A soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation

Abstract

This invention discloses a soft measurement and quality control method for the continuous reaction process of calcium formate carbonylation, belonging to the field of cheminformatics technology. The method includes: collecting auxiliary variables and offline analytical values ​​of the reaction process; establishing a mechanistic model branch based on the carbonylation reaction kinetics to calculate the calcium formate concentration C. 机理 ; Using auxiliary variables as input, and C 机理 The difference between the test value and the target neural network data-driven branch is used to output the correction amount ΔC; the result is C. soft =C 机理 The system calculates +ΔC and uses Kalman filtering to dynamically correct the concentration based on the new test values. The corrected soft measurement value is then used as a feedback signal to achieve closed-loop control of product quality by proportionally adjusting the reaction temperature. This invention achieves real-time, high-precision soft measurement and online quality control of calcium formate concentration, solving the problems of large lag, inability to close the loop, and model mismatch caused by catalyst deactivation in traditional methods. This significantly improves product quality stability and reduces production costs.

Description

Technical Field

[0001] This invention relates to the field of cheminformatics technology, and in particular to a soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation. Background Technology

[0002] Calcium formate (chemical formula Ca(HCOO)2) is an important fine chemical product, widely used as a feed additive (antibacterial calcium supplement), a building early-strength agent, a cement additive, and an oilfield drilling additive. Traditional calcium formate production processes involve the neutralization reaction of formic acid and calcium hydroxide. This process consumes high-cost formic acid raw materials, and the reaction is conducted in an intermittent manner, resulting in high energy consumption, low efficiency, and poor product quality stability.

[0003] In recent years, significant progress has been made in the carbonylation method for synthesizing calcium formate. This process uses carbon monoxide (CO) and calcium hydroxide (Ca(OH)2) as raw materials, which react directly to produce calcium formate in the presence of a catalyst. The chemical equation for this reaction can be expressed as: CO + Ca(OH)2 → Ca(HCOO)2.

[0004] This process achieves 100% atomic utilization and can utilize CO purified from steel plant converter gas and lime from calcium carbide slag solid waste as raw materials, giving it outstanding advantages such as economical raw material sources, low energy consumption, and environmental friendliness.

[0005] When investigating the reaction kinetics of calcium hydroxide carbonylation to synthesize calcium formate, a power function was used as the kinetic equation to process the kinetic data. The main kinetic parameters are: the reaction order of carbon monoxide is 0.8, the activation energy is 28.245 kJ / mol, and the pre-exponential factor is 24.23 mol. -1 ·L -1 ·min -1 ·MPa -1 .

[0006] Although the carbonylation method has significant advantages in terms of raw material economy and environmental friendliness, the key quality indicators of its continuous reaction process—calcium formate concentration and reaction conversion rate—currently rely on manual sampling and offline laboratory analysis. This detection method has the following prominent problems: 1) Severe delay: It usually takes several hours from sampling and delivery to laboratory analysis. The test results cannot reflect the operating status of the reactor in real time. By the time a quality abnormality is detected, a large number of unqualified products have already been produced. 2) High cost: Frequent manual sampling and laboratory analysis increase human resource consumption and testing costs; 3) Inability to achieve closed-loop control: Due to the lack of real-time quality feedback signals, the existing control system can only perform single-loop PID control on auxiliary variables such as temperature, pressure, and flow rate. It cannot actively adjust the reaction conditions according to changes in product quality. Product quality control mainly relies on experience judgment and lacks systematicness and precision. 4) Interference from fluctuations in the quality of steel converter gas: This invention relies on a converter gas purification and upgrading system. The purified CO gas source inevitably contains trace impurities such as iron carbonyl (Fe(CO)5). Under carbonylation reaction conditions, these impurities decompose into highly dispersed iron on the catalyst surface upon heating, occupying and covering the active sites of the catalyst, leading to a decrease in catalyst activity and consequently causing fluctuations in the reaction conversion rate. Minor changes in the quality of the CO gas source are difficult to detect in a timely manner under offline detection mode. Summary of the Invention

[0007] This invention aims to provide a soft measurement and quality control method for the continuous reaction process of calcium formate carbonylation, to solve the technical problems in existing technologies such as the inability to measure calcium formate concentration and reaction conversion rate in real time online, lagging product quality control, and difficulty in dealing with the impact of CO gas source impurities. The specific solution is as follows: A soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation includes the following steps: S1. Real-time acquisition of process variable data during the continuous reaction of calcium formate carbonylation, including the true value of calcium formate concentration C obtained from offline laboratory analysis. true ; S2. Calculate the calcium formate concentration C at the reactor outlet based on the above process variable data. 机理 ; S3. Train a prediction model based on process variable data to predict the true value of calcium formate concentration C in real time. true With the calculated value C 机理 The difference ΔC between them yields the baseline value C for soft measurement of calcium formate concentration. soft :C soft =C 机理 +ΔC, and for C soft Perform optimal estimation and correction using Kalman filtering to obtain C. soft_out ; S4, according to C soft_out Real-time calculation of reaction conversion rate X, based on C soft_out With target concentration C target The reaction temperature is automatically adjusted according to the deviation, realizing real-time closed-loop control of product quality.

[0008] Furthermore, the process variable data includes auxiliary variables and quality control variables, wherein the auxiliary variables include: CO gas flow rate at reactor inlet, calcium hydroxide slurry flow rate at reactor inlet, temperature (T1, T2, ..., Tn) of the multi-stage reactor in the reactor, pressure in the reactor, composition of reactor tail gas, stirring speed of reactor, effective volume of reactor, pH value of circulating slurry, and concentration of carbonyl iron in CO gas source. The quality control variable includes the true value C of calcium formate concentration obtained through offline laboratory analysis. true and reaction conversion rate; Between steps S1 and S2, a data preprocessing step is also included: the process variable data collected in real time during the continuous reaction of calcium formate carbonylation are cleaned, outlier removed, and normalized. The normalization formula is as follows: X norm =(XX min ) / (X max -X min ), where X norm X represents the normalized data value, and X represents the original data value. min and X max These are the minimum and maximum values ​​of the variable in the training sample set, respectively.

[0009] Furthermore, step S2 specifically includes: S21. Calculate the rate r of the calcium formate carbonylation reaction: ; Where r is the reaction rate, A is the pre-exponential factor, Ea is the activation energy of the reaction, R is the ideal gas constant, T is the absolute temperature of the reaction, exp is the natural exponential function, and P is the reaction rate. CO ρ is the partial pressure of CO in the reactor, and n is the reaction order of CO. S22. Calculate the calcium formate concentration C at the reactor outlet. 机理 :C 机理 =(r·τ·ρ slurry )·E reactor ·X Ca ; Where C 机理 ρ is the concentration of calcium formate at the reactor outlet, τ is the average residence time of the reactants in the reactor, and ρ is the concentration of calcium formate at the reactor outlet. slurry The density of the reaction slurry, E reactor X is the reactor efficiency coefficient. Ca The initial conversion rate of calcium hydroxide is given.

[0010] Further, step S3 uses a BP neural network to predict ΔC: the true value of calcium formate concentration C obtained from offline laboratory analysis. true The calculated concentration of calcium formate at the reactor outlet, C 机理The difference between the values ​​is used as the target output, and the auxiliary variables at the corresponding time points are used as input samples to construct training and validation sets (80% for training and 20% for validation). The network weights are updated using the backpropagation algorithm until the convergence condition is met.

[0011] Furthermore, the BP neural network structure used is as follows: Input layer: K nodes, where K corresponds to the number of auxiliary variables collected. This device uses 13 auxiliary variables: temperature of each reactor (4 typical reactors, T1~T4), reaction pressure P, and CO gas feed flow rate F. CO Calcium hydroxide slurry feed flow rate F Ca Stirring speed R, pH value of circulating slurry, carbonyl iron concentration C Fe CO content in exhaust gas Y CO And exhaust CO2 content Y CO2 That is, K=13; Hidden layer: H nodes, set H=2K+1=27; Output layer: 1 node, outputs the correction amount ΔC for calcium formate concentration; Activation function: The hidden layer uses the sigmoid function f(x) = 1 / (1+e^x). -x The output layer uses the linear activation function identity. Training method: The Levenberg-Marquardt algorithm is used for weight optimization. The learning rate μ is initially set to 0.1 (μ is dynamically adjusted as the training error decreases). The number of training epochs is 500. The termination condition is that the validation set error no longer decreases for 10 consecutive epochs or the maximum number of training epochs is reached.

