A real-time prediction method for the glucose content of the reaction solution during the production process

Abstract

The present invention belongs to the technical field of gluconate production, and particularly relates to a method for real-time prediction of the glucose content in the reaction solution during the production process. The present invention selects variables that can be measured during the production process, collects and preprocesses data, establishes a mathematical relationship or model through historical data, and uses software or algorithms to replace hardware functions to achieve real-time prediction of the glucose content in the reaction solution during the production process. The method and device disclosed by the present invention can achieve real-time prediction of the glucose content in the reaction solution during the production process, and the prediction results can meet the requirements of actual production.

Description

Technical Field

[0001] The present invention belongs to the technical field of gluconate production, and particularly relates to a method for real-time prediction of glucose content in a reaction solution during the production process. Background Art

[0002] Gluconates such as zinc gluconate, calcium gluconate, and ferrous gluconate are important food and pharmaceutical additives and are widely used in health products, medicine, and food, mainly for the supplementation and treatment of deficiencies in elements such as zinc, calcium, and iron; gluconates also have their respective extensive applications in the industrial field.

[0003] The main preparation process methods of gluconates in industrial production can generally be divided into ordinary chemical methods, catalytic oxidation methods, electrolytic oxidation methods, biochemical methods, double enzyme methods, etc. Among them, the double enzyme method is a relatively new process in recent years, and its advantages such as simple production operation, energy conservation and consumption reduction, and high product purity have been gradually recognized. Taking zinc gluconate as an example, the main principle of producing zinc gluconate by the "double enzyme method" is to use glucose oxidase and catalase to oxidize glucose into gluconic acid, and gluconic acid reacts with zinc oxide to form zinc gluconate. Later, the zinc gluconate solution is filtered, concentrated, crystallized, and dried to obtain the finished product. The glucose content in the reaction solution during the production of zinc gluconate by the double enzyme method is an important indicator to measure the reaction progress. Too high glucose content indicates that the reaction has not been fully carried out, resulting in a low yield of zinc gluconate finished product and the output not meeting the quality standard, bringing some economic losses to the production enterprise. When the glucose content is too low, the reaction time is prolonged, resulting in an increase in time cost and unnecessary energy waste. In actual production, the reaction progress is mainly determined by monitoring this parameter to control the production progress.

[0004] Currently, the commonly used methods for detecting glucose content are electrochemical detection methods and chemical titration methods. During the production of gluconates by the double enzyme method, the chemical titration method is used to detect the glucose content. This method involves manually sampling the solution in the reaction, and titrating and detecting the sample in the laboratory to obtain the glucose content. The time from sampling to the completion of detection is generally about 20 - 30 minutes, and the interval time between each sampling is generally 1 - 2 hours; the sampling period of this method has a certain impact on the glucose content, and the detection results obtained from each sampling have a significant lag in time; this method cannot reflect the change of glucose content in the production process in real time, resulting in the inability to perform real-time optimization control on the production process, and having many adverse effects on aspects such as output, efficiency, and production energy consumption.

[0005] Electrochemical detection methods are used for glucose content detection. They have advantages such as fast speed and good real-time performance, and are widely used in fields such as fructose detection and blood glucose detection. The detection mechanism is to immobilize glucose oxidase on the surface of the electrode of the detector, so that glucose undergoes an enzymatic reaction with glucose oxidase, is converted into another compound, releases electrons, and the electrons are conducted to the electrode, and the glucose content is indirectly reflected through the potential change. For the reaction process of producing gluconate by the double-enzyme method, since the reaction solution contains glucose oxidase, it has a great impact on the accuracy of the electrochemical detection method with the same principle, making it difficult to detect the glucose content in the reaction solution by the electrochemical detection method.

[0006] Therefore, during the production process of producing gluconate by the double-enzyme method, the glucose content in the reaction solution cannot be detected in real time, and there is no method in the prior art that can reflect the glucose content in the reaction solution in real time. Summary of the Invention

[0007] In view of the above problems, the present invention provides a real-time prediction method for the glucose content in the reaction solution during the production process. This method selects variables that can be measured during the production process, collects and preprocesses the data, establishes a mathematical relationship or model through historical data, and uses software or algorithms to replace the hardware function to achieve real-time prediction of the glucose content in the reaction solution during the production process.

[0008] Based on the above purpose, the present invention provides a real-time prediction method for the glucose content in the reaction solution during the production process, which specifically includes the following steps:

[0009] Step 1: Data collection; Prepare gluconate by the double-enzyme method. When starting the reaction device, start the data collection system at the same time. Collect and record the detection values of each variable every 2 seconds and store them. Each item of detection data is sorted and stored with the collection time as the index. The collected multiple variable data forms the input data set used by the neural network model; Regularly take artificial samples and use the titration method to measure the glucose content value of the reaction solution to form the output data set used by the neural network model; After the measured glucose content value is lower than the given value, end the current reaction and data collection;

[0010] Repeat the above double-enzyme method reaction and data collection multiple times to obtain multiple groups of input data sets and output data sets. The data collected in the first double-enzyme method reaction forms the first group of input data set and output data set in the first stage, and the data collected in the nth double-enzyme method reaction forms the nth group of input data set and output data set in the first stage, where n is the number of double-enzyme method reactions and n≥2;

[0011] Step 2. Data storage method: The automatically stored collected data in Step 1 is sorted and numbered with the collection time as the index. The data of each variable collected at the same collection time occupies one column and is arranged in a row to form a data record. Multiple data records form an input variable data set; the manually sampled titration measurement values of the glucose content are stored with the manual sampling time as the index. Select the row where the stored data has the same collection time as the manual sampling time in the input variable data set and fill it into the column where the row is located to form an output data set. The number of data records of the glucose content values in the output data set is less than the number of data records of the input variables in the input data set. The interpolation method is used to generate the missing glucose content value data, forming an output data set corresponding to each data record of the input variable, named the first group to the nth group of input data sets and output data sets in the second stage, where n is the number of glucose oxidase-catalase reaction times and n≥2;

[0012] Step 3. Data collection of the natural heat dissipation of the reactor and its accessories: Glucose oxidase and catalase are removed from the raw materials, and the other raw materials are exactly the same as those in Step 1. All reaction conditions and process parameters are exactly the same as those in Step 1. Start the reaction device and the data collection system. After the temperature of the raw material liquid is the same as the room temperature, continue to collect data for 0.5 hours, and then stop the operation of the reaction device and the data collection system to end the data collection; the storage method of the data collected in this step is similar to that in Step 2, and after storage, it is named the variable data set in the first stage of approximate value measurement; if there are no changes in the reaction raw materials and reaction conditions during the multiple glucose oxidase-catalase reaction processes and data collection processes in Step 1, the data collection in this step can be carried out only once;

[0013] Step 4. Merging of the best time period data: According to the determined best time period, process the data in the first group to the nth group of input data sets and output data sets in the second stage to form a data result set with a longer time period, named the first group to the nth group of input data sets and output data sets in the third stage (n is the number of glucose oxidase-catalase reaction times and n≥2); according to the same best time period, process the data in the variable data set in the first stage of approximate value measurement to form a data result set with a longer time period, named the variable data set in the second stage of approximate value measurement;

[0014] Step 5. Data calculation: Calculate some variable data in Step 4 to obtain new variables and variable data, and use the serial number of each row of data in the first group to the nth group of input data sets and output data sets in the third stage as the index to merge and store them together with the original data in the first group to the nth group of input data sets and output data sets in the third stage, forming the first group to the nth group of input data sets and output data sets in the fourth stage, where n is the number of glucose oxidase-catalase reaction times and n≥2;

[0015] Step 6. Data processing: For the output data sets of the 1st to nth groups in the fourth stage of Step 5, perform normalization processing on them. Name the obtained input variable data set and output variable data set as the input data sets and output data sets of the 1st to nth groups in the fifth stage, where n is the number of double-enzyme reaction times and n ≥ 2;

[0016] Step 7. Variable screening: Screen each input variable in the input data sets and output data sets of the 1st to nth groups in the fifth stage, and select the variables and data with a high correlation degree with the glucose content value to form an input variable data set. The original glucose content value data is the output variable data set, forming the input data sets and output data sets of the 1st to nth groups in the sixth stage, where n is the number of double-enzyme reaction times and n ≥ 2;

[0017] Step 8. Model training: Use the data in the input data sets and output data sets of the 1st to nth groups in the sixth stage as input variables and output variables to train the neural network model to obtain the trained model;

[0018] Step 9. Prediction of glucose content: Use the model trained in Step 8 to predict the glucose content in the reaction solution during a new reaction process. For the reaction process that requires predicting the glucose content in the reaction solution, select the variables for which data needs to be collected according to the types of variables with a high correlation degree screened in Step 7, collect the real-time data of these variables during the new reaction process, and then detect, collect, store, calculate, merge, and process the data of these variables according to Steps 1 to 6. Perform normalization processing on the processed data to form the input variable data for the prediction stage. Input the input variable data for the prediction stage into the model trained in Step 8, calculate to obtain the output value, and the result obtained by performing reverse processing on the output value according to the normalization processing method is the predicted value of the glucose content corresponding to the input variable data for the prediction stage.

