Online monitoring method for conversion of hydrochloric acid from carbon tetrachloride

Abstract

The invention belongs to the technical field of conversion process monitoring, and provides an on-line monitoring method for conversion of hydrochloric acid from carbon tetrachloride, which comprises the following steps: acquiring historical data of a conversion process in a steady state and operation under various preset abnormal working conditions, and constructing time sequence data including trichloromethane concentration and process parameters and a corresponding state label historical data set; performing statistical feature extraction on the historical data set, and establishing a correlation model between the statistical features of the trichloromethane concentration and the process parameter state based on a machine learning algorithm; the method comprises the following steps: detecting the concentration of trichloromethane in a reactor outlet product in real time through an online analysis instrument, calculating the real-time statistical characteristic of the real-time trichloromethane concentration as the input of a correlation model, and comparing the output process parameter monitoring combination with the currently activated process parameters to generate a control instruction. By monitoring a key reaction intermediate (trichloromethane), the number of monitored process parameters is optimized, and the preposition of abnormal early warning can also be realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of conversion process monitoring, in particular to an online monitoring method for hydrochloric acid converted from carbon tetrachloride. BACKGROUND

[0002] Carbon tetrachloride is a controlled ozone-depleting substance, and its harmless disposal and resource utilization are important issues in the environmental protection field. Among them, converting carbon tetrachloride into hydrochloric acid through hydrolysis or chlorination process is a key treatment technology. The conversion process is essentially a complex step-by-step substitution reaction, and the chlorine atom is gradually replaced by the hydrogen atom under the condition of the presence of water vapor and heating.

[0003] At present, the monitoring of the conversion process of carbon tetrachloride into hydrochloric acid mainly relies on the all-around and fixed mode monitoring of multiple key process parameters (such as temperature, pressure, flow rate, etc.) and the concentration of the final product (such as HCl). However, this monitoring method has obvious disadvantages. First, the monitoring parameters are numerous, resulting in a complex system and cumbersome maintenance. Second, this is a passive and lagging monitoring mode. The concentration change of the final product HCl is the comprehensive result of all the preceding reaction steps. When it is abnormal, it indicates that the process has deviated for a period of time, causing raw material waste or product unqualification, and the operator needs time to troubleshoot the root cause from a large number of parameters, which is slow in response.

[0004] Therefore, the application provides an online monitoring method for hydrochloric acid converted from carbon tetrachloride. SUMMARY

[0005] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background art.

[0006] The technical scheme adopted by the application to solve its technical problems is: an online monitoring method for hydrochloric acid converted from carbon tetrachloride, the conversion product of carbon tetrachloride into hydrochloric acid includes chloroform, comprising the following steps:

[0007] Obtain the historical data of the steady state and multiple preset abnormal working conditions of the conversion process, and construct a time series data set including the concentration of chloroform, process parameters and corresponding state labels;

[0008] Statistical feature extraction is performed on the historical data set, and a correlation model between the statistical features of the concentration of chloroform and the state of the process parameters is established based on a machine learning algorithm;

[0009] The concentration of chloroform in the reactor outlet product is detected in real time by an online analysis instrument, the real-time statistical features of the real-time concentration of chloroform are calculated as the input of the correlation model, the output process parameter monitoring combination is compared with the currently activated process parameter, and a control instruction is generated;

[0010] According to the real-time concentration of chloroform and the real-time monitoring data of process parameters, the fusion analysis of multiple process parameters is carried out to judge whether the conversion process is in an abnormal state;

[0011] If the conversion process is in an abnormal state, output the corresponding abnormal state type, and execute the preset matching control operation based on the abnormal state type.

[0012] The beneficial effects of the present application are as follows:

[0013] The present application optimizes the number of monitored process parameters by monitoring the key reaction intermediate (chloroform), and realizes the preposition of abnormal early warning;

[0014] The present application collects time series data under steady state and various preset abnormal conditions to mark state labels, and constructs a historical data set covering all scenarios. On the one hand, the unique optimal process parameter monitoring combination under each condition is selected through root cause tracing and effectiveness scoring, which optimizes the cumbersome problem of traditional full parameter monitoring. On the other hand, the multi-dimensional statistical features of chloroform concentration are extracted to realize the unique identification of the reactor operating state. Then, the multi-dimensional features are taken as input and the optimal monitoring combination is taken as output to train and verify the correlation model. Finally, the qualified model is integrated into the online monitoring system, thereby establishing a deterministic mapping of chloroform features and optimal monitoring combination, laying a foundation for subsequent reduction of invalid monitoring and reduction of resource consumption, and providing core algorithm and data support for real-time monitoring automation decision-making;