[0012] Furthermore, when new offline laboratory analysis samples C are obtained... true With C soft When the time is misaligned, the Kalman filter algorithm is used to adjust C. soft Perform optimal estimation correction to obtain C soft_out The Kalman filter recursive formula is as follows: State prediction: ; Covariance prediction: ; Kalman gain calculation: ; State update (C soft_out =Estimated value after state update ): ; Covariance update: ; Initial values ​​for recursion: Set as the current C soft Value; the initial covariance of P0 is set to 1 (indicating a large uncertainty in the initial estimate). Kalman filter correction allows for high-precision soft measurement output C even when offline test results lag by approximately 1 hour. soft_out ; Where k is the k-th offline sampling time (k=1,2,3,…); The state prediction based on the information from the previous step, i.e., C before Kalman filter correction. soft The estimated value; The state estimate after Kalman filtering correction represents the final soft-sensor output C. soft_out F is the state transition matrix, set to 1, indicating that the mass concentration shows no trend within adjacent sampling intervals; H is the observation matrix, set to 1, indicating that offline measurements and predicted values ​​can be directly compared under the same dimensions; Q is the process noise covariance, set to 0.001. 2 R represents the observation noise covariance. Based on the measurement standard deviation of the offline laboratory instrument, the standard deviation is set to 0.5%~1.0%, and R = 0.005 is chosen. 2 ;P k I is the posterior estimation error covariance matrix; I is the identity matrix; C true_k This represents the true value of calcium formate concentration obtained from the k-th offline test.

[0013] Furthermore, the reaction conversion rate X is indirectly obtained from the ratio of the real-time concentration of calcium formate to the maximum theoretical concentration: X = C soft_out / C max C max The calculation is based on the assumption that all calcium hydroxide feed is converted into calcium formate.

[0014] Furthermore, based on C soft_out With target concentration C target The formula for automatically adjusting the reaction temperature based on the deviation is: ΔT set =K p ·(C target -C soft_out ), where ΔT set K is the adjustment amount for the reaction temperature setpoint. p C is the proportional gain coefficient. target The target product quality concentration; proportional gain coefficient K p The tuning should satisfy the following safety boundary constraints: T reactor ∈[140℃, 200℃], where T reactor This is the measured temperature of the reactor.

[0015] Furthermore, it also includes step S5, online monitoring of gas source gas quality and adaptive model updating: an online carbonyl iron analyzer is installed on the feed pipeline before CO enters the reactor. When the carbonyl iron concentration is detected to exceed the threshold T, the analyzer will update the gas source gas quality and the model adaptive update. Fe When, or when the cumulative integral of carbonyl iron exceeds the cumulative threshold T Fe_int At that time, the system automatically introduces an activity decay factor α (0 < α ≤ 1) into the rate equation of the mechanistic model, and the corrected reaction rate is: The value of α is calculated using the exponential decay model: α = exp[-β·∫C Fe ·dt]; Where β is the attenuation coefficient, C Fe For carbonyl iron concentration, ∫C Fe ·dt is the cumulative carbonyl iron concentration over time from the start of catalyst operation, calculated in real time using the trapezoidal numerical integration method.

[0016] Furthermore, the incremental learning process will be automatically initiated when any of the following triggering conditions are met: 1) Cumulative integral of carbonyl iron ∫C Fe ·dt exceeds the cumulative threshold T Fe_int ; 2) Error of N consecutive soft measurements (|C) true -C soft | / C true Exceeding the set value; 3) Periodic fine-tuning will be mandatory every 168 hours (7 days) of accumulated running time; The incremental learning method employs a "transfer learning" strategy: freezing the weights of the first two hidden layers of the BP neural network and updating only the weights of the output layer (i.e., fine-tuning only the linear combination coefficients of the last layer of the model). The loss function is consistent with the error function during model training, and the minimum batch size for incremental learning is 50 new samples.

[0017] Compared with the prior art, the present invention has at least one of the following technical effects: 1. This invention utilizes easily measurable auxiliary variables (temperature, pressure, flow rate, pH value, iron carbonyl concentration, etc.) to calculate calcium formate concentration and reaction conversion rate in real time. The soft-sensor values ​​are refreshed every second or minute, replacing offline testing that occurs every few hours. This solution reduces the delay in acquiring quality information from hours to seconds, enabling operators or control systems to promptly detect and intervene in reaction anomalies, preventing a large number of defective products. The soft-sensor value refresh frequency can reach once per minute, compared to offline testing, such as once every four hours, significantly reducing information delay. Therefore, it can significantly reduce the defect rate caused by delayed quality feedback after implementation, reducing the number of defective products per month.

[0018] 2. This invention primarily uses online calculations based on soft measurement models, with offline testing only used for model calibration and Kalman filter updates. The testing frequency can be reduced from, for example, once every 4 hours to, for example, once every 8-12 hours, reducing the workload of manual sampling, sample preparation, and instrument analysis. This also reduces the high labor costs, reagent costs, and consumable costs caused by frequent offline testing, while simultaneously reducing laboratory waste discharge.

[0019] 3. This invention will use soft measurement values ​​(C) soft Or C after Kalman filter correction soft_out As a feedback signal, a proportional control loop is constructed: when the soft-sensor value is below the target concentration lower limit, the reaction temperature is automatically increased; when it is above the upper limit, the temperature is automatically decreased. This closed-loop control ensures that product quality always operates within the target range, eliminating reliance on human experience. Implementing closed-loop control reduces the standard deviation of calcium formate product concentration fluctuations, improves product quality stability, and correspondingly reduces customer complaint rates; simultaneously, due to more precise temperature regulation (compared to human experience), steam or cooling water consumption is reduced.

[0020] 4. This invention incorporates an online CO gas chromatography-mass spectrometry (ACMS) module (carbonyl iron analyzer) to monitor Fe(CO)₅ content in real time. When the cumulative exposure exceeds a threshold, an activity decay factor α is introduced into the mechanistic model, simultaneously triggering incremental learning (fine-tuning the neural network output layer) of the data-driven model. This mechanism enables the soft-sensor model to adapt online to the slow decay of catalyst activity and fluctuations in gas source quality, avoiding model output drift. This is achieved when carbonyl iron concentration fluctuates (0~20 mg / m³). 3 Under these conditions, the prediction error of the soft measurement model can still be stably controlled within a reasonable range; catalyst activity decay compensation extends the effective service life of the model, reduces the number of times offline retraining is required due to model mismatch, and ensures the reliability of the model for long-term continuous production.

[0021] 5. This invention utilizes offline laboratory values ​​to correct the soft measurement output C using Kalman filtering. soft Performing optimal estimation effectively reduces C soft Compared with the true value of calcium formate concentration C true The remaining uncompensated deviations improve the accuracy of calcium formate concentration prediction, while overcoming the information delay problem caused by the lag in test results, ensuring the continuous reliability of soft measurement values ​​and the effectiveness of real-time control.

[0022] 6. This invention can extend the effective service life of catalysts and save catalyst procurement costs by predicting the remaining life of catalysts and optimizing replacement schemes. At the same time, it can reduce the risk of unplanned shutdowns and avoid product quality fluctuations or abnormal reactions caused by overuse of catalysts. It can also improve the level of precision in production management by transforming experience-based judgments into quantifiable and traceable mathematical models. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic flowchart of a soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0025] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0026] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0027] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0028] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0029] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0030] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0031] See appendix Figure 1 A soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation is disclosed, including: Step 1: Data Acquisition and Processing.

[0032] Process variable data were collected during the continuous reaction of calcium formate carbonylation, including auxiliary variables and quality control variables.

[0033] The auxiliary variables are variables that can be measured in real time by online sensors, including: CO gas flow rate at the reactor inlet (unit: Nm³). 3 / h), Calcium hydroxide slurry flow rate at reactor inlet (unit: kg / h or m³) 3 / h), the temperature of the multi-stage reactor (T1, T2, ..., Tn), the pressure inside the reactor (P), the composition of the reactor tail gas (CO content, CO2 content, CH4 content, etc.), the stirring speed of the reactor (R), the pH value of the circulating slurry (the part of the material separated from the mixture after the reaction and sent back to the reactor to continue to participate in the reaction), and the concentration of carbonyl iron in the CO gas source (calculated as Fe(CO)5).

[0034] The quality control variables are the concentration of calcium formate and the reaction conversion rate obtained through offline laboratory analysis, which are used for training and validation of the soft measurement model.