[0019] Further, in Step 1, for manual sampling, the glucose content value is detected by chemical titration method in the laboratory and recorded. The manual sampling time period is once every 30 minutes or once every 60 minutes.

[0020] Further, in Step 4, the optimal time period is one of 1 minute, 3 minutes, 5 minutes, 10 minutes, or 30 minutes, and preferably the 3-minute time period. The method of merging the 2-second time period data of variables into 3-minute time period data is to sum the continuous 90 2-second time period data of the variables and then divide by 90. The obtained average value is the merged result. Each row of data in the processed data set is sorted in the order of the original data collection time and numbered, with the number as the index.

[0021] Further, in Step 7, the screening method is the Pearson correlation coefficient method or other existing technologies.

[0022] Further, in the step 8, the neural network model is a BP, LSTM or other type of neural network model.

[0023] Further, the reaction device includes a reaction kettle and instrument sensors, specifically including reaction kettle 1; a jacket 2 is arranged outside reaction kettle 1; reaction liquid 3 is contained inside reaction kettle 1; a motor 7 is arranged at the top of reaction kettle 1; a stirring paddle 4 is connected to the motor 7, and the stirring paddle 4 extends into reaction kettle 1; an air outlet pipeline 10 is arranged above one side of reaction kettle 1, and a water inlet pipeline 9 is arranged below the same side of the air outlet pipeline 10; a water outlet pipeline 11 is arranged above the other side of reaction kettle 1, and an air inlet pipeline 5 is arranged below the same side of the water outlet pipeline 11; a feed inlet 8 is arranged above the water outlet pipeline 11; a discharge outlet 6 is arranged below reaction kettle 1;

[0024] An air outlet temperature sensor 12, a gas flowmeter 13 and an oxygen analyzer 14 are arranged on the air outlet pipeline 10; a kettle internal temperature sensor 15, a pH meter 16, a dissolved oxygen meter 17, a turbidity meter 18 and a conductivity sensor 19 that extend into the reaction liquid 3 are arranged at the top of reaction kettle 1; a kettle internal pressure sensor 20 is separately arranged at the top of reaction kettle 1; a heat meter inlet water temperature sensor 21 and a heat meter 22 are arranged on the water inlet pipeline 9; a heat meter outlet water temperature sensor 23 is arranged on the water outlet pipeline 11; an air inlet temperature sensor 24 is arranged on the air inlet pipeline 5.

[0025] Further, the data acquisition system is composed of a PLC, a touch screen, communication lines and various instrument sensors; the PLC runs acquisition software to read the data monitored by various instrument sensors in real time and displays and stores them through a network cable connected to the touch screen.

[0026] Further, in the step 1, the variable data are:

[0027] A. Acquisition time TM, pressure PR inside the reaction kettle, reaction liquid temperature TRL;

[0028] B. Inlet air temperature TGI, outlet gas temperature TGO of the air introduced into the reaction kettle, cumulative outlet gas flow CFG, oxygen content OG in the outlet gas;

[0029] C. Cumulative inlet flow LW of the circulating water in the jacket of the reaction kettle, inlet water temperature WTI, outlet water temperature WTO;

[0030] D. Acid-base value RPH, dissolved oxygen rate RDO, oxygen saturation value ROS, oxygen partial pressure value ROPP, conductivity RC, resistivity RER, total dissolved substances RTDS, salinity RS, turbidity value RT of the reaction liquid inside the reaction kettle;

[0031] E. Glucose content GC in the reaction liquid;

[0032] F. Natural heat dissipation QR of the reaction kettle and its accessories

[0033] QG, the heat carried away by the gas; QL, the heat carried away by the circulating water; QA, the heat released by the reaction; AQA, the cumulative heat released by the reaction; AQG, the cumulative heat carried away by the gas; AQL, the cumulative heat carried away by the circulating water; AOG, the cumulative oxygen consumption of the reaction;

[0034] Among them, type A variables are conventional variables, all of which are collected and recorded; type B and C variables are partially or fully collected and recorded according to the heating, cooling, and oxygen consumption conditions of the reaction; type D variables are partially or fully collected and recorded according to the reactants and the requirements for monitoring the reaction process; type E variables are measured by manual titration in the laboratory after manual sampling;

[0035] Type F variables cannot be directly detected and need to be further calculated to obtain approximate values after collecting and recording data using the special detection method in step 3; type G variables are further calculated from type A, B, C, F, and G variables.

[0036] Furthermore, in step 3, the variables collected for the natural heat dissipation of the reactor and its accessories are:

[0037] H, the collection time TTM, the pressure TPR in the reactor, the reaction liquid temperature TTRL;

[0038] I, the inlet air temperature TTGI and the outlet gas temperature TTGO in the reactor, the cumulative outlet gas flow rate TCFG;

[0039] J, the cumulative inlet flow rate TLW, the inlet water temperature TWTI, and the outlet water temperature TWTO of the circulating water in the reactor jacket.

[0040] Furthermore, in step 5, the data calculation method for the natural heat dissipation QR of the reactor and its accessories is:

[0041] Step a: Calculate the variable data in the variable data set of the second stage of the approximate value measurement in step 4 to obtain new variables and variable data, and merge and store them with the original data in the data set with the serial number of each row in the variable data set of the second stage of the approximate value measurement as the index to form the variable data set of the third stage of the approximate value measurement. The calculation methods for the variables and data are as follows:

[0042] 1. Calculate the heat carried away by the gas TQG in the approximate value measurement stage

[0043] TQG i =(TCFG i - TCFG i-1 ) × ρG × CG × (TTGO i - TTGI i )

[0044] Wherein, i is the serial number of the current data record row in the second-stage data set of approximate value measurement, i≥2, and the value of TQG1 is a null value; ρG is the density of air, with a value of ρG = 1.29 kg / m 3 , CG is the specific heat capacity of air, with a value of CG = 1.005 kJ / (kg*K).

[0045] 2. Calculate the heat taken away by the circulating water in the approximate value measurement stage, TQL

[0046] TQL i =(TLW i -TLW i-1 )×ρL×CL×(TWTO i -TWTI i )

[0047] Wherein, i is the serial number of the current data record row in the second-stage data set of approximate value measurement, i≥2, and the value of TQL1 is a null value; ρL is the density of water, with a value of ρL = 1000 kg / m 3 , CL is the specific heat capacity of water, with a value of CL = 4.1829 kJ / (kg*K).

[0048] 3. Calculate the natural heat dissipation of the reactor and its accessories in the approximate value measurement stage, TQR

[0049] TQR i =mc(TTRL i-1 -TTRL i )-TQG i -TQL i

[0050] Wherein, i is the serial number of the current data record row in the second-stage data set of approximate value measurement, i≥2, and the value of TQR1 is a null value; m is the mass of the reaction liquid, and c is the specific heat capacity of the reaction liquid.