[0015] The present application collects chloroform concentration data at the reactor outlet in real time through an online analysis instrument, generates real-time multi-dimensional feature vectors and inputs them into the correlation model, obtains the optimal process parameter monitoring combination corresponding to the current condition, and compares it with the currently activated monitoring parameters to generate control instructions, thereby ensuring that the monitoring strategy can be dynamically adjusted according to the real-time state of the reaction, reducing the cumbersome and resource waste of traditional fixed full parameter monitoring, and providing real-time monitoring data for subsequent abnormal state judgment;

[0016] The present application calculates the instantaneous value and short-term change trend of each data through real-time concentration of chloroform and dynamic monitoring data of process parameters, and judges whether the conversion process is abnormal and the specific abnormal type through the double diagnosis mechanism of preliminary abnormal assumption and multi-source evidence verification, thereby reducing the hysteresis of monitoring only relying on instantaneous value and reducing the risk of misjudgment, and providing accurate and reliable abnormal state basis for executing targeted control operation. BRIEF DESCRIPTION OF DRAWINGS

[0017] The present application will be further described below with reference to the accompanying drawings.

[0018] Figure 1 is a flow chart of the steps of the online monitoring method for carbon tetrachloride conversion hydrochloric acid of the present application;

[0019] Figure 2 is a part of the flow chart of step one of the online monitoring method of carbon tetrachloride conversion hydrochloric acid according to the present application;

[0020] Figure 3 is a module architecture diagram of the online monitoring system of carbon tetrachloride conversion hydrochloric acid according to the present application. DETAILED DESCRIPTION

[0021] In order to make the technical means, creative features, purposes and effects of the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0022] Embodiment 1

[0023] Please refer to Figure 1 and Figure 2 , the online monitoring method of carbon tetrachloride conversion hydrochloric acid according to the present application embodiment, mainly aiming at the problem of numerous monitored process parameters and abnormal early warning lag in the process of carbon tetrachloride conversion hydrochloric acid, by selecting the key reaction intermediates for monitoring and constructing dynamic monitoring strategy, effectively reducing the number of process parameters that need to be continuously monitored, and through capturing the early concentration change of the reaction intermediates and monitoring results, realizing the pre-warning of the abnormal conversion process, specifically including the following steps:

[0024] First of all, it needs to be pointed out that since carbon tetrachloride is converted into hydrochloric acid, it is a step-by-step substitution process, usually in the presence of water vapor and under heating conditions, the chlorine atom is gradually replaced by hydrogen atom:

[0025] CCl4+H2O→CHCl3(three chloromethane)+HCl+other oxygen-containing organic matter;

[0026] CHCl3+H2O→CH2Cl2(dichloromethane)+HCl+other oxygen-containing organic matter;

[0027] CH2Cl2+H2O→CH3Cl(chloromethane)+HCl+other oxygen-containing organic matter;

[0028] CH3Cl+H2O→CH4(methane)+HCl+other oxygen-containing organic matter;

[0029] This embodiment selects three chloromethane as the core monitoring reaction intermediate, because:

[0030] Firstly, three chloromethane is the direct product of the first step reaction, which means that any fluctuation of process parameters will be most directly and most rapidly reflected in the generation rate and concentration change of three chloromethane;

[0031] Secondly, chloroform is both the product of CCl4 conversion and the reactant of the subsequent intermediate dichloromethane, so its concentration level directly reflects the dynamic balance between the first step reaction rate and the subsequent reaction rate. Stable concentration means that the entire reaction chain is in a stable state; rising concentration clearly indicates that the reaction has accumulated at the "chloroform" node, indicating that the reaction path is blocked or the conditions are unbalanced.

[0032] Thirdly, chloroform has sufficient chemical stability to ensure the reliability and reproducibility of monitoring data.