[0035] The collected data undergoes preprocessing, including data cleaning, outlier removal, and normalization. The normalization formula is as follows: X norm =(XX min ) / (X max -X min ). Where X norm X represents the normalized data value, and X represents the original data value. min and X maxThese are the minimum and maximum values ​​of the variable in the training sample set, respectively. This linear normalization method is applicable to all auxiliary variables in this invention, compressing each variable to the [0, 1] interval, eliminating numerical weight bias caused by differences in dimensions such as temperature, flow rate, and pH, and ensuring that each auxiliary variable is treated fairly during model training.

[0036] Step 2: Construct a mechanistic model branch for concentration prediction based on the reaction mechanism.

[0037] The carbonylation reaction of calcium formate, CO + Ca(OH)₂ → Ca(HCOO)₂, follows a power-law kinetic equation, and the reaction rate can be expressed as: .

[0038] Where: r is the reaction rate (unit: mol·L⁻¹) -1 ·min -1 k(T) is the reaction rate constant (unit: mol·L). -1 ·min -1 ·MPa -n ), P CO Let T be the partial pressure of CO in the reactor (unit: MPa), n be the reaction order of CO (experimental value, n=0.8), and k(T) follow the Arrhenius equation: k(T)=A·exp[-Ea / (R·T)].

[0039] Where: A is the pre-exponential factor (unit: mol·L⁻¹) -1 ·min -1 ·MPa -n After experimental calibration and correction, A = 24.23; Ea is the activation energy of the reaction (unit: J / mol), Ea = 28.245 × 10⁻⁶. 3 J / mol; R is the ideal gas constant, R = 8.314 J / (mol·K); T is the absolute temperature of the reaction (unit: K), T = T reactor +273.15 (T) reactor (where is the measured temperature of the reactor); exp is the natural exponential function.

[0040] T reactor The measured temperature of the reactor is a key temperature parameter used in the mechanistic model to calculate the reaction rate. The temperature value of a certain stage (such as the last stage main reaction zone) can be selected according to the process requirements, or the weighted average of the temperatures of each stage (usually with the residence time of the material in each reactor as the weight) can be taken to represent the characteristic reaction temperature of the entire reaction process.

[0041] The reaction rate formula is derived as follows: For the gas-liquid-solid three-phase carbonylation reaction CO + Ca(OH)₂ → Ca(HCOO)₂, it is generally assumed that CO dissolves in the liquid phase and reacts with Ca(OH)₂ on the solid surface. When liquid-solid mass transfer resistance is neglected or the reaction is assumed to be a surface-controlled step, the reaction rate mainly depends on the concentration of dissolved CO in the liquid phase. According to Henry's Law, the concentration of dissolved CO is related to the partial pressure P of CO in the gas phase. CO Proportional, therefore the rate can be expressed as: .

[0042] Where: k′(T) is the intrinsic reaction rate constant related to temperature (including the effect of activation energy, with units matching the reaction order), C CO,dissolved Let be the concentration of CO in the reaction liquid phase (in mol / L), and let m be the intrinsic reaction order of CO in the liquid phase. According to Henry's Law, C CO,dissolved =H·P CO ·C CO,dissolved =H·P CO Where H is the Henry's law constant for CO in the reaction slurry (unit: mol·L⁻¹). -1 ·MPa -1 ), P CO Let be the partial pressure of CO in the gas phase (in MPa). Therefore, the dissolved concentration is directly proportional to the partial pressure. Substituting this into the equation, we get... Where k(T) = k′(T)·H m Let n be the apparent rate constant, and the combined exponent n=m is the apparent reaction order.

[0043] This invention uses a reaction order of n=0.8 as an example. 0.8 is the apparent reaction order obtained by fitting kinetic data under specific experimental conditions (such as temperature, pressure, catalyst, etc.), which reflects the influence of CO concentration (or partial pressure) on the reaction rate in the system. Of course, the actual reaction order may fluctuate slightly due to changes in catalyst activity, gas-liquid mass transfer, and other conditions (e.g., 0.75 to 0.85).

[0044] Based on the above formulas, the specific expression for the carbonylation reaction rate is as follows: Based on the above rate equation, combined with material balance and residence time distribution, a predictive mechanism model for the concentration of calcium formate at the reactor outlet is established: C 机理 =r·τ·E reactor ·X Ca The detailed derivation process is as follows: Step 1: Calculate the outlet molar concentration of calcium formate based on CSTR material balance. Under the assumption of steady-state operation of a fully mixed-flow reactor and neglecting the change in total material volume before and after the reaction, the outlet molar flow rate of calcium formate is equal to its outlet molar concentration C. 机理Multiply by the total outlet volumetric flow rate, which is approximately equal to the total inlet feed volumetric flow rate Q. total Steady-state material balance of calcium formate product in a mixed-flow reactor: The molar flow rate of calcium formate outlet equals the molar rate of calcium formate formation, i.e., C. 机理 ·Q total =r·V reactor C 机理 Let C be the molar concentration (mol / L) of calcium formate at the reactor outlet. Solving for C, we get: 机理 =r·V reactor / Q total =r·τ.

[0045] Step 2: Introduce the reactor efficiency coefficient E reactor Actual industrial reactors exhibit non-ideal flow (such as dead zones, short-circuiting, and non-uniform mixing) and gas-liquid-solid phase transfer resistance, resulting in actual outlet concentrations lower than the calculated values ​​under the ideal assumption of perfectly mixed flow. Therefore, a dimensionless efficiency coefficient E is introduced. reactor ∈(0,1]:C 机理 =r·τ·E reactor .

[0046] Step 3: Introduce the initial conversion rate X of calcium hydroxide. Ca The calcium hydroxide feedstock is not 100% reactive (it may contain inert impurities, calcium carbonate, etc.), and only a portion of the solids in the feed slurry is available Ca(OH)2. Let X... Ca ∈(0,1] represents the "initial reactive proportion of calcium hydroxide" (an empirical constant, determined beforehand through raw material purity analysis). This factor further reduces the actual concentration of calcium formate that can be produced: C 机理 =r·τ·E reactor ·X Ca .

[0047] Where: C 机理 The concentration of calcium formate at the reactor outlet (unit: mol / L or mass fraction) is calculated based on the mechanistic model; τ is the average residence time of the reactants in the reactor (unit: min, τ=V). reactor / Q total ); Q total The total feed flow rate converted to standard conditions (CO gas flow rate + calcium hydroxide slurry flow rate, calculated based on the feed volumetric flow rate converted to standard conditions, unit: L / min); ρ slurry The density of the reaction slurry (unit: kg / L) is calculated by mass weighting of calcium hydroxide, calcium formate, and water: ρ slurry =(m Ca(OH)2 +m 产物 +m水 ) / V total ; V reactor The effective reaction volume of the reactor (unit: L), V is the effective reaction volume of the reactor. reactor Defined as the volume occupied by the slurry actually participating in the gas-liquid-solid three-phase reaction within the reactor, that is, the total geometric volume of the reactor minus the volume occupied by internal components (such as heat exchange coils, stirring blades, baffles, thermometer sleeves, etc.), and the volumes of the gas phase space at the top of the reactor and the sludge zone at the bottom that do not participate in effective mixing. In engineering, it can be estimated through the following steps: First, obtain the total reactor volume V based on the reactor equipment drawings. total Then subtract the discharge volume of all internal components (coils, stirring shafts and blades, baffles, etc.), which can be calculated by the component size and immersion depth or provided by the equipment supplier; finally deduct the gas phase above the normal operating liquid level and the dead zone that may exist below the liquid level (such as the bottom settling zone), usually taking the slurry volume corresponding to the actual operating liquid level.