[0051] Step b: Obtain the function formula for calculating QR:

[0052] Using the reaction liquid temperature TTRL j in the variable data set of the third stage of approximate value measurement as the input variable and TQR j as the output variable for curve fitting, the obtained second-order polynomial function is:

[0053] TQR j =0.4477×TTRL j 2 -28.82×TTRL j +465.5(1)

[0054] Wherein, j is the serial number of the current data record row of the variable data set in the third stage of approximate value measurement, j ≥ 2, (the value of TQR1 is null and not used), TTRL j is the reaction liquid temperature value of the j-th row of the variable data set in the third stage of the approximate value measurement stage.

[0055] The function (1) is converted into a function for calculating QR as follows:

[0056] QR x = 0.4477 × TRL x 2 - 28.82 × TRL x + 465.5(2)

[0057] Wherein, x is the serial number of the current data record row in the input data set and output data set from the 1st group to the nth group in the third stage, x ≥ 1, TRL x is the reaction liquid temperature value of the x-th data record row in the data set.

[0058] Furthermore, the calculation method of the G-type variable data in step 5 is specifically as follows:

[0059] 1. Heat carried away by gas QG

[0060] QG x = (CFG x - CFG x-1 ) × ρG × CG × (TGO x - TGI x )

[0061] Wherein, x is the serial number of the current data record row in the input data set and output data set from the 1st group to the nth group in the third stage, x ≥ 2, the value of QG1 takes the same value as QG2, ρG is the density of air, and the value ρG = 1.29 kg / m 3 , CG is the specific heat capacity of air, and the value CG = 1.005 kJ / (kg*K)

[0062] 2. Heat carried away by circulating water QL

[0063] QL x = (LW x - LW x-1 ) × ρL × CL × (WTO x - WTI x )

[0064] Wherein, x is the serial number of the current data record row in the input data set and output data set from the 1st group to the nth group in the third stage, x ≥ 2, the value of QL1 takes the same value as QL2, ρL is the density of water, and the value ρL = 1000 kg / m 3, where \(C_L\) is the specific heat capacity of water, with a value of \(C_L = 4.1829\ kJ / (kg\cdot K)\).

[0065] 3. Heat release of the reaction \(Q_A\)

[0066] \(Q_A\) x = \(mc(T_{RL}\) x - \(T_{RL}\) x-1 ) + \(Q_R\) x + \(Q_G\) x + \(Q_L\) x

[0067] In the formula, \(x\) is the serial number of the current data record row in the input data set and output data set from the 1st group to the \(n\)th group in the third stage, \(x\geq2\), and the value of \(Q_{A1}\) takes the same value as \(Q_{A2}\), \(m\) is the mass of the reaction solution, and \(c\) is the specific heat capacity of the reaction solution. The value of \(Q_R\) x is calculated according to the reaction solution temperature \(T_{RL}\) of the current data record row x The value is obtained by calculating through the function (2) in step b. Taking the \(T_{RL}\) x value as the input variable of the function, substituting it into the function (2), and calculating the corresponding \(Q_R\) x value.

[0068] 4. Cumulative heat release of the reaction \(AQ_A\)

[0069] \(AQ_A\) x = \(Q_{A1}+Q_{A2}+Q_{A3}+\cdots+Q_A\) x-1 + \(Q_A\) x

[0070] In the formula, \(x\) is the serial number of the current data record row in the input data set and output data set from the 1st group to the \(n\)th group in the third stage, \(x\geq1\).

[0071] 5. Cumulative heat carried away by the gas \(AQ_G\)

[0072] \(AQ_G\) x = \(Q_{G1}+Q_{G2}+Q_{G3}+\cdots+Q_G\) x-1 + \(Q_G\) x

[0073] In the formula, \(x\) is the serial number of the current data record row in the input data set and output data set from the 1st group to the \(n\)th group in the third stage, \(x\geq1\).

[0074] 6. Cumulative heat carried away by the circulating water \(AQ_L\)

[0075] \(AQ_L\) x = \(Q_{L1}+Q_{L2}+Q_{L3}+\cdots+Q_L\) x-1 + \(Q_L\) x

[0076] Where x is the serial number of the current data record row in the input data set and output data set of the 1st to nth groups in the third stage, and x≥1.

[0077] 7. Cumulative reactive oxygen consumption AOG

[0078] Since the double-enzyme method reaction is an oxygen-consuming reaction, the measured value of the oxygen content OG in the exhaust gas is less than 21%. The oxygen consumption is 21% minus the measured value OG. CFG x Is the cumulative exhaust gas flow rate of the current data record row, OCG x Is the reactive oxygen consumption of the current data record row.

[0079] OCG x =(21 - OG x )×(CFG x - CFG x-1 )

[0080] Where x is the serial number of the current data record row in the input data set and output data set of the 1st to nth groups in the third stage, x≥2, and the value of OCG1 takes the same value as OCG2.

[0081] AOG x = OCG1 + OCG2 + OCG3 + … + OCG x-1 + OCG x

[0082] Where x is the serial number of the current data record row in the input data set and output data set of the 1st to nth groups in the third stage, and x≥1.

[0083] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0084] Aiming at the problem that in the production process of preparing gluconate by the double-enzyme method, the measurement of the glucose content in the reaction solution is manually detected offline, lacking timeliness and convenience and being unfavorable for production guidance, the method and device disclosed by the present invention can realize the real-time prediction of the glucose content in the reaction solution during the production process, and the prediction result can meet the requirements of actual production.

[0085] The present invention has taken improvement measures in the following aspects, improving the accuracy of training the prediction model of the glucose content in the reaction solution, and further improving the accuracy of the prediction result of the glucose content in the reaction solution.

[0086] (1) The reaction for preparing gluconate by the double-enzyme method is an exothermic reaction. The amount of heat released during the reaction process can directly reflect the progress of the reaction and the consumption degree of glucose. The present invention proposes a method for measuring and calculating the natural heat dissipation of the reaction kettle and its accessories. By this method, an approximate value of the natural heat dissipation of the reaction kettle and its accessories is obtained, which is used to complete the calculation of the heat release and the cumulative heat release during the double-enzyme reaction process, so that the reaction heat release can be used as a variable for model training, thereby improving the accuracy of model training.

[0087] (2) Since the variable data collected by the data acquisition system is recorded and stored every 2 seconds, its results reflect the microscopic changes and local fluctuations of the detection instrument and the detected value in more detail, which affects the accurate training of the neural network model and leads to a decrease in the accuracy of the prediction model. The present invention processes the original data collected every 2 seconds according to the selected optimal time period (preferably a 3-minute time period) to form an input variable data set and an output variable data set with a longer time period. The processed data is used for the training of the neural network model, avoiding the influence of the 2-second time period data on the training accuracy of the neural network model.

[0088] (3) The present invention uses the Pearson correlation coefficient method to select highly correlated input variables for the training of the neural network model, removing the interference of low-correlation variables on the neural network model and improving the accuracy of model training.

[0089] In the embodiment, using the method and device disclosed in the present invention, two mathematical models, namely the BP neural network and the LSTM neural network, are used for model training and prediction, and two prediction results are obtained. Both prediction results can meet the actual production requirements. Except for the difference in the mathematical models, other factors are the same. By selecting four indicators, namely MSE, RMSE, MAE, and MAPE, to compare and analyze the two prediction results, in terms of the accuracy of the prediction results, the BP neural network model in Embodiment 1 is superior to the LSTM neural network model, and the LSTM neural network model in Embodiment 2 is superior to the BP neural network model. In the actual industrial production process, the result of predicting the glucose content in the reaction solution using the technical solution disclosed in the present invention can be used to judge the state of the reaction process in real time, and adjust the reaction control conditions and parameters in a timely manner according to the reaction situation, which plays a key role in improving the quality of gluconate products, saving enterprise costs, and building an intelligent production line. Description of the Drawings

[0090] Figure 1Schematic diagram of the structure of the reaction device of the present invention and the installation of instrument sensors (where: 1, reaction kettle; 2, jacket; 3, reaction liquid; 4, stirring paddle; 5, inlet gas pipeline; 6, discharge port; 7, motor; 8, feed port; 9, inlet water pipeline; 10, outlet gas pipeline; 11, outlet water pipeline; 12, outlet water pipeline; 13, gas flowmeter; 14, oxygen analyzer; 15, in-kettle temperature sensor; 16, pH meter; 17, dissolved oxygen meter; 18, turbidity meter; 19, conductivity sensor; 20, in-kettle pressure sensor; 21, heat meter inlet water temperature sensor; 22, heat meter; 23, heat meter outlet water temperature sensor; 24, inlet gas temperature sensor).

