the instantaneous voltage prediction fluctuations that may occur during model inference to obtain a smoother voltage prediction curve that better reflects the actual decay law of the oxygen cell;

[0155] S503.7. Obtain the minimum voltage threshold U_min for the oxygen cell to maintain normal operation of the analyzer from the hardware specifications or system preset parameters of the gas analyzer (e.g., for the oxygen cell of the AO2020 instrument, U_min is usually set to 1.0V) - this threshold is the core criterion for judging "insufficient voltage fault" and must be consistent with the fault label definition during model training in step S405 (i.e., the "actual fault occurrence time t_fail" marked during training, corresponding to the time point when the voltage first drops below U_min);

[0156] S503.8. Traverse the smoothed voltage prediction curve U_pred_smooth(t) and find the time point in the curve where "U_pred_smooth(t) ≤ U_min" is first satisfied. This time point is the preliminary oxygen cell voltage deficiency fault prediction inflection point t_pred. For example, if U_pred_smooth = 1.0V (equal to U_min) at t = 15 days and U_pred_smooth = 0.98V (lower than U_min) at t = 16 days, then t_pred is determined to be t = 15 days.

[0157] S503.9. Combining the open-circuit voltage reference value U0_avg retained in step S307 during the shutdown period, verify the rationality of t_pred—calculate the deviation rate between the predicted voltage U_pred_smooth(t_pred) and U0_avg at time t_pred, using the following formula:

[0158] Deviation rate=|U_pred_smooth(t_pred)-U0_avg×0.8| / (U0_avg×0.8)×100%

[0159] (Note: U0_avg×0.8 is the reference voltage for the oxygen cell to degrade to 80% health level under shutdown conditions, which can reflect the consistency of the cell degradation trend.)

[0160] If the deviation rate is ≤5%, then t_pred is considered valid; if the deviation rate is >5%, then return to the model inference stage, readjust the hidden layer weight parameters of the LSTM model (based on the calibration logic of U0_avg), and execute voltage prediction and inflection point determination again until the deviation rate meets the requirements.

[0161] S502.10. Convert the valid t_pred to the standard time format of "year-month-day hour:minute:second" (e.g., 2024-12-30 08:00:00), and associate it with the current prediction execution time t_now (e.g., 2024-12-15 08:00:00). Calculate the time difference Δt_now between t_pred and t_now (e.g., Δt_now = 15 days), providing a time difference basis for subsequent graded early warning (steps S602-S603).

[0162] S502.11. Store t_pred, the corresponding U_pred_smooth(t_pred), Δt_now, and the P_pred feature vector of the input model together in the "Fault Prediction Result Table" of the initial parameter library DB to be analyzed (constructed in step S206), and mark the model version number (recorded in step S408) and the validation deviation rate to facilitate the subsequent tracking of the effectiveness of the prediction process, and at the same time provide actual prediction data support for model iterative optimization (such as the version update in step S408).

[0163] Understandably, by extracting parameter data from the initial parameter library DB of the most recent 72 hours, and following the characteristic calculation method of "calculating voltage decay rate, deviation growth rate, and temperature influence coefficient" in step S400, a first prediction parameter set P_pred is generated, which includes the new oxygen battery voltage decay rate v_U_new, the new oxygen measurement deviation growth rate v_ΔC_new, and the new temperature influence coefficient k_T_new. Then, P_pred is input into the trained discharge fault inflection point prediction model to obtain the oxygen battery insufficient voltage fault prediction inflection point t_pred (defined as the time point when the oxygen battery voltage drops to the analyzer's normal operating threshold U_min). This realizes the transformation of preprocessed historical parameter data into accurate prediction of future fault time, establishing a key link of "historical parameter analysis → future fault prediction". This provides a clear time anchor for subsequent generation of graded early warning and matching of operation and maintenance plans based on t_pred, avoiding passive operation and maintenance caused by the inability to predict fault time.