[0048] E reactor The reactor efficiency coefficient, ranging from 0.80 to 0.95, is determined through cold model tracer experiments combined with operational parameter identification. The cold model experiment method involves using a room-temperature, ambient-pressure acrylic glass model reactor, geometrically similar to an industrial reactor, with brine as the liquid phase and compressed air as the gas phase. A pulsed injection of KCl tracer is employed to measure the residence time distribution (RTD) curve. By comparing the actual residence time distribution with the theoretical residence time distribution of an ideal total mixed reactor (CSTR) model, the efficiency coefficient is determined using equation "E". reactor The baseline value is calculated using the formula: "Ideal distribution variance / Actual distribution variance". During operation, the Recursive Least Squares (RLS) method is used to calculate E. reactor Perform online identification: calculate value C using the mechanistic model. 机理 Compared with offline test value C true With the objective of minimizing the residual, E is updated every 24 hours. reactor The estimated value is obtained by using valid data from the past 72 hours as the identification window. The detailed steps and RLS recursive formula of this method are achievable by those skilled in the art based on conventional knowledge of chemical process parameter identification, ensuring the accuracy of E. reactor The parameters can still accurately reflect the actual reactor efficiency even when the process changes; X Ca X represents the initial conversion rate of calcium hydroxide. CaDefined as an empirical constant pre-calculated based on the purity of the raw material calcium hydroxide and the solid content of the feed slurry, it does not change in real time with reaction conditions. Its specific determination method is as follows: At the first feed of the system or each time a batch of calcium hydroxide raw material is changed, the raw material is analyzed offline to detect the mass fraction of calcium hydroxide (by acid-base titration or thermogravimetric analysis), and simultaneously the solid content of the feed slurry is detected (by drying and weighing). From equation "X..." Ca The value is calculated as "Mass fraction of Ca(OH)2 in raw materials × Mass ratio of Ca(OH)2 in the solid phase", and is typically taken in the range of 0.95 to 0.99. For the typical process conditions of this invention (Ca(OH)2 purity ≥ 92% after pretreatment of carbide slag, solid content 8%–15%), X... Ca The default value is 0.98, which has been verified through 100 batches of experiments in the pilot plant. Introducing this correction factor can improve the mechanism model C. 机理 Under steady-state conditions, the systematic deviation between the offline and in-process test values ​​is controlled within ±2%, eliminating the need for real-time adjustments during operation. If the raw material source is changed or the pretreatment process is adjusted, X should be re-determined through the aforementioned offline analysis. Ca And update the soft measurement model parameters.

[0049] This mechanistic model branch directly embeds the kinetic nature of chemical reactions (the power-law relationship between reaction rate and temperature and CO partial pressure) into the reasoning framework, providing physically interpretable benchmarks for soft measurements.

[0050] Step 3: Construct a data-driven model branch for quality prediction based on neural networks.

[0051] To compensate for the biases caused by idealized assumptions in the mechanistic model branches (such as fully mixed reactor flow, constant catalyst activity, and inaccurate reactor efficiency coefficient), this invention constructs a data-driven model branch to learn the influence of complex nonlinear factors not covered by the mechanistic model.

[0052] The BP (backpropagation) neural network structure used is as follows: Input layer: K nodes, where K corresponds to the number of auxiliary variables collected in step one. This device uses 13 auxiliary variables: temperature of each reactor (4 typical reactors, T1~T4), reaction pressure P, and CO gas feed flow rate F. CO Calcium hydroxide slurry feed flow rate F Ca Stirring speed R, pH value of circulating slurry, carbonyl iron concentration C Fe CO content in exhaust gas Y CO And exhaust CO2 content Y CO2 That is, K=13.

[0053] Hidden layer: H nodes, set H=2K+1=27 (based on the universal approximation theorem proposed by Hecht-Nielsen, a three-layer neural network with one hidden layer can approximate any continuous function with arbitrary precision, and 2K+1 is a commonly used empirical number of nodes).

[0054] Output layer: 1 node, outputs the correction amount ΔC for calcium formate concentration.

[0055] Activation function: The hidden layer uses the sigmoid function f(x) = 1 / (1+e^x). -x The output layer uses a linear activation function (identity).

[0056] Training method: The Levenberg-Marquardt algorithm is used for weight optimization. The learning rate μ is initially set to 0.1 (μ is dynamically adjusted as the training error decreases). The number of training epochs is 500. The termination condition is that the validation set error no longer decreases for 10 consecutive epochs or the maximum number of training epochs is reached.

[0057] Training process of data-driven model branch: using the true value C of calcium formate concentration obtained from offline laboratory analysis. true With the branch output C of the mechanism model 机理 The difference between them (ΔC) target =C true -C 机理 Using the target output and auxiliary variables at corresponding time points as input samples, a training set and a validation set are constructed (80% for training and 20% for validation). The network weights are updated using the backpropagation algorithm until the convergence condition is met.

[0058] The core significance of introducing this data-driven branch lies in the fact that in the complex system of gas-liquid-solid three-phase carbonylation reaction, the assumptions of the actual mechanism model cannot be fully met in continuous industrial production processes—the residence time distribution function deviates from the ideal perfectly mixed flow, the catalyst activity slowly decays with operating time, and the fluctuation and accumulation effects of CO source impurities. The mechanism model cannot express these dynamic nonlinear effects with concise explicit formulas, while the BP neural network, with its multi-layer nonlinear mapping capability, can effectively approximate these deviations, serving as a supplement and correction to the mechanism model, thereby achieving accurate fitting of the quality output driven by a large amount of production operation data.

[0059] Step 4: Construct a hybrid soft measurement model.

[0060] Output C from the mechanism model branch 机理 By fusing the data-driven model branch output ΔC, the baseline value for soft measurement of calcium formate concentration is obtained: C soft =C 机理 +ΔC.

[0061] Where C soft The concentration of calcium formate obtained by soft measurement (unit: mol / L, which is finally converted to mass fraction based on the molar mass of calcium formate, 129.11 g / mol).

[0062] During the use of the hybrid soft sensor model, when a new offline laboratory analysis sample C is obtained... true With C soft When the time is misaligned (offline analysis lags by approximately 1 hour), the Kalman filter algorithm is used to process C. soft Perform optimal estimation and correction. The Kalman filter recursive formula is as follows: State prediction: ; Covariance prediction: ; Kalman gain calculation: ; State update (C soft_out =Estimated value after state update ): ; Covariance update: .

[0063] in: k is the kth offline sampling time (k=1,2,3,…); The state prediction based on the information from the previous step (i.e., C before Kalman filter correction) soft (estimated value) The state estimate after Kalman filtering correction (representing the final soft-sensor output C) soft_out F is the state transition matrix (set to 1, indicating that the mass concentration shows no trend within adjacent sampling intervals); H is the observation matrix (set to 1, indicating that offline measurements and predicted values ​​can be directly compared under the same dimensions); Q is the process noise covariance (set to 0.001). 2 R represents the observation noise covariance (the standard deviation is set to 0.5%~1.0% based on the measurement standard deviation of the offline laboratory instrument, and R=0.005 is used). 2 ); P k I is the posterior estimation error covariance matrix; I is the identity matrix; C true_k This represents the true value of calcium formate concentration obtained from the k-th offline test.

[0064] Initial values ​​for recursion: Set as the current C softValue; the initial covariance of P0 is set to 1 (indicating a large uncertainty in the initial estimate). Kalman filter correction allows for high-precision soft measurement output C even when offline test results lag by approximately 1 hour. soft_out This is the key to ensuring model stability and eliminating the effects of time delay in the engineering application of this invention.

[0065] Step 5: Online quality control.

[0066] The reaction conversion rate X is indirectly obtained from the ratio of the real-time concentration of calcium formate to the maximum theoretical concentration: X = C soft_out / C max C max Calculated based on the complete conversion of calcium hydroxide feed to calcium formate (C max =Calcium hydroxide feed molar flow rate / Total slurry volumetric flow rate (mol / L). Wherein, the calcium hydroxide feed molar flow rate (unit: mol / min) is calculated based on the solid content (mass fraction), slurry density, and feed flow rate of the calcium hydroxide slurry; the total slurry volumetric flow rate (unit: L / min) is usually taken as the sum of the CO gas and calcium hydroxide slurry volumetric flow rates at the reactor inlet (ignoring volume changes during the reaction process). Dividing the two yields the maximum number of moles of calcium formate that can be generated per unit volume of reactants.

[0067] The reaction conversion rate X is used to monitor the reaction process of calcium hydroxide to calcium formate in real time and to evaluate the overall efficiency of the carbonylation reaction. X serves as an important indicator of whether the catalyst activity is normal and whether the reaction conditions are optimal: when X is consistently low, it may indicate catalyst deactivation, excessive CO gas source impurities, or unsuitable temperature and pressure, which can trigger alarms or automatically adjust process parameters; simultaneously, X is combined with the product concentration C soft_out It allows for a more comprehensive assessment of production efficiency, providing real-time data support for raw material ratio optimization and output calculation. In addition, online acquisition of conversion rate helps to promptly detect the trend of side reactions (such as the water-gas conversion of CO and water to produce CO2), thereby avoiding the risk of calcium carbonate precipitation and ensuring the long-term stable operation of the unit.