[0091] Figure 2 Schematic diagram of the composition of the data acquisition system.

[0092] Figure 3 Thermodynamic diagram of the correlation of each input variable in Example 1.

[0093] Figure 4 Comparison between the predicted values and the measured values of the BP and LSTM models in Example 1.

[0094] Figure 5 Thermodynamic diagram of the correlation of each input variable in Example 2.

[0095] Figure 6 Comparison between the predicted values and the measured values of the BP and LSTM models in Example 2. Specific implementation manners

[0096] Some embodiments of the present invention are disclosed below. Those skilled in the art can implement them by appropriately modifying the process parameters according to the content herein. It should be particularly noted that all similar substitutions and modifications are obvious to those skilled in the art, and they are all considered to be included in the present invention. The methods and applications of the present invention have been described through preferred embodiments, and those skilled in the art can obviously make changes or appropriate modifications and combinations to the methods and applications described herein without departing from the content, spirit and scope of the present invention to implement and apply the technology of the present invention.

[0097] The types of variable data involved in the present invention are as follows:

[0098] A. Acquisition time TM, pressure PR in the reaction kettle, and reaction liquid temperature TRL;

[0099] B. Inlet gas temperature TGI, outlet gas temperature TGO of the air introduced into the reaction kettle, cumulative discharge gas flow CFG, and oxygen content OG in the discharge gas;

[0100] C. Cumulative inlet flow LW, inlet water temperature WTI, and outlet water temperature WTO of the circulating water in the reaction kettle jacket;

[0101] D. The pH value RPH, dissolved oxygen rate RDO, oxygen saturation value ROS, oxygen partial pressure value ROPP, conductivity RC, resistivity RER, total dissolved solids RTDS, salinity RS, and turbidity value RT of the reaction liquid in the reactor;

[0102] E. The glucose content GC in the reaction liquid;

[0103] F. The natural heat dissipation QR of the reactor and its accessories;

[0104] G. The heat carried away by the gas QG, the heat carried away by the circulating water QL, the heat released by the reaction QA, the cumulative heat released by the reaction AQA, the cumulative heat carried away by the gas AQG, the cumulative heat carried away by the circulating water AQL, and the cumulative oxygen consumption AOG of the reaction;

[0105] Among them, type A variables are conventional variables, all of which are collected and recorded; type B and C variables are partially or all collected and recorded according to the heating, cooling, and oxygen consumption conditions of the reaction; type D variables are partially or all collected and recorded according to the reactants and the requirements for monitoring the reaction process; type E variables are measured by manual titration in the laboratory after manual sampling;

[0106] Type F variables cannot be directly detected and obtained. It is necessary to collect and record data through the special detection method in step 3 and then further calculate to obtain an approximate value; type G variables are obtained by further calculating type A, B, C, F, and G variables.

[0107] A real-time prediction method for the glucose content during the production process specifically includes the following contents:

[0108] Step 1. Data collection: Install various sensors, detection instruments, and data collection systems on the reaction device, as shown in Figure 1 , Figure 2 . Add raw materials to the reaction device and start the reaction device to start the enzymatic catalytic oxidation reaction for preparing gluconate by the double-enzyme method. Start the data collection system simultaneously when starting the reaction device. After the data collection system runs, it automatically queries the detection data of each instrument sensor, collects and records the detection values of each variable every 2 seconds, sorts and stores the detection data of each item according to the collection time as the index, and the collected multiple variable data forms the input data set used by the neural network model. The glucose content value of the reaction liquid is obtained by manual titration after regular manual sampling, forming the output data set used by the neural network model. Manual sampling of the reaction liquid is performed once at regular intervals, and the glucose content value is detected and recorded by chemical titration in the laboratory. The manual sampling time period is once every 30 minutes or once every 60 minutes. After the glucose content value measured by manual sampling and titration is lower than the given value, the double-enzyme method reaction ends, the reaction device discharges materials, and the operation of the reaction device and the data collection system stops, ending the current reaction and data collection.

[0109] Repeat the above double-enzyme reaction and data collection multiple times. The data collected in the first double-enzyme reaction forms the first-stage group 1 input data set and output data set. The data collected in the nth double-enzyme reaction forms the first-stage group n input data set and output data set, where n is the number of double-enzyme reactions and n≥2. The double-enzyme reaction and data collection can be continued multiple times to obtain multiple groups of data. The total number of times of the double-enzyme reaction and data collection is greater than or equal to 2 times. Collecting a large number of data groups can improve the accuracy of the prediction results.

[0110] Step 2, data storage method: The automatically stored collected data in Step 1 is sorted and numbered with the collection time as the index. The data of each variable collected at the same collection time each occupies a column and is arranged in a row to form a data record. Multiple data records form the input variable data set; the manual sampling titration measurement values of the glucose content are stored with the manual sampling time as the index. Select the row where the stored data with the same collection time as the collection time in the input variable data set is located, and fill in the column where the row is located to form the output data set. The glucose content value in the output data set is obtained by manual sampling detection in Step 1. One piece of data is formed every 30 minutes or 60 minutes. The number of its data records is less than the number of data records of the input variables in the input data set. The interpolation method is used to generate the missing glucose content value data to form an output data set corresponding to each data record of the input variable, named the second-stage group 1 to group n input data sets and output data sets, where n is the number of double-enzyme reactions and n≥2. The storage example is shown in Table 1.

[0111] Table 1 Storage example of collected variables

[0112]

[0113] Step 3, data collection of the natural heat dissipation of the reactor and its accessories: For small-scale, pilot-scale and actual production devices, due to the limitations of the device volume and detection conditions, the natural heat dissipation QR of the reactor and its accessories cannot be directly measured, and the approximate value of QR is obtained indirectly by measuring other variables under the approximate operating conditions of the actual reaction device.

[0114] Use the same conditions and methods in Step 1 to collect the data required to calculate the natural heat dissipation of the reactor and its accessories. Remove glucose oxidase and catalase from the raw materials added to the reaction device, and the other raw materials are exactly the same as those in Step 1. All reaction conditions and process parameters are exactly the same as those in Step 1. Because no enzyme is added and the enzyme-catalyzed oxidation reaction does not occur, there is no reaction heat release, while other operating conditions and conditions are the same as the actual reaction in Step 1. In this case, the measured data is closest to the actual value of the natural heat dissipation of the reactor and its accessories. Start the reaction device and the data collection system. After the temperature of the raw material liquid is the same as the room temperature, continue to collect data for 0.5 hours, and then stop the operation of the reaction device and the data collection system to end the data collection.

[0115] The types of data collected in this step are as follows:

[0116] H, collection time TTM, pressure TPR inside the reactor, reaction solution temperature TTRL;

[0117] I, intake air temperature TTGI and exhaust gas temperature TTGO of the air introduced into the reactor, cumulative exhaust gas flow rate TCFG;

[0118] J, cumulative intake flow rate TLW, intake water temperature TWTI, and outlet water temperature TWTO of the circulating water in the reactor jacket.

[0119] The data collected in this step is stored separately. The storage method is similar to that in Step 2. The collected data is sorted and numbered by the collection time as the index and stored. The data of each variable collected at the same collection time occupies one column and is arranged in a row to form a data record, which is named the variable dataset of the first stage of approximate value measurement after storage. The storage example is shown in Table 2.

[0120] Table 2 Storage example of the collected data of the natural heat dissipation of the reactor and its accessories

[0121]

[0122] If there are no changes in the reaction raw materials and reaction conditions during the double-enzyme reaction process and data collection process repeated multiple times in Step 1, the data collection in this step can be carried out once only and does not need to be repeated multiple times.