[0164] P_pred is generated by selecting parameter data from the most recent 72 hours. This ensures sufficient coverage to reflect the dynamic trends of parameter changes (such as the continuous pattern of voltage decay and the continuous influence of temperature on voltage) while avoiding outdated interference introduced by excessively long data (such as large differences between early parameters and the current battery state), thus guaranteeing the timeliness and relevance of the predicted features. The feature calculation method in step S400 is strictly followed to ensure that the feature format and calculation logic of P_pred are completely consistent with the dataset used for model training, eliminating model input bias caused by differences in feature calculation and improving prediction accuracy. t_pred is explicitly defined as the "time point when the voltage drops to U_min," directly related to the critical state where the oxygen battery cannot support the normal operation of the analyzer. This strongly links the prediction results to the actual fault impact, rather than a vague state judgment. Maintenance personnel can accurately plan the shutdown and replacement timing based on this time point, avoiding unplanned downtime and preventing distortion of environmental monitoring data due to untimely replacement. This fully meets the needs of gas analyzers for accurate and practical fault prediction in environmental monitoring scenarios.

[0165] Furthermore, step S600 specifically includes:

[0166] S601. Obtain the next planned shutdown time t_stop from the unit operation and maintenance system, and calculate the difference between the predicted inflection point and the shutdown time Δt_stop=t_stop-t_pred;

[0167] S602. When Δt_stop<0, that is, the predicted inflection point is earlier than the planned shutdown time, a level one warning is triggered. Among them, when the difference between t_pred and the current time t_now, Δt_now=t_pred-t_now<7 days, the warning level is level one.

[0168] S603. When 7 days ≤ Δt_now ≤ 30 days, the warning level is Level II, indicating that the oxygen cell should be replaced during the next shutdown.

[0169] When a Level 1 warning is triggered, the red LED on the control analyzer flashes at a frequency of 1Hz and sends an alarm message containing the predicted inflection point time and the current voltage value to the remote operation and maintenance platform through the communication module.

[0170] When a Level 2 warning is triggered, the yellow LED light is controlled to flash at a frequency of 0.5Hz, and the warning record is stored locally, including information such as the warning time, the predicted inflection point, and the current calibration deviation.

[0171] When Δt_stop≥0, meaning the predicted inflection point is later than the planned shutdown time, no warning is triggered, but the model prediction results are updated, and the predicted inflection point is recalculated in the next parameter acquisition cycle.

[0172] Specifically, by obtaining the next planned shutdown time t_stop from the unit operation and maintenance system and calculating the difference Δt_stop between the predicted inflection point and the shutdown time, when Δt_stop < 0 (the predicted inflection point is earlier than the shutdown time), a first-level warning (Δt_now < 7 days) and a second-level warning (7 days ≤ Δt_now ≤ 30 days) are distinguished based on the difference Δt_now between t_pred and the current time t_now. A first-level warning triggers a red LED to flash at 1Hz and sends an alarm message containing the predicted inflection point and the current voltage to the remote platform. A second-level warning triggers a yellow LED to flash at 0.5Hz and stores a record containing the warning time, the predicted inflection point, and the current calibration deviation locally. When Δt_stop ≥ 0, no warning is triggered, but the model prediction results are updated. This achieves deep linkage between fault warning and unit operation and maintenance plan, transforming fault prediction into actionable operation and maintenance guidelines. This avoids unplanned shutdowns for maintenance and ensures that oxygen cells are replaced within the shutdown window, guaranteeing continuous and accurate environmental monitoring data from the gas analyzer during normal unit operation.

[0173] Based on Δt_stop, the system determines the necessity of early warnings, accurately matching shutdown plans with fault risks. This avoids invalid warnings when the predicted inflection point is later than the shutdown time, reducing interference for maintenance personnel. Early warnings are tiered by Δt_now. Level 1 warnings use a dual alert system (red LED) and remote push notifications to ensure rapid response to emergency risks (faults within 7 days). Level 2 warnings use yellow LEDs and local recordings to provide advance planning for shutdown and replacement, meeting the differentiated handling needs of different risk levels. Warning information includes key data such as the predicted inflection point time, current voltage value, and calibration deviation. Maintenance personnel can grasp the oxygen cell status and fault trends without disassembling the meter, lowering the maintenance threshold. When Δt_stop ≥ 0, the model prediction results are updated and recalculated in the next data collection cycle, dynamically tracking cell degradation changes and avoiding risk omissions due to single prediction errors. This comprehensively adapts to the core needs of power generation companies and other scenarios for continuous operation and efficient maintenance of environmental monitoring equipment.