[0068] The output C of the hybrid soft sensor model in step S4 soft_out The feedback signal is input into the closed-loop control loop that sets the reactor temperature. The control logic and method are as follows: 1) Quality Control Objectives: The concentration of calcium formate varies depending on the product grade. The quality standard for industrial-grade calcium formate is 85% ± 2% by mass, and the quality standard for feed-grade calcium formate is 94% ± 1.5% by mass. This control method sets C when targeting feed-grade products. target =94%.

[0069] 2) Lower and upper quality thresholds: Lower quality threshold Ltarget =C target -2.0% (for feed grade L) target =92.0%), upper limit of quality threshold U target =C target +1.0% (for feed grade U) target =95.0%, because exceeding 96% may generate byproducts such as calcium carbonate or hydrated crystals, affecting product flowability. When C soft_out <L target When C is high, it indicates that the reaction conversion rate is low; when C is low, the reaction conversion rate is low. soft_out >U target When this occurs, it indicates a possible overreaction or exacerbation of side effects.

[0070] 3) Calculation of control adjustment amount: ΔT set =K p ·(C target -C soft_out ).in: ΔT set The adjustment amount for the reaction temperature setpoint (unit: °C); K p C is the proportional gain coefficient (unit: °C / % concentration), ranging from 0.6 to 1.2. target Target product mass concentration (set according to product brand); C soft_out The real-time concentration value of calcium formate output by the hybrid soft measurement model (after Kalman filter correction).

[0071] The ΔT set The calculation formula is directly derived from the proportional feedback control law (P control) in classical control theory. Its form is reasonable and effective for the calcium formate carbonylation reaction process because: within a small range near the operating point, the relationship between reaction temperature and outlet concentration can be approximated as a monotonic linear relationship (increasing temperature leads to a faster reaction rate and higher concentration). Therefore, directly and linearly adjusting the temperature setpoint using the concentration deviation can achieve negative feedback closed-loop control. Although the effect of temperature on the reaction rate in actual reaction kinetics is exponential (Arrhenius equation), under the premise of small deviations, linear approximation is a universally accepted and stable simplification in engineering, avoiding overly complex nonlinear controllers.

[0072] proportional gain coefficient K p The tuning should satisfy the following safety boundary constraints: T reactor ∈[140℃, 200℃].

[0073] The adjusted temperature should not exceed the safe operating range of 140~200℃ (based on catalyst thermal stability test data). The optimal temperature window for catalyst activity in the carbonylation synthesis of calcium formate is 160~190℃. When the temperature exceeds 200℃, the catalyst deactivates rapidly, and when the temperature is below 140℃, the reaction rate is too low, affecting the conversion rate and production capacity.

[0074] K p The engineering tuning method involves measuring the reactor's temperature step response curve through experiments on the reactor's thermal response characteristics to determine the reactor's thermal response time constant τ. t The time is approximately 300–600 s (depending on the reactor volume and heat exchanger jacket area). K is calculated using the Ziegler-Nichols closed-loop tuning method. p The initial value is 0.8, and it is fine-tuned based on the overshoot and response time of the soft-sensor closed-loop control during field commissioning. K is triggered when the temperature overshoot exceeds 5°C or the soft-sensor residual consistently exceeds the threshold. p Dynamic correction.

[0075] Temperature regulation commands are executed through the temperature controller of a DCS (Distributed Control System). The controller outputs corresponding valve position signals to adjust the flow rate of cooling water or heating steam in the reactor, thereby achieving closed-loop control of the reaction temperature.

[0076] Step Six: Online monitoring of gas source and adaptive model update.

[0077] To address the slow deactivation of the catalyst caused by impurities such as iron carbonyl (Fe(CO)5) in the CO gas source, and the risk of other trace impurities (H2S, COS, O2, etc.) in the converter gas affecting the catalyst, an online CO gas quality monitoring module was set up.

[0078] Selection and Installation of Carbonyl Iron Analyzer: Install an online carbonyl iron analyzer (infrared gas analyzer or gas chromatograph recommended) on the feed line before CO enters the reactor. Specific requirements: Detection range 0~100 mg / m³ 3 (Calculated as Fe), accuracy ±2%FS, response time <60s, with automatic zero-point calibration and range calibration functions. The carbonyl iron analyzer collects carbonyl iron concentration data C every 30 minutes. Fe (Unit: mg / m³).

[0079] The introduction of the activity decay factor α: when the carbonyl iron concentration exceeds the threshold T. Fe =5mg / m 3 When, or when the cumulative integral of carbonyl iron exceeds the cumulative threshold T Fe_int =10mg·h / m 3 At that time, the system automatically introduces an activity decay factor α (0 < α ≤ 1) into the rate equation of the mechanistic model, and the corrected reaction rate is: The value of α is calculated using the exponential decay model: α = exp[-β·∫C Fe ·dt].

[0080] Where: β is the attenuation coefficient (unit: m) 3 The concentration (mg / h) was determined through offline catalyst activity evaluation experiments. The reference range for this process is 0.01~0.05 mg / h. 3 / (mg·h), the specific value depends on the catalyst formulation and reaction conditions. The attenuation coefficient β is determined by calibrating through catalyst activity-accelerated deactivation experiments. The specific steps include: in a laboratory high-pressure reactor, using the same catalyst and reaction conditions (temperature, pressure, CO partial pressure, slurry composition) as in an actual industrial reactor, continuously and quantitatively adding carbonyl iron (Fe(CO)5) to the CO gas source while maintaining a constant C concentration. Fe,test (Typical value is 5~20 mg / m³) 3 The formation rate of calcium formate or the conversion rate of calcium hydroxide were measured at fixed intervals (e.g., every 12 hours) to obtain the catalyst activity as a function of cumulative carbonyl iron exposure, S = ∫C Fe The decay curve of dt is obtained; a nonlinear regression is performed on the relationship between the activity retention rate a = r / r0 (r0 is the initial rate) and the cumulative exposure S, and an exponential decay model a = exp(-β·S) is fitted to obtain an estimate of β. This experiment is repeated at least three times, and the average value is taken as the β value under the process conditions. When the catalyst formulation, support, or reaction temperature changes, β needs to be recalibrated according to the above method to ensure the accuracy of the activity decay factor.

[0081] ∫C Fe ·dt is the integral of the cumulative carbonyl iron concentration over time from the start of catalyst operation (unit: mg·h / m). 3 The trapezoidal numerical integration method is used for real-time calculation.

[0082] The rationale for calculating the value of α using the exponential decay model: assuming the catalyst activity decay rate is related to the instantaneous impurity concentration C. Fe The concentrations are directly proportional, and the decay process conforms to first-order deactivation mechanics—that is, the number of poisoned active sites per unit time is directly proportional to the number of remaining active sites, and the total impurity exposure accumulated in integral form is ∫C. Fe ·dt is proportional to the total amount of iron deposited on the catalyst surface, while the exponential function reflects the gradual coverage of active sites according to an exponential law. This form ensures that α is always between 0 and 1, and decreases monotonically from 1 to 0 as the cumulative exposure increases. The physical meaning is clear, and there is no need to introduce multiple adjustable parameters, which is convenient for engineering calibration.

[0083] The linear product form of the α factor is based on the engineering assumption that the coverage of carbonyl iron on the active sites of the carbonylation catalyst surface is approximately proportional to the cumulative exposure of carbonyl iron. Its physical meaning is that carbonyl iron thermally decomposes on the catalyst surface to generate metallic iron, which occupies and covers the active sites, and the catalyst activity is linearly related to the number of remaining active sites. In practical applications, when α < 0.9, a catalyst activity decline warning is triggered (requiring inspection of catalyst regeneration or adjustment of process parameters); when α < 0.7, a short-term shutdown of the unit is recommended to replace the catalyst.

[0084] Data-driven incremental learning of models: The incremental learning process of the data-driven model branch will be automatically initiated when any of the following triggering conditions are met: 1) Cumulative integral of carbonyl iron ∫C Fe ·dt exceeds the cumulative threshold T Fe_int ; 2) Error of N consecutive soft measurements (|C) true -C soft | / C true ) Exceeds the set value (e.g., error > 5% for 5 consecutive times); 3) Periodic fine-tuning will be mandatory every 168 hours (7 days) of accumulated running time.