[0033] Step one: Obtain historical data of the conversion process under stable state and multiple preset abnormal conditions, construct a time series data historical dataset including chloroform concentration and process parameters, extract statistical features from the historical dataset, and based on machine learning algorithm, establish the correlation model between the statistical features of chloroform concentration and the state of process parameters;

[0034] The correlation model is used to define the unique correspondence between different chloroform concentration features and process parameter monitoring combinations;

[0035] In step one, the industrial device or pilot device of the conversion process is operated under different stable states and multiple preset abnormal conditions respectively;

[0036] Abnormal conditions include but are not limited to: carbon tetrachloride feed flow fluctuation, water vapor ratio imbalance, reactor specific bed temperature anomaly, catalyst activity attenuation simulation, and using existing sensors and online analysis instruments, a series of time series data are collected and recorded synchronously and frequently to construct the historical dataset;

[0037] The historical dataset includes: time series data of chloroform;

[0038] Time series data of process parameters, including but not limited to: temperature, carbon tetrachloride feed rate, water vapor feed rate, system pressure;

[0039] Each piece of time series data is labeled with a state label, such as "stable state", "abnormal - low water carbon ratio", "abnormal - catalyst deactivation initial stage", and each group of state labels corresponds to a unique and effective process parameter monitoring combination under the current condition;

[0040] It should be noted that each state label corresponds to a unique and effective process parameter monitoring combination. The determination method of process parameter monitoring combination is as follows:

[0041] Based on the historical database, all cases under the same label are traced back to the root cause parameters and the correlation parameter set is extracted;

[0042] By constructing an effectiveness comprehensive score model that fuses the two dimensions of diagnostic path length and diagnostic confidence, all candidate parameter sets are quantitatively evaluated;

[0043] Specifically, the historical database is called to filter out all historical cases marked with the same abnormal state label (for example, abnormality-low water-carbon ratio);

[0044] For each historical case, the domain expert or through automated log analysis, traces and confirms which one or more process parameters are the first to appear abnormal and are the root cause of the corresponding working condition, defined as the key root cause parameter of the corresponding abnormal type;

[0045] For each historical case, extract other parameters that are strongly correlated with the key root cause parameter in statistics and change in time sequence ahead of or synchronously before and after the occurrence of the abnormality, to form a candidate monitoring parameter set;

[0046] Optionally, strong correlation can be expressed as: the absolute value of the correlation coefficient is greater than the correlation coefficient threshold (the correlation coefficient threshold is set to 0.8);

[0047] For each candidate monitoring parameter set, the number of parameters analyzed from triggering the alarm to confirming the root cause is counted as the diagnostic path length, and the mean value of the diagnostic path length of all historical cases is calculated, and the ratio of the mean path length to the total number of all monitoring parameters is calculated. The average path length ratio;

[0048] The number of historical cases in which the candidate monitoring parameter set successfully locates the root cause is counted, and the ratio to the total number of all historical cases is calculated as the diagnostic confidence;

[0049] The average path length ratio and the diagnostic confidence are multiplied to obtain the effectiveness comprehensive score;

[0050] It should be noted that the average path length ratio quantifies the diagnostic efficiency of the candidate monitoring parameter set, and the diagnostic confidence quantifies the reliability and accuracy of the candidate monitoring combination;

[0051] The candidate parameter set with the highest effectiveness comprehensive score is determined as the most effective process parameter monitoring combination under the corresponding state label.

[0052] The historical data set is preprocessed, including at least data cleaning, removing outliers, and timestamp alignment processing;

[0053] Statistical feature extraction is performed on the historical data set after preprocessing;

[0054] Specifically, for the time series data of trichloromethane concentration, the corresponding statistical features are calculated according to the set time window (for example, one window every 10 minutes), and the statistical feature set constitutes a multi-dimensional feature vector, including:

[0055] Central tendency feature: the average value of the trichloromethane concentration in the time window;

[0056] Dynamic change feature: the first derivative (i.e. change rate) of the trichloromethane concentration in the time window, used to represent the rising or falling trend of the concentration; and the second derivative (i.e. change acceleration) of the trichloromethane concentration, used to identify the intensification or mitigation of the trend;

[0057] Distribution fluctuation feature: the standard deviation of the trichloromethane concentration in the time window, used to quantify the intensity of the concentration fluctuation;

[0058] It should be noted that the core purpose of constructing a multi-dimensional feature vector is to ensure that the input information received by the correlation model can uniquely determine the dynamic running state of the current reactor. For example, a single 100ppm concentration value may be caused by multiple abnormalities, but the feature vector "100ppm, with a change rate of +2ppm / min and a standard deviation of 0.5" may uniquely point to the initial stage of the "slow decline of water vapor flow" specific fault. By introducing change rate and fluctuation, the uniqueness of the state label representation is improved, laying a foundation for the correlation model to output a unique process parameter monitoring combination.