[0123] Step 4: Merge the data of the optimal time period: According to the determined optimal time period, process the data in the input datasets and output datasets of the 1st to nth groups in the second stage to form a data result set with a longer time period, named the input datasets and output datasets of the 1st to nth groups in the third stage (n is the number of double-enzyme reactions, n ≥ 2); according to the same optimal time period, process the data in the variable dataset of the first stage of approximate value measurement to form a data result set with a longer time period, named the variable dataset of the second stage of approximate value measurement. The selectable optimal time periods are 1 minute, 3 minutes, 5 minutes, 10 minutes, and 30 minutes. For the reaction device in the present invention, the 3-minute time period is the best. The method of merging the 2-second time period data of the variable into the 3-minute time period data is to sum the consecutive 90 2-second time period data of the variable and then divide by 90, and the obtained average value is the result after merging. Each row of data in the processed dataset is sorted and numbered in the order of the original data collection time, with the serial number as the index.

[0124] Step 5: Calculation of data: Calculate the data of some variables in Step 4 to obtain new variables and variable data, and use the serial numbers of each row of data in the input data sets and output data sets of the 1st to nth groups in the third stage as indexes to merge and store them together with the original data in the input data sets and output data sets of the 1st to nth groups in the third stage, forming the input data sets and output data sets of the 1st to nth groups in the fourth stage. n is the number of double enzyme reaction times, and n≥2. The specific calculation method is as follows:

[0125] First, obtain the function formula for calculating the natural heat dissipation QR of the reaction kettle and its accessories. The specific method is as follows:

[0126] Step a: Calculate the variable data in the variable data set of the second stage of approximate value measurement in Step 4 to obtain new variables and variable data, and use the serial numbers of each row of data in the variable data set of the second stage of approximate value measurement as indexes to merge and store them together with the original data in the data set, forming the variable data set of the third stage of approximate value measurement. The calculation method of variables and data is as follows:

[0127] (1) Calculate the heat carried away by the gas TQG in the approximate value measurement stage

[0128] TQG i =(TCFG i -TCFG i-1 )×ρG×CG×(TTGO i -TTGI i )

[0129] In the formula, i is the serial number of the current data record row in the variable data set of the second stage of approximate value measurement, i≥2, and the value of TQG1 is a null value; ρG is the density of air, and the value is ρG = 1.29 kg / m 3 , CG is the specific heat capacity of air, and the value is CG = 1.005 kJ / (kg*K).

[0130] (2) Calculate the heat carried away by the circulating water TQL in the approximate value measurement stage

[0131] TQL i =(TLW i -TLW i-1 )×ρL×CL×(TWTO i -TWTI i )

[0132] In the formula, i is the serial number of the current data record row in the variable data set of the second stage of approximate value measurement, i≥2, and the value of TQL1 is a null value; ρL is the density of water, and the value is ρL = 1000 kg / m 3 , CL is the specific heat capacity of water, and the value is CL = 4.1829 kJ / (kg*K).

[0133] (3) Calculate the natural heat dissipation TQR of the reactor and its accessories during the approximate value measurement stage

[0134] TQR i = mc(TTRL i-1 - TTRL i ) - TQG i - TQL i

[0135] In the formula, i is the serial number of the current data record row in the dataset of the second stage of approximate value measurement, i ≥ 2, and the value of TQR1 is a null value; m is the mass of the reaction liquid, and c is the specific heat capacity of the reaction liquid.

[0136] The calculation results of the above TQG, TQL, and TQR are filled into the corresponding data rows according to the serial number of the current data record row to form a variable dataset for the third stage of approximate value measurement. The storage example is shown in Table 3.

[0137] Table 3 Storage example after calculating TQR

[0138]

[0139] Step b: Use the reaction liquid temperature TTRL in the variable dataset for the third stage of approximate value measurement j as the input variable and TQR j as the output variable for curve fitting. The obtained second-order polynomial function is:

[0140] TQR j = 0.4477 × TTRL j 2 - 28.82 × TTRL j + 465.5(1)

[0141] In the formula, j is the serial number of the current data record row in the variable dataset for the third stage of approximate value measurement, j ≥ 2, (the value of TQR1 is a null value and is not used), and TTRL j is the reaction liquid temperature value of the j-th row in the variable dataset for the third stage of approximate value measurement.

[0142] Convert the function (1) into a function for calculating QR as follows:

[0143] QR x = 0.4477 × TRL x 2 - 28.82 × TRL x + 465.5(2)

[0144] In the formula, x is the serial number of the current data record row in the input dataset and output dataset from the first group to the n-th group in the third stage, x ≥ 1, and TRL xIt is the reaction solution temperature value of the x-th data record row in the dataset.

[0145] Secondly, calculate the G-class variable data. The specific method is as follows:

[0146] (1) Heat carried away by gas QG

[0147] QG x =(CFG x -CFG x-1 )×ρG×CG×(TGO x -TGI x )

[0148] In the formula, x is the serial number of the current data record row in the input dataset and output dataset from the 1st group to the nth group in the third stage, x≥2, the value of QG1 takes the same value as QG2, ρG is the density of air, and the value ρG = 1.29 kg / m 3 , CG is the specific heat capacity of air, and the value CG = 1.005 kJ / (kg*K)

[0149] (2) Heat carried away by circulating water QL

[0150] QL x =(LW x -LW x-1 )×ρL×CL×(WTO x -WTI x )

[0151] In the formula, x is the serial number of the current data record row in the input dataset and output dataset from the 1st group to the nth group in the third stage, x≥2, the value of QL1 takes the same value as QL2, ρL is the density of water, and the value ρL = 1000 kg / m 3 , CL is the specific heat capacity of water, and the value CL = 4.1829 kJ / (kg*K).

[0152] (3) Heat released by reaction QA

[0153] QA x =mc(TRL x -TRL x-1 )+QR x +QG x +QL x

[0154] In the formula, x is the serial number of the current data record row in the input dataset and output dataset from the 1st group to the nth group in the third stage, x≥2, the value of QA1 takes the same value as QA2, m is the mass of the reaction solution, and c is the specific heat capacity of the reaction solution. QR x The value of is based on the reaction solution temperature TRL of the current data record row xThe value is calculated through function (2) in step b, with TRL x The value is the input variable of the function. Substitute it into function (2) to calculate the corresponding QR x value.

[0155] (4) Cumulative heat release of reaction AQA

[0156] AQA x = QA1 + QA2 + QA3 + … + QA x-1 + QA x

[0157] In the formula, x is the serial number of the current data record row in the input data set and output data set from the 1st group to the nth group in the third stage, and x ≥ 1.

[0158] (5) Cumulative heat taken away by gas AQG

[0159] AQG x = QG1 + QG2 + QG3 + … + QG x-1 + QG x

[0160] In the formula, x is the serial number of the current data record row in the input data set and output data set from the 1st group to the nth group in the third stage, and x ≥ 1.

[0161] (6) Cumulative heat taken away by circulating water AQL

[0162] AQL x = QL1 + QL2 + QL3 + … + QL x-1 + QL x

[0163] In the formula, x is the serial number of the current data record row in the input data set and output data set from the 1st group to the nth group in the third stage, and x ≥ 1.

[0164] (7) Cumulative oxygen consumption of reaction AOG

[0165] Since the double - enzyme reaction is an oxygen - consuming reaction, the detected value of the oxygen content OG in the exhaust gas is less than 21%. The oxygen consumption is 21% minus the detected value OG. CFG x is the cumulative flow rate of the exhaust gas at the current data record row, and OCG x is the oxygen consumption of the reaction at the current data record row.

[0166] OCG x =(21 - OG x )×(CFG x - CFG x-1 )

[0167] Wherein, x is the serial number of the current data record row in the input data set and the output data set of the 1st to nth groups in the third stage, x ≥ 2, and the value of OCG1 takes the same value as OCG2.