[0085] The incremental learning method employs a "transfer learning" (fine-tuning) strategy: freezing the weights of the first two hidden layers of the BP neural network and updating only the weights of the output layer (i.e., fine-tuning only the linear combination coefficients of the last layer of the model), with the loss function consistent with the error function used during model training. The minimum batch size for incremental learning is 50 new samples. Those skilled in the art know that the commonly used error function in neural network regression tasks is the mean squared error (MSE, which is the sum of the squares of the differences between the predicted and target values ​​divided by the number of samples), and the training process described above explicitly uses the backpropagation algorithm; the loss function is a conventional default choice and will not be elaborated further.

[0086] As reactor operating time increases, time-varying factors such as slow catalyst deactivation and fluctuations in CO gas source impurities cause the process characteristics reflected in the original training data to gradually drift, leading to a continuous increase in the prediction bias of the previously fixed neural network model. The purpose of incremental learning is precisely to fine-tune the network weights using newly acquired recent operating data (e.g., updating only the output layer or the last few layers) without discarding the effective knowledge already learned. This allows the model to quickly adapt to changes in catalytic activity and feedstock properties, thereby ensuring that the soft measurement method of this invention maintains continuous soft measurement accuracy throughout its entire lifecycle. This avoids significant drift in the soft measurement model output caused by periodic fluctuations in feedstock CO quality, catalyst replacement, or process adjustments, and avoids the high costs and data storage pressure associated with periodic retraining.

[0087] Current methods passively correct reaction rates based solely on the current value of the α factor, without providing a predictive model for the remaining catalyst operating time. In actual production, when α falls below a certain threshold (e.g., 0.6), reactor efficiency drops significantly, necessitating shutdown for catalyst replacement or regeneration. However, premature replacement leads to catalyst waste (high cost), while delayed replacement results in decreased yield and increased energy consumption. Currently, there is a lack of a model that dynamically predicts remaining lifespan based on α decay history, current load, and raw material impurity fluctuations, as well as a decision-making method to calculate the economically optimal replacement timing (e.g., considering catalyst cost, shutdown losses, and product value). Solving this problem can reduce catalyst consumption and minimize unplanned shutdown losses.

[0088] Based on this, the present invention further includes step S6: predicting the remaining catalyst lifetime and optimizing replacement. Specifically, it includes: Step S61: Calibration of the catalyst economic death threshold.

[0089] Catalyst economic death threshold α eco This refers to the activity decay factor value at which the average daily economic loss caused by continuing to use the current catalyst is exactly equal to the average daily amortized cost of replacing it with a new catalyst. Below this threshold, continued use is no longer economical; above this threshold, use should be extended as much as possible to amortize catalyst costs.

[0090] Calibration method: Before the unit is first put into operation or before each batch of new catalysts is put into use, the following economic parameters are determined through offline calculations: Catalyst replacement cost C cat (Including new catalyst procurement costs, loading and unloading labor costs, downtime losses, start-up and commissioning costs, etc.), in yuan.

[0091] The market unit price of calcium formate products is P. prod The unit is yuan / ton.

[0092] The unit energy cost E required to maintain normal operation of the reactor cost The unit is yuan / ton of calcium formate.

[0093] The reactor is designed to have a maximum capacity Q. max The unit is tons per day.

[0094] Within the range where the activity decay factor α decreases from 1 to 0.5, using the soft sensor model in steps S3-S4 and the real-time calculated value of α, the daily average output decline rate and the energy consumption increase rate per ton of product corresponding to different α values ​​are statistically analyzed. Specifically: When α=1, the reactor can achieve full-load production under rated energy consumption; when α decreases to a certain value... x At that time, in order to maintain the product concentration at the standard, the control system automatically increased the reaction temperature, resulting in an increase of Δ in steam consumption per ton of product.E (α), while the maximum achievable output decreases to Δ due to residence time limitations. Q (α).

[0095] The economic loss L(α) from continuing to use the catalyst for one more day can be expressed as: L(α)=Δ Q (α)·Q max ·P prod +Δ E (α)·Q actual ·E cost .

[0096] The first item is the decrease in sales revenue due to reduced production, and the second item is the additional costs (Q) caused by increased energy consumption. actual (This refers to actual production). The daily amortized cost of the catalyst, D... cat =C cat / T expected T expected This represents the expected total number of operating days for this batch of catalyst (estimated based on historical experience or accelerated experiments using twin reactors).

[0097] Let L(α) = D cat Solving for α gives the economic mortality threshold α. eco The calibration process is performed once during the initial operation of the unit, and then recalibrated quarterly or when there are significant changes in the catalyst formulation or feedstock prices.

[0098] Step S62: Dynamic prediction of the remaining effective life of the catalyst.

[0099] Define the remaining effective lifetime L of the catalyst rem For: at the current activity decay factor α now If the catalyst is not replaced, it will continue to degrade to α. eco The remaining runtime required.

[0100] Based on the above exponential decay model α=exp(–β·∫C Fe Let dt), and the current accumulated carbonyl iron exposure be dt. Then the current activity α now =exp(–β·S now Let S be the cumulative exposure level at which the economic mortality threshold is reached. eco =-ln(α eco If ) / β, then the remaining tolerable carbonyl iron exposure ΔS=S eco –S now .

[0101] Average accumulation rate of carbonyl iron in the future It can be obtained in one of the following two ways: Short-term forecast: Take the moving average of carbonyl iron concentration over the past 72 hours, and take into account the trend of raw material quality changes (if an online carbonyl iron analyzer is installed, read the real-time value directly).

[0102] Long-term suppression: If the CO gas source comes from the purification of converter gas, its carbonyl iron content often shows periodic fluctuations, which can be predicted using historical data from the same period or time series models (such as exponential smoothing).

[0103] Then the remaining lifespan The unit is hour or day.

[0104] The forecast is automatically updated every 24 hours, and the results are written to the operator interface of the DCS system as the basis for replacement decisions.

[0105] Step S7-3: Dynamic optimization of the optimal replacement timing.

[0106] After obtaining the remaining useful life prediction, the system does not simply suggest "use until the end of the useful life", but introduces a dynamic optimization window to cope with changes in production plans, market environment and raw material quality.

[0107] The optimization logic is as follows: 1) Daily monitoring: The system automatically calculates the average daily economic loss L(α) of continuing to use the catalyst. now ) and the daily diluted cost of catalyst D cat The ratio. When the ratio is ≥0.8, the system enters "early warning status"; when it is ≥1.0, "economic replacement recommendation" is triggered.

[0108] 2) Short-term rolling optimization: Under early warning status, the system substitutes the production plan (planned output, planned shutdown time, and expected carbonyl iron concentration) for the next 48 hours into the economic loss model, simulating the cumulative net profit difference between the two strategies of continued use and immediate replacement hourly. If the cumulative net profit of continued use is lower than that of immediate replacement, the system outputs the "optimal replacement time window" (e.g., "It is recommended to arrange a shutdown for replacement between 8:00 on March 25 and 20:00 on March 26").

[0109] 3) Safety boundary constraints: Regardless of economic considerations, when α decays to below 0.4 (this value is determined by the technical manual provided by the catalyst supplier; below this value there is a risk of runaway reaction or a sharp increase in side reactions), the system will forcibly issue an "emergency replacement" alarm. At this point, the economic loss model is no longer applicable, and safety takes the lead.

[0110] Step S7-4: Closed-loop implementation and benefit evaluation of decision results.

[0111] The system will send the optimal replacement timing decision to the DCS operator station in the form of a "recommendation command," and at the same time provide the following auxiliary information: Current residual value of catalyst (calculated based on future net profit, in yuan); The estimated additional losses for delaying replacement by one day (in yuan / day); The depreciation cost of replacing the catalyst one day ahead of schedule.

[0112] Operators can confirm or reject the suggestion. The system records the deviation between each decision and the actual implementation result, and continuously optimizes the parameters of the economic model (such as α) through an incremental learning mechanism. eco (e.g., β, etc.) to make subsequent predictions more accurate.

[0113] The technical solution in step S6 of this invention further enhances the economic optimization function of catalyst life, which can extend the effective service life of the catalyst and save the annual catalyst procurement cost; at the same time, it reduces the risk of unplanned shutdowns and avoids product quality fluctuations or abnormal reactions caused by excessive use of catalysts; and improves the level of production management refinement by transforming experience judgment into a quantifiable and traceable mathematical model.