[0174] Furthermore, a gas analyzer fault early warning system based on battery state data is proposed to implement the early warning method described above, including:

[0175] The data acquisition module is used to acquire real-time terminal voltage, open-circuit voltage, temperature and cumulative charge-discharge cycle count of the oxygen cell to obtain essential battery parameters. It is also used to acquire auxiliary analyzer parameters, including the working mode of the gas analyzer, the oxygen measurement value and the calibration deviation of the standard gas, and the sensor output voltage signal.

[0176] The data processing module is used to preprocess the mandatory battery parameters and auxiliary analyzer parameters respectively, construct an initial parameter library to be analyzed, obtain the time nodes corresponding to outliers and missing values ​​during the data preprocessing process as interpolation nodes, and mark the time intervals corresponding to the unit's operating status. Based on the initial parameter library to be analyzed and the interpolation nodes, it obtains the first initial parameter set and the second initial parameter set corresponding to the interpolation nodes. Simultaneously, according to a preset time window, it determines whether there are additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, the first initial parameter set and / or the second initial parameter set with additional interpolation nodes are... If the parameter set is not removed, the first initial parameter set and / or the second initial parameter set are truncated according to a preset time window to obtain the third initial parameter set. It is also used to obtain the discharge fault inflection point prediction dataset based on the first initial parameter set, the second initial parameter set and the third initial parameter set, combined with the oxygen battery discharge characteristics of the gas analyzer, and to construct the discharge fault inflection point prediction model. The model input features include the oxygen battery voltage decay rate, the oxygen measurement value deviation growth rate and the temperature influence coefficient. It is used to generate the first prediction parameter set according to the initial parameter library to be analyzed, and import the first prediction parameter set into the discharge fault inflection point prediction model to obtain the oxygen battery insufficient voltage fault prediction inflection point.

[0177] The communication module is used to generate alarm information based on the inflection point of oxygen cell voltage deficiency fault prediction, and send alarm information including the predicted inflection point time and the current voltage value to the remote operation and maintenance platform.

[0178] The early warning module is used to issue early warning signals based on alarm information.

[0179] Specifically, according to the alarm information, issuing an early warning signal means that when a level one early warning is triggered, the red LED on the control analyzer flashes at a frequency of 1Hz and sends an alarm message containing the predicted inflection point time and the current voltage value to the remote operation and maintenance platform through the communication module.

[0180] When a Level 2 warning is triggered, the yellow LED light is controlled to flash at a frequency of 0.5Hz, and the warning record is stored locally, including information such as the warning time, the predicted inflection point, and the current calibration deviation.

[0181] Furthermore, the data acquisition module includes:

[0182] The first acquisition unit is used to acquire the real-time terminal voltage, open-circuit voltage, battery temperature and cumulative charge-discharge cycle number of the oxygen battery to obtain the essential battery parameters.

[0183] The second acquisition unit is used to acquire auxiliary analyzer parameters, including the gas analyzer's operating mode, oxygen measurement value and standard gas calibration deviation, and sensor output voltage signal.

[0184] Furthermore, the data processing module includes:

[0185] The preprocessing unit is used to preprocess the parameters of the required battery and the parameters of the auxiliary analyzer respectively, and to build an initial parameter library to be analyzed.

[0186] The marking unit is used to obtain the time nodes corresponding to outliers and missing values ​​in the data preprocessing process as interpolation nodes, and to mark the time intervals corresponding to the unit's operating status.

[0187] The filtering unit is used to obtain the first initial parameter set and the second initial parameter set corresponding to the interpolation node based on the initial parameter library to be analyzed and the interpolation node. Simultaneously, according to a preset time window, it determines whether there are any additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, the first initial parameter set and / or the second initial parameter set with additional interpolation nodes are removed. If not, the first initial parameter set and / or the second initial parameter set are truncated according to the preset time window to obtain the third initial parameter set.

[0188] The model building unit is used to obtain a discharge fault inflection point prediction dataset based on the first initial parameter set, the second initial parameter set, and the third initial parameter set, combined with the oxygen cell discharge characteristics of the gas analyzer, and to build a discharge fault inflection point prediction model. The model input features include the oxygen cell voltage decay rate, the oxygen measurement value deviation growth rate, and the temperature influence coefficient.