[0059] The multi-dimensional feature vector of the trichloromethane concentration is used as the input feature of the machine learning model, and the corresponding parameter monitoring combination is used as the output target. A supervised learning algorithm is used to train the machine learning model, which learns from a large amount of historical data to establish a deterministic mapping relationship from the "trichloromethane feature space" to the "monitoring strategy space", and obtains a trained correlation model;

[0060] Optionally, the machine learning model can use gradient boosting decision trees or support vector machines;

[0061] The trained correlation model is verified using a historical data set that did not participate in the training, and the accuracy of correctly classifying different feature vectors into unique monitoring combinations is evaluated. After confirming that the performance of the correlation model meets the mapping accuracy requirements of the project, the correlation model is integrated into the online monitoring system to provide core decision support for real-time diagnosis.

[0062] In this embodiment, by constructing the correlation model, the trichloromethane concentration features monitored online can be reliably converted into a set of clear and unique process parameter monitoring instructions;

[0063] Step one is the core basis link of the online monitoring method of carbon tetrachloride conversion hydrochloric acid, which provides full-process support for subsequent real-time monitoring, abnormal early warning and accurate diagnosis. Specifically, by collecting time series data (including trichloromethane concentration and key process parameters) under steady state and various preset abnormal conditions and marking state labels, a historical data set covering all scenarios is constructed. After data preprocessing to purify data quality, on the one hand, the unique optimal process parameter monitoring combination under each working condition is selected through root cause tracing and effectiveness scoring, which optimizes the cumbersome problem of traditional full-parameter monitoring; on the other hand, the multi-dimensional statistical characteristics (including concentration trend, dynamic change and distribution fluctuation) of trichloromethane concentration are extracted to realize the unique identification of reactor operating state; then taking multi-dimensional characteristics as input and optimal monitoring combination as output, the correlation model is trained and verified, and finally the qualified model is integrated into the online monitoring system, thereby establishing the deterministic mapping of "trichloromethane characteristics-optimal monitoring combination", which lays the foundation for subsequent reduction of invalid monitoring and reduction of resource consumption, and provides core algorithm and data support for automatic decision-making of real-time monitoring.

[0064] Step two: real-time detection of trichloromethane concentration in reactor outlet product by online analytical instrument, calculation of real-time statistical characteristics of real-time trichloromethane concentration as input of correlation model, comparison of output process parameter monitoring combination with currently activated process parameters to generate control instructions;

[0065] Among them, the control instructions include starting the monitoring of unmonitored parameters or stopping the monitoring of unnecessary parameters.

[0066] In step two, the online analytical instrument (such as online gas chromatograph or online mass spectrometer) deployed on the reactor outlet pipeline is used to detect the trichloromethane concentration in the product gas stream in real time;

[0067] The same algorithm and time window as in the aforementioned step one are used to process the real-time measured trichloromethane concentration time series data, calculate the real-time statistical characteristics, and construct the real-time multi-dimensional feature vector consistent with the historical data format;

[0068] The real-time multi-dimensional feature vector is input into the correlation model, which performs real-time reasoning according to the input and outputs the corresponding process parameter monitoring combination;

[0069] The process parameter monitoring combination output by the correlation model is compared with the currently activated process parameters, and based on the comparison result, a control instruction is generated;

[0070] The control instruction is: if there is a process parameter monitoring combination output by the correlation model that is not in the currently activated process parameters, a "start monitoring" instruction is generated to include it in the real-time monitoring range.

[0071] If there is no current activated monitoring process parameters not in the process parameters monitoring combination of the associated model output, a "stop monitoring" instruction is generated, removed from real-time monitoring, or the monitoring frequency is reduced to a minimum.

[0072] Step two is the key link between modeling and real-time monitoring, which realizes the dynamic adaptation of monitoring strategy and real-time landing. Specifically, real-time acquisition of trichloromethane concentration data at the reactor outlet is performed by an online analytical instrument, real-time statistical characteristics are calculated according to the same algorithm and time window as in step one, and a real-time multi-dimensional feature vector consistent with the input format of the associated model is generated. The vector is input into the associated model to obtain the optimal process parameter monitoring combination corresponding to the current working condition, and then compared with the currently activated monitoring parameters to generate a "start monitoring necessary parameters" or "stop unnecessary parameters" control instruction. Thus, the monitoring strategy can be dynamically adjusted according to the real-time state of the reaction, reducing the complexity and resource waste of traditional fixed full-parameter monitoring, and providing real-time monitoring data for subsequent multi-parameter fusion analysis and abnormal state judgment in step three.