[0114] To ensure the successful implementation of this patented technology in the industrial and chemical fields, the following detailed information on key process conditions and raw materials is disclosed: 1. Range of key process parameters: CO / Ca(OH)2 molar ratio: 1.2~2.0 (preferably 1.5~1.8); Reaction temperature: 140~200℃ (preferably 160~190℃); Reaction pressure: 2.0~4.0MPa (preferably 2.5~3.5MPa); Solid content of calcium hydroxide slurry: 5%~20% (mass fraction), preferably 8%~15% (mass fraction); Catalyst addition amount: 0.5%~2.0% of calcium hydroxide by mass (mass fraction); The pH value of the slurry should be controlled within the range of 8.5 to 11.0.

[0115] 2. Pretreatment of converter gas and carbide slag: The CO gas source is converter gas, which is purified by a Peking University Pioneer PSA pressure swing adsorption unit to a CO purity ≥ 98.5% (volume fraction). The typical composition of converter gas is: CO 50%~70%, CO2 15%~20%, N2 10%~15%, O2 0~2%, and trace amounts of H2S (<10ppm), COS (<5ppm), and iron carbonyl (Fe(CO)5, typically <20mg / m³). 3Impurities such as CO gas source fluctuations are present. The short-term impact of CO gas source fluctuations on catalyst activity is compensated by the activity decay factor in step S6; the long-term impact is adaptively addressed by the incremental learning mechanism.

[0116] The calcium source is calcium carbide slag solid waste. The pretreatment steps for calcium carbide slag include: iron removal (magnetic separation to remove metallic impurities) → sieving (filtering large particles through a 200-mesh sieve) → water washing and sedimentation (removing silicates, sulfides, and other impurities; sedimentation time ≥ 4 hours) → slurrying (mixing the precipitated calcium hydroxide with water in a certain proportion to form a slurry) → concentration adjustment (adding water to adjust the solid content to 5%~15% by mass). The iron content in the calcium carbide slag slurry is controlled to <50ppm to prevent iron ions from affecting catalyst performance and to avoid the derivation of Fe(CO)5. The pretreated calcium hydroxide slurry and high-purity CO are continuously fed into the reactor group for carbonylation reaction.

[0117] 3. Composition and preparation of catalysts: The catalyst for the carbonylation synthesis of calcium formate is a supported transition metal catalyst (preferably Ni or Fe-based supported catalysts, with γ-Al₂O₃ or activated carbon as the support). The catalyst composition, by mass fraction, is: active metal content 5%–15%, support 85%–95%. The catalyst preparation method includes an equal-volume impregnation method, where a metal salt solution is impregnated onto a γ-Al₂O₃ support, followed by drying (110℃, 8h), calcination (400℃, 4h), and reduction (H₂ atmosphere, 350℃, 6h). The catalyst is packed in the reactor in a fixed bed or suspended state, using a single reactor or multiple reactors in series. The catalyst activity decay model parameter β is calibrated through offline catalyst aging experiments: the catalyst is placed in a CO atmosphere with a constant carbonyl iron concentration, and samples are taken every 24h to evaluate the calcium formate conversion rate, measure the activity data, and fit to obtain β = 0.025m. 3 / (mg·h) (This value applies to the reaction conditions and catalyst formulation of this process).

[0118] 4. Online measurement of slurry pH value: Due to the high solids content of the reaction slurry (approximately 15%–30% by mass), to avoid clogging, contamination, and drift of the pH electrode, the online pH measurement device employs a pH meter with automatic cleaning function (such as automatic ultrasonic cleaning or high-pressure water jet cleaning, with a cleaning cycle of 4 hours / time). The pH electrode is installed on the circulation pipeline, with a measurement range of 2–12 pH, an accuracy of ±0.1 pH, a temperature compensation range of 0–130℃, and samples taken every 30 seconds, with a 5-minute average value used for soft measurement input.

[0119] pH is crucial for the carbonylation reaction environment: when the free Ca(OH)2 concentration is too high, resulting in a slurry pH > 11, CO may undergo a side reaction to form calcium carbonate precipitate, causing scaling and blockage in the system; when pH < 8.5, it may mean that available Ca...2+ A decrease in concentration leads to insufficient reaction rate and affects the yield of calcium formate. Stable pH monitoring under high solids content conditions is an engineering guarantee for ensuring the continuity and accuracy of soft-sensor input data.

[0120] 5. DCS System Integration Architecture: The soft measurement server (an operator station deployed on the process control network) communicates with the DCS system using the OPC UA (OLE for Process Control Unified Architecture) protocol. Soft measurement calculations are refreshed every 5-10 seconds and written to the DCS real-time database. Temperature control loop: The temperature controller (PID) setpoint SP receives the ΔT output from the soft measurement module via the DCS control program. set The controller outputs a 4-20mA current signal to adjust the opening of the reactor jacket cooling water inlet regulating valve (or the heating steam flow regulating valve). The action response time is <2s, and the entire closed-loop control cycle (from soft-sensor calculation to valve action) is ≤15s. The overall system reliability and real-time performance meet the requirements of carbonyl chemical industry production.

[0121] 6. Safety measures to prevent reactor clogging and scaling: During continuous carbonylation, CO2 accumulation can lead to the formation and deposition of calcium carbonate (CaCO3), causing a chain reaction of problems such as reactor scaling, pH drop, and circulation pipeline blockage. The apparatus of this invention incorporates the following auxiliary engineering measures to ensure the effective implementation of soft sensing and control: An online CO2 analyzer is installed in the exhaust gas recirculation system to monitor the CO2 percentage. If CO2 ≥ 5% (vol), an air purging and replacement operation is triggered. A dual filter (one on and one on standby) is added to the circulation pipeline, which automatically switches to cleaning when the pressure difference is >0.2MPa; Regularly perform low-pressure acid washing on the reactor (1%~2% dilute formic acid solution, pH=2~3, soak for 2 hours) to remove scale blockage caused by CaCO3 precipitation.

[0122] See attached document Figure 2 The present invention also provides a computer device, including: a memory and a processor, and a computer program stored in the memory. When the computer program is executed on the processor, it implements a soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation as described in any of the above methods.

[0123] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. This computer device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that... Figure 2The examples of computer devices are merely examples and do not constitute a limitation on computer devices. They may include more or fewer components than shown in the illustration, or combinations of certain components, or different components. For example, they may also include input / output devices, network access devices, etc.

[0124] The processor referred to can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0125] In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard drive or RAM. In other embodiments, the memory may be an external storage device of the computer device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory may include both internal and external storage units of the computer device. The memory is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory can also be used to temporarily store data that has been output or will be output.

[0126] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a soft measurement and quality control method for a continuous reaction process of calcium formate carbonylation as described in any of the above methods.

[0127] In this embodiment, if the integrated control unit is implemented as a software functional unit and sold or used as an independent product for soft measurement and quality control of a continuous calcium formate carbonylation reaction process, it can be stored in a computer-readable storage medium specifically designed for soft measurement and quality control of a continuous calcium formate carbonylation reaction process. Based on this understanding, all or part of the processes in the above embodiments of this application can be implemented by instructing relevant hardware for soft measurement and quality control of a continuous calcium formate carbonylation reaction process through a specific computer program. This computer program can be stored in a dedicated computer-readable storage medium for soft measurement and quality control of a continuous calcium formate carbonylation reaction process. When executed by a processor, this computer program can implement the application steps of the above-described method embodiments in soft measurement and quality control of a continuous calcium formate carbonylation reaction process. The computer program includes computer program code for soft measurement and quality control of a continuous calcium formate carbonylation reaction process, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium may include at least: any entity or device capable of carrying computer program code for soft measurement and quality control of a continuous reaction process of calcium formate carbonylation to a soft measurement and quality control device for a continuous reaction process of calcium formate carbonylation, recording media, machine tool computer memory, read-only memory (ROM), random access memory (RAM), and other media suitable for distributing software for soft measurement and quality control of a continuous reaction process of calcium formate carbonylation, such as a dedicated soft measurement and quality control card for a continuous reaction process of calcium formate carbonylation, a data storage card, etc.