[0189] The fault prediction unit is used to generate a first prediction parameter set based on the initial parameter library to be analyzed, and import the first prediction parameter set into the discharge fault inflection point prediction model to obtain the fault prediction inflection point of insufficient oxygen battery voltage.

[0190] The advantages of this invention are as follows: The gas analyzer fault early warning method and system based on battery status data proposed in this article have the overall advantage of taking oxygen battery status monitoring as the core. Through full-dimensional parameter acquisition (covering battery voltage, temperature, cycle life and analyzer working mode, detection deviation, etc.), scenario-based data preprocessing (optimizing parameter processing logic in combination with unit operation / shutdown status), precise model construction (building a discharge fault inflection point prediction model with LSTM neural network and calibrating through shutdown open circuit voltage), and operation and maintenance linkage hierarchical early warning (generating hierarchical early warnings in association with planned shutdown time and adapting to local LED prompts and remote push), a complete closed loop of "parameter acquisition - data processing - model prediction - early warning operation and maintenance" is formed. This not only solves the problems of early warning lag caused by reliance on manual maintenance in traditional methods, easy misjudgment due to lack of combination with unit status, and lack of operation and maintenance linkage, but also can accurately predict oxygen battery voltage deficiency faults in advance, guide operation and maintenance personnel to complete replacement during the shutdown window, effectively reduce unplanned downtime, and ensure the continuous and accurate operation of gas analyzers in environmental monitoring and other scenarios. It has both technological innovation, scenario adaptability and practical operation and maintenance value.

[0191] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for early warning of faults in a gas analyzer based on battery state data, characterized in that, include: S100. Collects real-time terminal voltage, open-circuit voltage, temperature and cumulative charge-discharge cycle count of oxygen cell to obtain essential battery parameters. Simultaneously collects auxiliary analyzer parameters, including the working mode of the gas analyzer, the oxygen measurement value and the standard gas calibration deviation, and the sensor output voltage signal. S200. Perform data preprocessing on the mandatory battery parameters and auxiliary analyzer parameters respectively, construct the initial parameter library to be analyzed, synchronously obtain the time nodes corresponding to the outliers and missing values ​​in the data preprocessing process as interpolation nodes, and mark the time interval corresponding to the unit's operating status. S300. Based on the initial parameter library to be analyzed and the interpolation nodes, obtain the first initial parameter set and the second initial parameter set corresponding to the interpolation nodes. Simultaneously, according to a preset time window, determine whether there are any additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, remove the first initial parameter set and / or the second initial parameter set with additional interpolation nodes. If not, truncate the first initial parameter set and / or the second initial parameter set according to the preset time window to obtain the third initial parameter set. S400. Based on the first initial parameter set, the second initial parameter set, and the third initial parameter set, combined with the oxygen cell discharge characteristics of the gas analyzer, obtain the discharge fault inflection point prediction dataset, and construct the discharge fault inflection point prediction model. The model input features include the oxygen cell voltage decay rate, the oxygen measurement value deviation growth rate, and the temperature influence coefficient. S500. Based on the initial parameter library to be analyzed, generate the first prediction parameter set, import the first prediction parameter set into the discharge fault inflection point prediction model, and obtain the oxygen battery voltage insufficient fault prediction inflection point. S600 generates alarm information based on the inflection point of oxygen cell voltage deficiency fault prediction.