[0168] AOG x = OCG1 + OCG2 + OCG3 + … + OCG x-1 + OCG x

[0169] Wherein, x is the serial number of the current data record row in the input data set and the output data set of the 1st to nth groups in the third stage, x ≥ 1.

[0170] Step 6, data processing: For the output data sets of the 1st to nth groups in the fourth stage of Step 5, perform normalization processing on them. The obtained input variable data set and output variable data set are named as the input data sets and output data sets of the 1st to nth groups in the fifth stage. n is the number of double-enzyme reaction times, n ≥ 2;

[0171] Step 7, variable screening: Screen each input variable in the input data sets and output data sets of the 1st to nth groups in the fifth stage. The screening method is the Pearson correlation coefficient method or other existing technologies. Select and retain the variable data with a high degree of correlation with the glucose content GC value, and remove the unselected input variable data. Use the selected input variables and data as the input variable data set, and the original glucose content GC value data as the output variable data set to form the input data sets and output data sets of the 1st to nth groups in the sixth stage. n is the number of double-enzyme reaction times, n ≥ 2;

[0172] Step 8, model training: Use the data in the input data sets and output data sets of the 1st to nth groups in the sixth stage as input variables and output variables to train the neural network model to obtain the trained model. The neural network model for training can be a BP, LSTM or other types of models.

[0173] Step 9, prediction of glucose content: In practical applications, use the model trained in Step 8 to predict the glucose content in the reaction solution for a new reaction process. For the reaction process that needs to predict the glucose content in the reaction solution, select the variables for which data needs to be collected according to the types of variables with a high degree of correlation screened in Step 7, collect the real-time data of these variables in the new reaction process, and then detect, collect, store, calculate, merge and process the data of these variables according to Steps 1 to 6, and perform normalization processing on the processed data to form the input variable data in the prediction stage. Input the input variable data in the prediction stage into the model trained in Step 8, calculate to obtain the output value, and the result obtained by performing reverse processing on the output value according to the normalization processing method is the predicted value of the glucose content corresponding to the input variable data in the prediction stage.

[0174] Further, the reaction device is as Figure 1 shown. The reaction device includes a reaction kettle and instrument sensors, specifically including reaction kettle 1; a jacket 2 is arranged outside reaction kettle 1; reaction liquid 3 is contained inside reaction kettle 1; a motor 7 is arranged at the top of reaction kettle 1; a stirring paddle 4 is connected to the motor 7, and the stirring paddle 4 extends into reaction kettle 1; an air outlet pipeline 10 is arranged above one side of reaction kettle 1, and a water inlet pipeline 9 is arranged below the same side of the air outlet pipeline 10; a water outlet pipeline 11 is arranged above the other side of reaction kettle 1, and an air inlet pipeline 5 is arranged below the same side of the water outlet pipeline 11; a feed inlet 8 is arranged above the water outlet pipeline 11; a discharge outlet 6 is arranged below reaction kettle 1.

[0175] Further, an outlet air temperature sensor 12, a gas flowmeter 13, and an oxygen analyzer 14 are arranged on the air outlet pipeline 10.

[0176] Further, an in-kettle temperature sensor 15, a pH meter 16, a dissolved oxygen meter 17, a turbidity meter 18, and a conductivity sensor 19 that extend into the reaction liquid 3 are arranged at the top of reaction kettle 1.

[0177] Further, an in-kettle pressure sensor 20 is separately arranged at the top of reaction kettle 1.

[0178] Further, a heat meter inlet water temperature sensor 21 and a heat meter 22 are arranged on the water inlet pipeline 9.

[0179] Further, a heat meter outlet water temperature sensor 23 is arranged on the water outlet pipeline 11.

[0180] Further, an inlet air temperature sensor 24 is arranged on the air inlet pipeline 5.

[0181] The reference manufacturers and specification models of the instrument sensors are shown in Table 4.

[0182] Table 4 Instrument Sensors for Acquired Variables

[0183]

[0184] As Figure 2 shown, the data acquisition system consists of a PLC, a touch screen, communication lines, and various instrument sensors; sensors with 4 - 20mA signals are connected to the PLC using double-core shielded communication lines, and sensors with Modbus signals are connected to the PLC using RS485 communication lines; the PLC runs acquisition software to read the data monitored by each instrument sensor in real time and displays and stores it through a network cable connection to the touch screen.

[0185] The reference manufacturer and model of the PLC is SIEMENS S7-200 SMART ST20, equipped with the corresponding number of analog input modules EMAE04 and RS485 communication modules SB CM01. The reference manufacturer and model of the touch screen is WEINVIEW MT8012IP.

[0186] Example 1.

[0187] According to Figure 1 、 Figure 2 , prepare the reaction device, various sensors and data acquisition system.

[0188] The computer configuration for data processing and neural network model training is as follows: CPU is AMD R7 4800H, memory is 16GB, and GPU is NVIDIA GEFORCE GTX1650.

[0189] Prepare the raw materials according to the production process of preparing gluconate by the "double enzyme method" as follows: 0.8 kg of glucose (Qiqihar Longjiang Fufeng Biotechnology Co., Ltd.), 20 kg of water, 160 g of zinc oxide (Qingdao Gongshenglian New Material Technology Co., Ltd.), 3 g of glucose oxidase (10000 u / mL, Shandong Longkete Enzyme Preparation Co., Ltd.), 1.6 g of catalase (580000 u / mL, Shandong Longkete Enzyme Preparation Co., Ltd.), appropriate amount of antifoaming agent (Liaoyang Hongwei Chemical Reagent Factory), the specific heat capacity of the reaction solution is taken as 6.8694 kJ / (kg*K), stir with the stirring paddle of the reaction kettle, supply air with an air compressor, and the initial temperature of the reaction solution is 40°C. Since the raw materials in this example are less and the heat released by the reaction is less, circulating water is not used during the reaction process, and the relevant data of the circulating water is not collected. The variable data collected in this example is shown in Table 5.

[0190] Table 5 Variable names collected in Example 1

[0191] Number Variable Name Unit 1 Sampling Time TM 2 Cumulative Exhaust Gas Flow CFG L 3 Inlet Gas Temperature TGI ℃ 4 Exhaust Gas Temperature TGO ℃ 5 Reaction Liquid Temperature TRL ℃ 6 Pressure Inside the Reactor PR Mpa 7 Turbidity Value RT FTU 8 Oxygen Content in Exhaust Gas OG ％ 9 Glucose Content in Reaction Liquid GC ％

[0192] Start the reaction device and data acquisition system according to the content of step 1, collect the actual data of the variables listed in Table 5, continue to collect data for 3 minutes after manually sampling and titrating to detect that the glucose content value in the reaction solution is lower than 0.5, stop the reaction and data acquisition, discharge the reaction device, and clean the reaction device. During the reaction process, the glucose content value in the reaction solution is detected by manual sampling and manual titration methods. The time interval between each sampling and detection is about 1 hour. The detection value is indexed by the sampling time and corresponds to the data collected by the data acquisition system to form the first-stage group 1 input data set and output data set. After storage, the data content is shown in Table 6 (the data volume is huge, only ten are listed here).

[0193] Table 6 Partial collected variable data of Example 1

[0194]

[0195]

[0196] Repeat the above reaction process and data acquisition process to obtain the input data set and output data set of the second group in the first stage.

[0197] Use the interpolation method to generate and fill in the missing data in the GC column of the input data set and output data set of the first group in the first stage, and the input data set and output data set of the second group in the first stage, to form an output data result set corresponding to each item in the INDEX column, that is, the input data set and output data set of the first group in the second stage, and the input data set and output data set of the second group in the second stage. The interpolation method is implemented through method='akima' in the pandas library of python.

[0198] Collect the data used to calculate the natural heat dissipation of the reactor and its accessories according to the content of "Step 3: Data on the Natural Heat Dissipation of the Reactor and Its Accessories". Remove glucose oxidase and catalase from the raw materials added to the reactor, and the other raw materials are the same as those in the previous reaction. Specifically, it is 0.8 kg of glucose, 20 kg of water, 160 g of zinc oxide, and an appropriate amount of antifoaming agent. The stirring conditions of the reactor, the air supply conditions of the air compressor, and the initial temperature of the reaction solution are exactly the same as those in the previous reaction. During the collection process, stop using circulating water and do not collect the relevant data of the circulating water. The collected data is stored and named as the approximate value measurement first-stage variable data set, and the content is shown in Table 7 (the data volume is huge, and only ten items are listed here).