[0073] Step three: according to the real-time concentration of trichloromethane and the real-time monitoring data of process parameters, multi-process parameter fusion analysis is performed to determine whether the conversion process is in an abnormal state.

[0074] In step three, the dynamic monitoring data output from step two is obtained, wherein the dynamic monitoring data includes the real-time trichloromethane concentration and the real-time monitoring data of the synchronously monitored process parameters.

[0075] A preset short-time window is set to calculate the instantaneous value and short-term change trend of the real-time trichloromethane concentration and the real-time monitoring data of the synchronously monitored process parameters.

[0076] Specifically, the instantaneous value refers to the latest measurement reading in the current short-time window, which is used to determine whether it exceeds the preset safe operation threshold.

[0077] For example, in a 1-minute window cycle, the system reads the "water vapor feed rate" as 105.3 kg / h at the 60th second, which is the instantaneous value of the parameter in the current window.

[0078] The short-term change trend refers to the change direction and rate in the current short-time window, which is quantified by calculating the first derivative (i.e. slope) of the data sequence in the short-time window, and is used to realize the key of pre-warning, i.e. even if the instantaneous value is still within the normal range, but its continuous and rapid deterioration trend (such as accelerated rise of temperature) is often an early sign of major failure.

[0079] For example, also within the 1-minute time window, the short-term trend of the second bed temperature is calculated to be +0.8 °C / min, indicating that the temperature is rising at a rate of 0.8 degrees per minute, which is a signal of an accelerating temperature rise that needs to be alerted, even though the current temperature value may not have reached the absolute alarm line.

[0080] The fusion analysis and verification based on the calculated instantaneous value and short-term trend include preliminary abnormality hypothesis and multi-source evidence verification, which specifically include:

[0081] The preliminary abnormality hypothesis process: obtain the state label output by the correlation model, which is the preliminary abnormality hypothesis and the starting point of abnormality diagnosis;

[0082] The multi-source evidence verification process includes consistency cross-verification and abnormality state matching;

[0083] Specifically, the consistency cross-verification is to verify whether the instantaneous value and the short-term trend logically support the preliminary abnormality hypothesis;

[0084] The preliminary abnormality hypothesis is compared with the pre-defined "abnormality-evidence" mapping table, which defines the expected state of the key correlation parameters when each abnormality hypothesis is true;

[0085] It should be noted that the "abnormality-evidence" mapping table is a logical rule library based on the chemical reaction mechanism of the conversion process and the engineering control principle, which defines the logical relationship of "if the preliminary hypothesis is A, then to prove A, we should see evidence B and C";

[0086] For example, when the water-carbon ratio is too low, according to stoichiometry, insufficient water vapor will lead to imbalance of reactant ratio, directly showing abnormality of water vapor feed parameters, and causing accumulation of intermediate product trichloromethane, whose concentration rises, so 'low or decreasing water vapor feed rate' and 'rising trichloromethane concentration' are defined as expected supporting evidence of'suspected low water-carbon ratio' hypothesis.

[0087] If it meets the expected state, it is determined to be strong consistency, if it does not meet the expected state, it is determined to be weak consistency or inconsistency;

[0088] For example, if the preliminary abnormality hypothesis is "suspected low water-carbon ratio", the mapping table is queried to find that the expected evidence for its establishment is that the instantaneous value of "water vapor feed rate" is low or shows a downward trend, and the "trichloromethane concentration" shows an upward trend. If it is found that the "water vapor feed rate" is indeed decreasing and the "trichloromethane concentration" is rising, it is determined to be strong consistency, and the preliminary hypothesis is strongly supported. If there is no significant change in the key parameters or the change trend is opposite, it is determined to be weak consistency or inconsistency, and the preliminary hypothesis is of low credibility.

[0089] Anomaly matching involves matching a multi-parameter feature pattern, which combines the instantaneous values ​​and short-term trends of all current dynamic monitoring parameters (including chloroform) with a predefined anomaly pattern library, based on consistent cross-validation.