2. The gas analyzer fault early warning method based on battery status data according to claim 1, characterized in that, Step S100 specifically includes: S101. Deploy the voltage acquisition module and connect it to the positive and negative terminals of the oxygen cell. Set the acquisition frequency to 1Hz and record the oxygen cell terminal voltage data U(t) in real time, where t is the acquisition timestamp in seconds. S102. When the unit is shut down, disconnect the oxygen cell from the analyzer load by switching the switch, collect the oxygen cell open circuit voltage U0 for 5 minutes, and take the average voltage during this period as the open circuit voltage reference value under the shutdown state. S103. Attach a temperature sensor to the surface of the oxygen battery casing, set the acquisition frequency to 0.1Hz, record the battery temperature data T(t), and synchronously associate it with the corresponding voltage acquisition timestamp; S104. Record the cumulative working time t_total of the oxygen battery from installation and activation to the current moment through the counter, and calculate the cumulative number of charge and discharge cycles N by combining the working mode conversion records of the analyzer. A single cycle is defined as a complete working cycle from the start-up of the analyzer to the shutdown. S105. Read the operating mode signal M(t) from the control system of the gas analyzer, where M(t) = 1 indicates continuous monitoring mode and M(t) = 0 indicates intermittent calibration mode; S106. Acquire the oxygen measurement value C(t) output by the analyzer, and simultaneously obtain the standard gas concentration value C_std introduced at the same time, and calculate the real-time calibration deviation ΔC(t)=|C(t)-C_std| / C_std×100%; S107. Acquire the raw voltage signal V_s(t) output by the sensor. This signal is the raw response voltage of the oxygen cell when it participates in the electrochemical reaction. It forms a correlation and comparison data with the oxygen cell terminal voltage U(t). S108. Align the collected U(t), U0, T(t), N, M(t), ΔC(t), and V_s(t) in steps S101-S107 according to the timestamps, and store them in the real-time database to form the original record of parameter collection.

3. The gas analyzer fault early warning method based on battery status data according to claim 1, characterized in that, Step S200 specifically includes: S201. Noise filtering is applied to the real-time terminal voltage U(t) of the oxygen cell in the mandatory battery parameters. The sliding window averaging method is used, with a window size of 5 sampling points. The filtered voltage value is then calculated. U_f(t)=[U(t-4)+U(t-3)+U(t-2)+U(t-1)+U(t)] / 5, where t≥4; S202. Perform outlier detection on battery temperature data T(t), set a normal temperature range, and mark T(t) as an outlier when it exceeds the normal temperature range, and record the corresponding time node t_err; S203. The real-time calibration deviation ΔC(t) in the auxiliary analyzer parameters is smoothed by using the exponential weighted average method to calculate the smoothed deviation value ΔC_f(t)=α×ΔC(t)+(1-α)×ΔC_f(t-1), where α is the smoothing coefficient; S204. Set a missing duration threshold Δt_std. For all missing values ​​in the mandatory battery parameters and auxiliary analyzer parameters, determine the missing duration Δt. When Δt ≤ Δt_std, use linear interpolation to fill the missing value. That is, for the missing point t_m, the filled value is: X(t_m)=X(t_m-1)+[X(t_m+1)-X(t_m-1)]×(t_m-t_m-1) / (t_m+1-t_m-1); S205. When the missing duration Δt > Δt_std, mark the time period as a data interruption interval and record the start time t_start and end time t_end of the interval as interpolation nodes; S206. Integrate the filtered, smoothed, and interpolated parameter data to construct the initial parameter library DB to be analyzed. Each record in the library contains a timestamp t and the corresponding U_f(t), T_f(t), N, M(t), ΔC_f(t), and V_s_f(t). S207. Obtain the unit operating status signal S(t) from the unit control system, where S(t) = 1 indicates normal operation and S(t) = 0 indicates shutdown. Associate S(t) with the timestamp in the parameter library DB to mark the time interval of the unit operating status corresponding to each parameter. S208. Statistically calculate the number of outliers N_err and the number of missing value intervals N_miss generated during the preprocessing process.

4. The gas analyzer fault early warning method based on battery status data according to claim 1, characterized in that, Step S300 specifically includes: S301. Extract parameter data corresponding to all interpolation nodes from the initial parameter library DB to be analyzed, forming the first initial parameter set P1, where each record in P1 contains the interpolation node time t_k and the corresponding parameter value set {U_f(t_k),T_f(t_k),ΔC_f(t_k)}; S302. Extract parameter data within a 30-second range before and after the interpolation node to form a second initial parameter set P2, where each record in P2 contains a time series [t_k-30, t_k+30] and the corresponding continuous parameter values; S303. Set a preset time window W, where W = 3600 seconds when the unit is in normal operation and W = 300 seconds when it is in shutdown state; S304. Traverse each record in P1 and check if there are other interpolation nodes in its corresponding time window. If there are, mark the record as invalid data and remove it from P1. S305. Perform the same validity check on P2, remove time series containing additional interpolation nodes, and retain the valid subset of P2; S306. Cut the valid subsets of P1 and P2 according to the time window W, ensuring that the time length of each cut segment is W and there are no interpolation nodes in the segment, to form the third initial parameter set P3; S307. Select data segments from P3 under the unit shutdown state, extract the oxygen cell open circuit voltage U0 data, calculate the average value U0_avg of U0 within the same shutdown cycle, and use it as the model calibration benchmark value for that period. S308. Establish a parameter set index table to record the time range, parameter types, and corresponding unit operating status of each parameter set in P1, P2, and P3.