[0199] INDEX TTM TTGI TTGO TTRL TPR TCFG 9050 14:18:18 29.36974 32.74256 33.67845 0.159024 312958 9051 14:18:20 29.3607 32.74256 33.68523 0.1601 312959 9052 14:18:22 29.35166 32.76969 33.68523 0.161023 312961 9053 14:18:24 29.38783 32.72448 33.69202 0.162176 312962 9054 14:18:26 29.36974 32.78777 33.67845 0.163021 312963 9055 14:18:28 29.35166 32.78777 33.67845 0.163943 312965 9056 14:18:30 29.3607 32.77873 33.68523 0.164558 312966 9057 14:18:32 29.36974 32.77873 33.6988 0.165788 312968 9058 14:18:34 29.34262 32.80586 33.68523 0.166403 312969 9059 14:18:36 29.3607 32.78777 33.69202 0.167248 312970

[0200] In Step 1 of this embodiment, the reaction raw materials and reaction conditions in the two double-enzyme reaction processes and data acquisition processes remain unchanged, and the data collection of the natural heat dissipation of the reactor and its accessories is only carried out once.

[0201] Merge the data of the input data set and output data set of the first group and the second group in the second stage according to the content of Step 4 to form a data result set with a longer time period, named the input data set and output data set of the first group and the second group in the third stage. According to the content of Step 4, merge the data in the approximate value measurement first-stage variable data set to form a data result set with a longer time period, named the approximate value measurement second-stage variable data set.

[0202] In the reaction device of this embodiment, 3 minutes is the optimal time period. The original data was in a 2-second time period, which was combined into a 3-minute (180 seconds) time period. The specific method is to sum the data of 90 consecutive 2-second time periods of the variable data and then divide by 90. The obtained average value is the result after combination. Each row of data in the processed dataset is sorted in the order of the original data collection time and numbered, and the data in the collection time column is deleted, with the serial number as the index.

[0203] Calculate the data of the variable dataset in the second stage of approximate value measurement according to the content of step a in the "Data calculation method of the natural heat dissipation QR of the reactor and its accessories" in step 5 to obtain the TQR value, and form the variable dataset in the third stage of approximate value measurement. The content is shown in Table 8 (the data volume is huge, only ten are listed here).

[0204] Table 8 Data after calculating TQR in Embodiment 1

[0205] INDEX TTGI TTGO TTRL TPR TCFG TQG TQR 11 27.48178959 32.64440225 38.57936221 0.160596548 301631.4667 0.812984802 20.94042578 12 27.57030473 32.65022958 38.43460831 0.162113261 301753.1889 0.801646557 20.04039919 13 27.65148547 32.6913223 38.30145884 0.161085039 301874.8778 0.795102601 18.37610762 14 27.73005395 32.86423325 38.16996715 0.161139695 301996.7556 0.811243688 18.12127574 15 27.82218605 32.73412304 38.04533261 0.161951 302118.3444 0.774287857 17.17091967 16 27.90045313 32.68951382 37.92898695 0.160550432 302239.9556 0.755056359 15.99669722 17 27.96525706 32.79410435 37.81158634 0.169592639 302359.7 0.749643219 16.1540045 18 28.03809868 32.73100843 37.71490792 0.186476324 302483.2667 0.751794779 13.16821808 19 28.12902513 32.81017974 37.6172499 0.187660829 302615.4889 0.802441184 13.25861623 20 28.1741367 32.83831168 37.52411309 0.188092102 302747.5667 0.798657106 12.61142543

[0206] According to the content of step b in the "Data calculation method of the natural heat dissipation QR of the reactor and its accessories" in step 5, obtain the function formula for calculating QR.

[0207] Using the reaction liquid temperature TTRL in the variable dataset in the third stage of approximate value measurement j as the input variable and TQR j as the output variable for curve fitting, the obtained second-order polynomial function is:

[0208] TQR j = 0.4477 × TTRL j 2 - 28.82 × TTRL j + 465.5 (1)

[0209] In the formula, j is the serial number of the current data record row in the variable dataset in the third stage of approximate value measurement, j ≥ 2, (the value of TQR1 is a null value and is not used), and TTRL j is the reaction liquid temperature value of the j-th row in the variable dataset in the third stage of the approximate value measurement stage. The curve fitting is implemented through the polyfit method of the numpy library in python.

[0210] Convert the function (1) into a function for calculating QR as follows:

[0211] QR x = 0.4477 × TRL x 2 - 28.82 × TRL x + 465.5 (2)

[0212] Wherein, x is the serial number of the current data record row in the input data set and the output data set from the 1st group to the nth group in the third stage, x ≥ 1, and TRL x is the reaction solution temperature value of the xth data record row in the data set.

[0213] Calculate new variable data such as AQA according to the "Calculation Method of G-Type Variable Data" in Step 5, and store the calculated data in a data table after calculation. The data is shown in Table 9, and the QR column data in each row of the table is obtained by using function (2). Specifically, according to the TRL value of the xth row, the corresponding QR value is calculated by inputting it into function (2) and filled into the QR column of that row to obtain the input data set and output data set of the 1st group in the fourth stage and the input data set and output data set of the 2nd group in the fourth stage (the data volume is huge, and only ten are listed here).

[0214] Table 9 Data after calculating new variables in Example 1

[0215]

[0216] According to the content in "Step 6: Data Processing", perform normalization processing on the input and output variable data of the input data set and output data set of the 1st group in the fourth stage and the input data set and output data set of the 2nd group in the fourth stage, which is completed by the MinMaxScaler method in the sklearn library of python, to obtain the input data set and output data set of the 1st group in the fifth stage and the input data set and output data set of the 2nd group in the fifth stage. As shown in Table 10 (the data volume is huge, and only ten are listed here).

[0217] Table 10 Results after normalization processing of input variables and output variables in Example 1

[0218]

[0219] According to the content in "Step 7: Variable Screening", use the Pearson correlation coefficient method to screen variables for the input data set and output data set of the 1st group in the fifth stage and the input data set and output data set of the 2nd group in the fifth stage. The correlation coefficients of each input variable are as Figure 3 shown.

[0220] According to Figure 3 , finally select a total of 4 variables, namely the cumulative reaction oxygen consumption AOG, turbidity value RT, cumulative reaction heat release AQA, and cumulative gas heat carried away AQG, as the input variables of the model, and use the glucose content GC in the reaction solution as the output variable, and organize the foregoing data accordingly to obtain the input data set and output data set of the 1st group in the sixth stage and the input data set and output data set of the 2nd group in the sixth stage.

[0221] According to the content in "Step 8: Model Training", the BP neural network and the LSTM neural network model are used for training respectively. The mathematical model uses pytorch as the underlying library to build the neural network, and the coding platform is pycharm. In the pytorch environment, the BP and LSTM neural network models are built. The input data set and output data set of the first group in the sixth stage and the input data set and output data set of the second group in the sixth stage are sent into the neural network model for training. The number of training times is 100 times. The optimization algorithm uses the Adma algorithm, and the learning rate is set to 0.01. After completion, the trained neural network model is obtained.

[0222] After the neural network model training is completed, according to the content in "Step 9: Prediction of Glucose Content" of the present invention, a new set of input variable data is collected and processed, and input into the model trained in Step 8 to obtain the predicted value of the glucose content corresponding to the input variable data. Synchronously collect and process the corresponding glucose content data as the measured value for comparison with the prediction result to test the accuracy of the prediction result.

[0223] The comparison of the predicted values and measured values of the BP neural network and the LSTM neural network models is as Figure 4 shown.