[0090] It should be noted that the anomaly pattern library originates from the historical dataset constructed in step one. The predefined anomaly pattern library construction process and the association model training process in step one share the same historical dataset. Specifically, after labeling the data with status in step one, for all data segments under each 'confirmed anomaly' label, the corresponding multi-parameter feature vector (i.e., including at least the instantaneous values ​​and trends of trichloromethane and each key process parameter) is calculated, and the typical numerical range of the feature vector (such as mean ± standard deviation) is statistically analyzed, thereby forming the feature fingerprint or pattern of this anomaly type. The anomaly pattern library is the collection of all feature fingerprints.

[0091] Final diagnostic output: Based on the matching results, determine whether the conversion process is in an abnormal state;

[0092] If a specific abnormal pattern is matched (e.g., "water-carbon ratio too low"), and the consistency cross-validation result is strong consistency, then the final abnormal state type is output (e.g., "Confirmed abnormality: water-carbon ratio too low").

[0093] If no known pattern can be matched and the consistency cross-validation result is weakly consistent or inconsistent, then the output transformation process is in a steady state.

[0094] If no known pattern can be matched but the consistency cross-validation result is strong consistency, the consistency cross-validation result is weak consistency, or the consistency cross-validation result is inconsistent but a specific abnormal pattern is matched, then the abnormal state type is output as unknown abnormality, and the advanced diagnostic process of full parameter scanning and expert system intervention analysis is automatically started to determine the abnormal state type.

[0095] It should be noted that the above-mentioned dual diagnostic mechanism is the key to the high reliability of this embodiment. The correlation model in step two enables rapid early warning and monitoring focus, while the multi-source evidence verification in step three provides the final diagnosis, thereby avoiding false alarms caused by model misjudgment, operating condition fluctuations or unknown interference, and ensuring the accuracy and safety of subsequent control operations.

[0096] Step three, as the core of abnormal state diagnosis, aims to accurately judge and provide early warning of abnormal states during the conversion process based on real-time data. Specifically, it involves using real-time monitoring data of chloroform concentration and process parameters to calculate the instantaneous values ​​of each data point (to determine whether the safety threshold is exceeded) and short-term trends (to quantify the direction and rate of change and capture early signs of failure). Then, through a dual diagnostic mechanism of preliminary abnormal hypothesis (based on the state label of the correlation model) and multi-source evidence verification (consistent cross-validation logic supports the hypothesis, and abnormal state matching is compared with the abnormal pattern library), it determines whether the conversion process is abnormal and the specific type of abnormality. This reduces the lag of traditional monitoring that relies solely on instantaneous values ​​and reduces the risk of misjudgment. At the same time, it provides accurate and reliable abnormal state basis for executing targeted control operations.

[0097] Step 4: If in an abnormal state, execute a preset matching control operation based on the abnormal state type.

[0098] In step four, the corresponding abnormal state type output in step three is obtained, the corresponding abnormal state type preset abnormal control logic is triggered, and the matching control command is automatically executed. The control logic clearly defines the process parameters, adjustment direction and control amount that need to be adjusted to correct the abnormality.

[0099] It should be noted that the preset abnormal control logic (e.g., "increase the opening of the steam feed regulating valve by 5%" for "too low water-carbon ratio"), the control strategy and specific parameters (such as adjustment range and maintenance time) are summarized and determined based on the abnormal operating conditions simulated and their effective correction measures when building the historical dataset in step one. In other words, the control logic is the solidification of empirical strategies that have been verified and can effectively correct the corresponding abnormal states during the process development, historical data accumulation and correlation model training stages.

[0100] For example: If the diagnosis conclusion is "Confirmed abnormality: water-carbon ratio is too low", then execute the preset abnormal control logic: increase the opening of the steam feed regulating valve by a step size (e.g., 5%) based on historical correction experience, and maintain the adjustment until the preset evaluation cycle (e.g., 3 minutes) is completed. Then, automatically return to step two to re-evaluate the control effect with the new characteristics of the trichloromethane concentration.

[0101] If the diagnosis is "Abnormality confirmed: The temperature of the first bed layer of the reactor is too high", then the preset control logic is executed: the power setting value of the corresponding bed heating unit is adjusted according to the range set based on historical operating experience (such as a reduction of X%).