5. The gas analyzer fault early warning method based on battery status data according to claim 1, characterized in that, Step S400 specifically includes: S401. Extract oxygen cell voltage data U_f(t) from the first initial parameter set P1, calculate the voltage difference ΔU = U_f(t+1) - U_f(t) between adjacent time points, and calculate the oxygen cell voltage decay rate v_U = ΣΔU / 100 based on the cumulative value of ΔU over 100 consecutive sampling points, in units of V / second; S402. Extract calibration deviation data ΔC_f(t) from the second initial parameter set P2, and calculate the oxygen measurement deviation growth rate v_ΔC=(ΔC_f(t_end)-ΔC_f(t_start)) / (t_end-t_start), in % / second, where t_start and t_end are the start and end times of the time series in P2; S403. Extract temperature data T_f(t) from the third initial parameter set P3, combine it with oxygen cell voltage data U_f(t), and calculate the temperature influence coefficient k_T using a multiple linear regression method: U_f(t)=a×T_f(t)+b×v_U+c, where k_T=a is the coefficient of temperature on voltage; S404. Integrate oxygen cell voltage decay rate v_U, oxygen measurement deviation growth rate v_ΔC, and temperature influence coefficient k_T as input features to construct a discharge fault inflection point prediction dataset D, where each sample contains a feature vector [v_U,v_ΔC,k_T] and the corresponding actual fault occurrence time label t_fail; S405. An LSTM neural network is used to construct a discharge fault inflection point prediction model, and the output value is the predicted fault inflection point time t_pred; S406. Set the oxygen cell open-circuit voltage error threshold U0_avg_std, use the oxygen cell open-circuit voltage U0_avg during shutdown as a benchmark, calibrate the model, and adjust the model parameters so that the error of the predicted voltage curve at U0_avg is ≤U0_avg_std. S407. Use 70% of dataset D as the training set and 30% as the test set. Train the model using the backpropagation algorithm until the model's prediction error MAE on the test set is ≤24 hours. S408. Save the parameters of the trained model, establish a model version management mechanism, and record the model training time, the range of the dataset used, and the calibration benchmark value.

6. The gas analyzer fault early warning method based on battery status data according to claim 1, characterized in that, Step S500 specifically includes: S501. Extract parameter data from the initial parameter library DB to be analyzed for the most recent 72 hours, and generate a first prediction parameter set P_pred that includes the new oxygen cell voltage decay rate v_U_new, the new oxygen measurement deviation growth rate v_ΔC_new, and the new temperature influence coefficient k_T_new according to the feature calculation method in step S400. S502. Input P_pred into the trained discharge fault inflection point prediction model to obtain the oxygen battery voltage insufficient fault prediction inflection point t_pred. The oxygen battery voltage insufficient fault prediction inflection point is the time point when the oxygen battery voltage drops to the normal operating threshold U_min of the analyzer.

7. The gas analyzer fault early warning method based on battery status data according to claim 1, characterized in that, Step S600 specifically includes: S601. Obtain the next planned shutdown time t_stop from the unit operation and maintenance system, and calculate the difference between the predicted inflection point and the shutdown time Δt_stop=t_stop-t_pred; S602. When Δt_stop<0, that is, the predicted inflection point is earlier than the planned shutdown time, a level one warning is triggered. Among them, when the difference between t_pred and the current time t_now, Δt_now=t_pred-t_now<7 days, the warning level is level one. S603. When 7 days ≤ Δt_now ≤ 30 days, the warning level is Level II, indicating that the oxygen cell should be replaced during the next shutdown.