[0128] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0129] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0130] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for soft-sensing and quality control of a continuous reaction process of calcium formate carbonylation, characterized in that, Includes the following steps: S1. Real-time acquisition of process variable data during the continuous reaction of calcium formate carbonylation, including the true value of calcium formate concentration C obtained from offline laboratory analysis. true ; S2, calculating the calcium formate concentration C at the reactor outlet based on the process variable data 机理 ; S3. Train a prediction model based on process variable data to predict the true value of calcium formate concentration C in real time. true With the calculated value C 机理 The difference ΔC between them yields the baseline value C for soft measurement of calcium formate concentration. soft :C soft =C 机理 +ΔC, and for C soft Perform optimal estimation and correction using Kalman filtering to obtain C. soft_out ; S4, according to C soft_out Real-time calculation of reaction conversion rate X, based on C soft_out With target concentration C target The reaction temperature is automatically adjusted according to the deviation, realizing real-time closed-loop control of product quality.

2. The soft-sensing and quality control method for the continuous reaction process of calcium formate carbonylation according to claim 1, characterized in that, The process variable data includes auxiliary variables and quality control variables. The auxiliary variables include: CO gas flow rate at reactor inlet, calcium hydroxide slurry flow rate at reactor inlet, temperature of multi-stage reactor (T1, T2, ..., Tn), pressure inside reactor, composition of reactor tail gas, reactor stirring speed, effective volume of reactor, pH value of circulating slurry, and concentration of carbonyl iron in CO gas source. The quality control variable comprises a true value C of the calcium formate concentration obtained by an offline assay analysis true and the reaction conversion rate; Between steps S1 and S2, a data preprocessing step is also included: the process variable data collected in real time during the continuous reaction of calcium formate carbonylation are cleaned, outlier removed, and normalized. The normalization formula is as follows: X norm =(XX min ) / (X max -X min ), where X norm X represents the normalized data value, and X represents the original data value. min and X max These are the minimum and maximum values ​​of the variable in the training sample set, respectively.

3. The soft-sensing and quality control method for continuous reaction process of calcium formate carbonylation according to claim 2, characterized in that, Step S2 specifically includes: S21, the rate of the calcium formate carbonylation reaction r is calculated: ; Where r is the reaction rate, A is the pre-exponential factor, Ea is the activation energy of the reaction, R is the ideal gas constant, T is the absolute temperature of the reaction, exp is the natural exponential function, and P is the reaction rate. CO ρ is the partial pressure of CO in the reactor, and n is the reaction order of CO. S22. Calculate the calcium formate concentration C at the reactor outlet. 机理 :C 机理 =(r·τ·ρ slurry )·E reactor ·X Ca ; Where C 机理 ρ is the concentration of calcium formate at the reactor outlet, τ is the average residence time of the reactants in the reactor, and ρ is the concentration of calcium formate at the reactor outlet. slurry The density of the reaction slurry, E reactor X is the reactor efficiency coefficient. Ca The initial conversion rate of calcium hydroxide is given.

4. The soft-sensing and quality control method for continuous reaction process of calcium formate carbonylation according to claim 3, characterized in that, Step S3 uses a BP neural network to predict ΔC: the true value of calcium formate concentration C obtained from offline laboratory analysis. true The calculated concentration of calcium formate at the reactor outlet, C 机理 The difference between the values ​​is used as the target output, and the auxiliary variables at the corresponding time points are used as input samples to construct training and validation sets (80% for training and 20% for validation). The network weights are updated using the backpropagation algorithm until the convergence condition is met.

5. The soft-sensing and quality control method for continuous reaction process of calcium formate carbonylation according to claim 4, characterized in that, The BP neural network structure used is as follows: Input layer: K nodes, where K corresponds to the number of auxiliary variables collected. This device uses 13 auxiliary variables: temperature of each reactor (including four typical reactors, T1~T4), reaction pressure P, and CO gas feed flow rate F. CO Calcium hydroxide slurry feed flow rate F Ca Stirring speed R, pH value of circulating slurry, carbonyl iron concentration C Fe CO content in exhaust gas Y CO And exhaust CO2 content Y CO2 That is, K=13; Hidden layer: H nodes, set H=2K+1=27; Output layer: 1 node, outputs the correction amount ΔC for calcium formate concentration; Activation function: sigmoid function f(x) = 1 / (1+e -x ) is used in the hidden layer, and linear activation function identity is used in the output layer; Training method: The Levenberg-Marquardt algorithm is used for weight optimization. The learning rate μ is initially set to 0.

1. As the training error decreases, μ is dynamically adjusted. The number of training epochs is 500. The termination condition is that the validation set error no longer decreases for 10 consecutive epochs or the maximum number of training epochs is reached.

6. The soft-sensing and quality control method for continuous reaction process of calcium formate carbonylation according to claim 1, wherein, When a new offline test analysis sample C is obtained true is not aligned with the time of C soft , the optimal estimation correction is made to C soft by a Kalman filtering algorithm to obtain C soft_out ; the Kalman filtering recursive formula is as follows: State prediction: ； Covariance prediction: ； Kalman gain calculation: ； State update (C soft_out = updated estimate ): ； Covariance update: ； Initial values ​​for recursion: Set as the current C soft Value; the initial covariance of P0 is set to 1 (indicating a large uncertainty in the initial estimate). Kalman filter correction allows for high-precision soft measurement output C even when offline test results lag by approximately 1 hour. soft_out ; Where k is the k-th offline sampling time (k=1,2,3,…); The state prediction based on the information from the previous step, i.e., C before Kalman filter correction. soft The estimated value; The state estimate after Kalman filtering correction represents the final soft-sensor output C. soft_out F is the state transition matrix, set to 1, indicating that the mass concentration shows no trend within adjacent sampling intervals; H is the observation matrix, set to 1, indicating that offline measurements and predicted values ​​can be directly compared under the same dimensions; Q is the process noise covariance, set to 0.

001. 2 R represents the observation noise covariance. Based on the measurement standard deviation of the offline laboratory instrument, the standard deviation is set to 0.5%~1.0%, and R = 0.005 is chosen. 2 ;P k I is the posterior estimation error covariance matrix; I is the identity matrix; C true_k This represents the true value of calcium formate concentration obtained from the k-th offline test.

7. The method for soft measurement and quality control of a continuous reaction process for the carbonylation of calcium formate as described in claim 6, characterized in that, The reaction conversion rate X is indirectly obtained from the ratio of the real-time concentration of calcium formate to the maximum theoretical concentration: X = C soft_out / C max C max The calculation is based on the assumption that all calcium hydroxide feed is converted into calcium formate.

8. The method for soft measurement and quality control of a continuous reaction process for calcium formate carbonylation as described in claim 6, characterized in that, Based on C soft_out With target concentration C target The formula for automatically adjusting the reaction temperature based on the deviation is: ΔT set =K p ·(C target -C soft_out ), where ΔT set K is the adjustment amount for the reaction temperature setpoint. p C is the proportional gain coefficient. target The target product quality concentration; Proportional gain coefficient K p The tuning of the proportional gain coefficient K reactor ∈ [140°C, 200°C], where T reactor is the measured temperature of the reactor.

9. The soft-sensing and quality control method for continuous reaction process of calcium formate carbonylation according to claim 6, characterized in that, It also includes step S5, online monitoring of gas source gas quality and adaptive model updating: an online carbonyl iron analyzer is installed on the feed pipeline before CO enters the reactor. When the carbonyl iron concentration is detected to exceed the threshold T, the analyzer will update the gas source gas quality and the model adaptive update. Fe When, or when the cumulative integral of carbonyl iron exceeds the cumulative threshold T Fe_int At that time, the system automatically introduces an activity decay factor α (0 < α ≤ 1) into the rate equation of the mechanistic model, and the corrected reaction rate is: The value of α is calculated using the exponential decay model: α = exp[-β·∫C Fe ·dt]; Where β is the attenuation coefficient, C Fe For carbonyl iron concentration, ∫C Fe ·dt is the cumulative carbonyl iron concentration over time from the start of catalyst operation, calculated in real time using the trapezoidal numerical integration method.

10. The soft-sensing and quality control method for continuous reaction process of calcium formate carbonylation according to claim 9, wherein, The incremental learning process will automatically start when any of the following trigger conditions are met: 1) the integrated amount of the carbonyl iron accumulation ∫C Fe exceeding the accumulation threshold T Fe_int ; 2) the soft measurement error (|C true - C soft | / C true ) exceeds a set value for N consecutive times; 3) Periodic fine-tuning will be mandatory every 168 hours (7 days) of accumulated running time; The incremental learning method employs a "transfer learning" strategy: freezing the weights of the first two hidden layers of the BP neural network and updating only the weights of the output layer (i.e., fine-tuning only the linear combination coefficients of the last layer of the model). The loss function is consistent with the error function during model training, and the minimum batch size for incremental learning is 50 new samples.

Images