[0102] This embodiment monitors the conversion process by real-time concentration of the key reaction intermediate (chloroform) in conjunction with the corresponding combination of process parameters, thereby reducing the number of process parameters that need to be continuously monitored and reducing the consumption of monitoring instruments and data storage and computing.

[0103] Since the concentration change of chloroform as a reaction intermediate is a process, when the reaction conditions deviate slightly (such as a slight drop in temperature or initial deactivation of the catalyst), the concentration of HCl may not yet show a change, but the balance of chloroform generation / consumption has been broken, and the concentration will show a detectable anomaly, thus realizing the early warning of anomalies in the conversion process.

[0104] By monitoring key reaction intermediates (chloroform), not only is the number of process parameters to be monitored optimized, but also the ability to provide early warnings of anomalies can be achieved.

[0105] Example 2

[0106] Based on the same inventive concept as the online monitoring method for the conversion of carbon tetrachloride to hydrochloric acid in the foregoing embodiments, such as Figure 3 As shown, this application provides an online monitoring system for the conversion of carbon tetrachloride to hydrochloric acid, wherein the system specifically includes:

[0107] Model analysis construction module: Obtain historical data of the steady state and various preset abnormal operating conditions of the conversion process, construct a time series historical dataset including trichloromethane concentration and process parameters, extract statistical features from the historical dataset, and establish a correlation model between the statistical features of trichloromethane concentration and the state of process parameters based on machine learning algorithms.

[0108] Dynamic monitoring strategy module: Through online analysis instruments, the concentration of chloroform in the reactor outlet product is detected in real time. The real-time statistical characteristics of the chloroform concentration are calculated as input to the correlation model. The output process parameter monitoring combination is compared with the currently activated process parameters to generate control commands.

[0109] Among them, control commands include starting the monitoring of unmonitored parameters or stopping the monitoring of unnecessary parameters;

[0110] Anomaly analysis and judgment module: Based on the real-time concentration of chloroform and the real-time monitoring data of process parameters, it performs a fusion analysis of multiple process parameters to determine whether the conversion process is in an abnormal state.

[0111] Anomaly control execution module: If in an abnormal state, execute a preset matching control operation based on the abnormal state type.

[0112] 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 illustrative of the 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 present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for online monitoring of the conversion of carbon tetrachloride to hydrochloric acid, wherein the product of the carbon tetrachloride-hydrochloric acid conversion process includes chloroform, characterized in that: Includes the following steps: Acquire historical data of the steady-state operation and various preset abnormal operating conditions of the conversion process, and construct a time-series dataset including trichloromethane concentration, process parameters and corresponding state labels; Statistical features were extracted from historical datasets, and a correlation model between the statistical features of chloroform concentration and process parameter status was established based on machine learning algorithms. The concentration of chloroform in the reactor outlet product is detected in real time by online analysis instruments. The real-time statistical characteristics of the chloroform concentration are calculated and used as input for the correlation model. The output process parameter monitoring combination is compared with the currently activated process parameters to generate control commands. Based on the real-time concentration of chloroform and the real-time monitoring data of process parameters, a fusion analysis of multiple process parameters is performed to determine whether the conversion process is in an abnormal state. If the conversion process is in an abnormal state, output the corresponding abnormal state type and execute the preset matching control operation based on the abnormal state type.

2. The method for online monitoring of carbon tetrachloride-to-hydrochloric acid conversion according to claim 1, characterized in that: Each segment of time-series data for each trichloromethane concentration and process parameter is marked with a corresponding status label. Each status label uniquely corresponds to a combination of process parameter monitoring parameters. The determination process is as follows: Based on the historical database, root cause parameters are traced and related parameter sets are extracted for all cases under the same label; By constructing a comprehensive effectiveness scoring model that integrates two dimensions—diagnostic path length and diagnostic confidence—a set of candidate parameters is obtained and evaluated. The candidate parameter set with the highest overall effectiveness score is determined as the process parameter monitoring combination under the corresponding status label.

3. The method for online monitoring of carbon tetrachloride conversion to hydrochloric acid according to claim 2, characterized in that: The process of obtaining the candidate parameter set is as follows: Retrieve the historical database and filter out all historical cases that are marked with the same abnormal status label; For each historical case, the first abnormal process parameter is traced and identified through automated log analysis, and is defined as the key root cause parameter for the corresponding abnormality type. For each historical case, parameters that are statistically strongly correlated with the key root cause and whose changes precede the occurrence of the anomaly are extracted to form a set of candidate monitoring parameters.