8. A gas analyzer fault early warning system based on battery state data, used to implement the early warning method as described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire the real-time terminal voltage, open-circuit voltage, temperature and cumulative charge-discharge cycle number of the oxygen cell to obtain the essential battery parameters. It is also used to acquire auxiliary analyzer parameters, including the working mode of the gas analyzer, the oxygen measurement value and the calibration deviation of the standard gas, and the sensor output voltage signal. The data processing module is used to preprocess the mandatory battery parameters and auxiliary analyzer parameters respectively, construct an initial parameter library to be analyzed, obtain the time nodes corresponding to outliers and missing values ​​in the data preprocessing process as interpolation nodes, and mark the time intervals corresponding to the unit operating status. Based on the initial parameter library to be analyzed and the interpolation nodes, it obtains the first initial parameter set and the second initial parameter set corresponding to the interpolation nodes, and simultaneously determines whether there are additional interpolation nodes in the first initial parameter set and / or the second initial parameter set according to a preset time window. If so, the first initial parameter set and / or the second initial parameter set with additional interpolation nodes are then processed. If the parameter set is removed, the first initial parameter set and / or the second initial parameter set are truncated according to a preset time window to obtain the third initial parameter set. It is also used to obtain the discharge fault inflection point prediction dataset based on the first initial parameter set, the second initial parameter set and the third initial parameter set, combined with the oxygen battery discharge characteristics of the gas analyzer, and to construct the discharge fault inflection point prediction model. The model input features include the oxygen battery voltage decay rate, the oxygen measurement value deviation growth rate and the temperature influence coefficient. It is used to generate the first prediction parameter set according to the initial parameter library to be analyzed, and import the first prediction parameter set into the discharge fault inflection point prediction model to obtain the oxygen battery insufficient voltage fault prediction inflection point. The communication module is used to generate alarm information based on the predicted inflection point of insufficient oxygen cell voltage fault, and send alarm information including the predicted inflection point time and the current voltage value to the remote operation and maintenance platform. The early warning module is used to issue early warning signals based on alarm information.

9. A gas analyzer fault early warning system based on battery status data according to claim 8, characterized in that, The acquisition module includes: The first acquisition unit is used to acquire the real-time terminal voltage, open-circuit voltage, battery temperature and cumulative charge-discharge cycle number of the oxygen battery to obtain the essential battery parameters. The second acquisition unit is used to acquire auxiliary analyzer parameters, including the gas analyzer's operating mode, oxygen measurement value and standard gas calibration deviation, and sensor output voltage signal.

10. A gas analyzer fault early warning system based on battery state data according to claim 8, characterized in that, The data processing module includes: The preprocessing unit is used to preprocess the parameters of the required battery and the parameters of the auxiliary analyzer respectively, and to build an initial parameter library to be analyzed; The marking unit is used to obtain the time nodes corresponding to outliers and missing values ​​in the data preprocessing process as interpolation nodes, and to mark the time intervals corresponding to the unit's operating status. The filtering unit is used to obtain the first initial parameter set and the second initial parameter set corresponding to the interpolation node based on the initial parameter library to be analyzed and the interpolation node. Simultaneously, according to a preset time window, it determines whether there are any additional interpolation nodes in the first initial parameter set and / or the second initial parameter set. If so, the first initial parameter set and / or the second initial parameter set with additional interpolation nodes are removed. If not, the first initial parameter set and / or the second initial parameter set are truncated according to the preset time window to obtain the third initial parameter set. The model building unit is used to obtain a discharge fault inflection point prediction dataset based on the first initial parameter set, the second initial parameter set, and the third initial parameter set, combined with the oxygen cell discharge characteristics of the gas analyzer, and to build a discharge fault inflection point prediction model. The model input features include the oxygen cell voltage decay rate, the oxygen measurement value deviation growth rate, and the temperature influence coefficient. The fault prediction unit is used to generate a first prediction parameter set based on the initial parameter library to be analyzed, and to import the first prediction parameter set into the discharge fault inflection point prediction model to obtain the oxygen battery voltage insufficient fault prediction inflection point.

Images