4. The method for online monitoring of carbon tetrachloride conversion to hydrochloric acid according to claim 2, characterized in that: The process of obtaining the comprehensive effectiveness scoring model is as follows: For each candidate monitoring parameter set, the number of parameters that need to be analyzed from triggering the warning to confirming the root cause is counted as the diagnostic path length. The average diagnostic path length of all historical cases is calculated, and the proportion of the average diagnostic path length to the total number of all monitoring parameters is calculated as the average path length ratio. The number of historical cases in which the candidate monitoring parameter set successfully located the root cause was counted, and the proportion of these cases to the total number of historical cases was calculated as the diagnostic confidence level. The average path length ratio is multiplied by the diagnostic confidence level to obtain the comprehensive effectiveness score.

5. The method for online monitoring of carbon tetrachloride-to-hydrochloric acid conversion according to claim 1, characterized in that: The correlation model uses the multidimensional feature vector of chloroform concentration as input features, takes the corresponding parameter monitoring combination as output target, uses a supervised learning algorithm to train the machine learning model, and uses a historical dataset that was not used in the training to verify the trained correlation model. The correlation model that meets the performance verification is integrated into the online monitoring system. The multidimensional feature vector of chloroform concentration includes: the average value of chloroform concentration within the time window, the first derivative of chloroform concentration within the time window, the second derivative of chloroform concentration, and the standard deviation of chloroform concentration within the time window.

6. The method for online monitoring of carbon tetrachloride-to-hydrochloric acid conversion according to claim 1, characterized in that: The process of generating control commands is as follows: The concentration of chloroform in the product gas stream is detected in real time, and real-time statistical characteristics are calculated to construct a real-time multidimensional feature vector. The real-time multidimensional feature vector is input into the correlation model, and the corresponding process parameter monitoring combination is output. The process parameter monitoring combination output by the correlation model is compared with the currently activated process parameters, and control commands are generated based on the comparison results.

7. The method for online monitoring of carbon tetrachloride conversion to hydrochloric acid according to claim 6, characterized in that: The control command is: If the process parameter monitoring combination output by the associated model is not among the currently activated process parameters, a monitoring start command will be generated and included in the real-time monitoring range. If there is a process parameter monitoring combination that is not currently being monitored in the associated model output, a stop monitoring command will be generated and the process parameter will be removed from the real-time monitoring.

8. The method for online monitoring of carbon tetrachloride-to-hydrochloric acid conversion according to claim 1, characterized in that: The process of performing fusion analysis of multiple process parameters is as follows: Acquire real-time monitoring data of chloroform concentration and synchronously monitored process parameters; A preset short-term window is used to calculate the instantaneous values ​​and short-term trends of real-time trichloromethane concentration and real-time monitoring data of synchronously monitored process parameters. The analysis and verification are based on the fusion of calculated instantaneous values ​​and short-term trends, including: Preliminary anomaly hypothesis process: Obtain the state labels output by the association model; these output state labels constitute the preliminary anomaly hypothesis. The multi-source evidence verification process includes: consistency cross-validation and anomaly state matching; Consistent cross-validation verifies whether instantaneous values ​​and short-term trends logically support the initial anomalous hypothesis. Anomaly matching involves matching a multi-parameter feature pattern, which combines the instantaneous values ​​and short-term trends of all current dynamic monitoring parameters, with a predefined anomaly pattern library, based on consistent cross-validation.

9. The method for online monitoring of carbon tetrachloride conversion to hydrochloric acid according to claim 8, characterized in that: The instantaneous value refers to the measurement reading in the latest sampling period within the current short-time window; The short-term trend refers to the direction and rate of change within the current short-term window, which is quantified by calculating the first derivative of the data sequence within the short-term window.

10. The method for online monitoring of carbon tetrachloride-to-hydrochloric acid conversion according to claim 8, characterized in that: The process for determining whether the conversion process is in an abnormal state is as follows: If a specific anomaly pattern is matched and the consistency cross-validation result is strong consistency, then the final anomaly state type is output. If no known pattern can be matched but the consistency cross-validation result is strong consistency, the consistency cross-validation result is weak consistency, or the result is inconsistent but a specific abnormal pattern is matched, then the output abnormal state type is unknown abnormality.

Images