[0224] Four indicators, namely Mean Square Error, Root Mean Square Error, Mean Absolute Error, and Mean Absolute Percentage Error, are used to calculate the predicted values and measured values of the trained BP and LSTM models according to the above evaluation indicators. The calculation results are shown in Table 11:

[0225] Table 11 Model Performance Evaluation Indexes of Example 1

[0226] Evaluation Index BP LSTM MSE 0.0119514 0.03132373 RMSE 0.1093225 0.1769851 MAE 0.09647331 0.1530494 MAPE 0.17581609 0.41715953

[0227] In this embodiment, the methods and devices disclosed in the present invention are used to establish two neural network models, BP and LSTM, and the two models are respectively used to predict the glucose content in the production process, and two prediction results are obtained respectively. Both prediction results can meet the actual production requirements. By selecting four indicators, MSE, RMSE, MAE, and MAPE, to compare and analyze the two prediction results, in terms of the accuracy of the prediction results, the BP neural network model is better than the LSTM neural network model.

[0228] Example 2.

[0229] According to Figure 1 、 Figure 2, Prepare the reaction device, various sensors and data acquisition system.

[0230] The computer used for data processing and neural network model training is configured as follows: the CPU is AMD R7 4800H, the memory is 16GB, and the GPU is NVIDIA GEFORCE GTX1650.

[0231] Prepare the raw materials according to the production process of preparing gluconate by the "double enzyme method" as follows: 0.8 kg of glucose (Qiqihar Longjiang Fufeng Biotechnology Co., Ltd.), 20 kg of water, 160 g of zinc oxide (Qingdao Gongshenglian New Material Technology Co., Ltd.), 3 g of glucose oxidase (10000 u / mL, Shandong Longkete Enzyme Preparation Co., Ltd.), 1.6 g of catalase (580000 u / mL, Shandong Longkete Enzyme Preparation Co., Ltd.), an appropriate amount of antifoaming agent (Liaoyang Hongwei Chemical Reagent Factory), the specific heat capacity of the reaction solution is taken as 6.8694 kJ / (kg*K), stir with the stirring paddle of the reaction kettle, supply air with an air compressor, and the initial temperature of the reaction solution is 40°C. Since the raw materials in this example are few and the heat released by the reaction is small, the circulating water is stopped during the reaction process, and the relevant data of the circulating water is not collected. The variable data collected in this example is shown in Table 12.

[0232] Table 12 Collection variable names of Example 2.

[0233] Number Variable Name Unit 1 Sampling Time TM 2 Cumulative Exhaust Gas Flow CFG L 3 Inlet Gas Temperature TGI ℃ 4 Exhaust Gas Temperature TGO ℃ 5 Reaction Liquid Temperature TRL ℃ 6 Pressure Inside the Reactor PR Mpa 7 Conductivity RC S / m 8 Resistivity RER Ω·m 9 Total Dissolved Solids RTDS mg / L 10 Salinity RS ％ 11 Oxygen Content in Exhaust Gas OG ％ 12 Glucose Content in Reaction Liquid GC ％

[0234] Start the reaction device and data acquisition system according to the content of step 1, collect the actual data of the variables listed in Table 13, continue to collect data for 3 minutes after manually sampling and titrating to detect that the glucose content value in the reaction solution is lower than 0.5, stop the reaction and data acquisition, discharge the reaction device, and clean the reaction device. During the reaction process, the glucose content value in the reaction solution is detected by manual sampling and manual titration methods. The time interval between each sampling and detection is about 1 hour. The detection value is indexed by the sampling time and corresponds to the data collected by the data acquisition system to form the first-stage first-group input data set and output data set. After storage, the data content is shown in Table 13 (the data volume is huge, only ten are listed here).

[0235] Table 13 Partial collection variable data of Example 2

[0236]

[0237]

[0238] Repeat the above reaction process and data acquisition process to obtain the first-stage second-group input data set and output data set.

[0239] The interpolation method is used to generate and fill in the missing data in the GC column of the input and output data sets of the first group in the first stage, and the input and output data sets of the second group in the first stage, to form an output data result set corresponding to each row of the INDEX column, that is, the input and output data sets of the first group in the second stage, and the input and output data sets of the second group in the second stage. The interpolation method is implemented through method='akima' in the pandas library in python.

[0240] The reaction raw materials and reaction conditions in this embodiment are exactly the same as those in Embodiment 1. Therefore, directly use the data collected in Embodiment 1 according to "Step 3: Data on the natural heat dissipation of the reaction kettle and its accessories" for calculating the natural heat dissipation of the reaction kettle and its accessories, and the TQR value, approximate value measured in the third-stage variable data set and the function formula for calculating QR obtained according to "Calculation method of the natural heat dissipation QR data of the reaction kettle and its accessories" in Step 5, that is, directly use Formula (2) in Embodiment 1. This embodiment does not perform the content of Step 3.

[0241] Merge the data of the input and output data sets of the first and second groups in the second stage according to the content of Step 4 to form a data result set with a longer time period, named the input and output data sets of the first and second groups in the third stage.

[0242] In the reaction device of this embodiment, 3 minutes is the optimal time period. The original data is in a 2-second time period, which is merged into a 3-minute (180 seconds) time period. The specific method is to sum the continuous 90 2-second time period data of the variable data and then divide by 90. The obtained average value is the merged result. Each row of data in the processed data set is sorted in the order of the original data collection time and numbered, and the data in the collection time column is deleted, with the serial number as the index.

[0243] Calculate new variable data such as AQA according to the "Calculation method of G-class variable data" in Step 5. After calculation, store the data table, as shown in Table 14. Each row of QR column data in the table is obtained by using Function (2). The specific method is to input the TRL value of the x-th row into Function (2) to calculate the corresponding QR value, and fill it into the QR column of that row to obtain the input and output data sets of the first group in the fourth stage, and the input and output data sets of the second group in the fourth stage (the data volume is huge, only ten are listed here).

[0244] Table 14 Data after calculating new variables

[0245]

[0246] According to the content in "Step 6: Data Processing", normalize the input and output variable data of the input and output data sets of the first group in the fourth stage and the input and output data sets of the second group in the fourth stage. This is completed through the MinMaxScaler method in the sklearn library of python, and the input and output data sets of the first group in the fifth stage and the input and output data sets of the second group in the fifth stage are obtained. As shown in Table 15 (the data volume is huge, only ten are listed here).

[0247] Table 15 Results of Normalization Processing of Input and Output Variable Data in Example 2

[0248]

[0249] According to the content in "Step 7: Variable Screening", use the Pearson correlation coefficient method to screen variables for the input and output data sets of the first group in the fifth stage and the input and output data sets of the second group in the fifth stage. The correlation coefficients of each input variable are as Figure 5 shown.

[0250] According to Figure 5 , finally, a total of 4 variables, namely the cumulative reactive oxygen consumption AOG, conductivity RC, cumulative reactive heat release AQA, and cumulative heat carried away by gas AQG, are selected as the input variables of the model, and the glucose content GC in the reaction solution is used as the output variable. Based on this, the above-mentioned data is sorted out to obtain the input and output data sets of the first group in the sixth stage and the input and output data sets of the second group in the sixth stage.

[0251] According to the content in "Step 8: Model Training", use the BP neural network and LSTM neural network models for training respectively. The mathematical model uses pytorch as the underlying library to build the neural network, and the coding platform is pycharm. In the pytorch environment, build the BP and LSTM neural network models, and send the input and output data sets of the first group in the sixth stage and the input and output data sets of the second group in the sixth stage into the neural network model for training. The number of training times is 100 times, and the optimization algorithm uses the Adma algorithm, with the learning rate set to 0.01.

[0252] After the neural network model training is completed, collect and process a set of new input variable data according to the content in "Step 9: Prediction of Glucose Content" of the present invention, and input it into the model trained in Step 8 to obtain the predicted value of the glucose content corresponding to the input variable data. Synchronously collect and process the corresponding glucose content data as the measured value for comparison with the prediction result to test the accuracy of the prediction result.

[0253] The comparison of the predicted values and measured values of the two models, namely the BP neural network and the LSTM neural network, is as Figure 6as shown

[0254] Four indicators, namely Mean Square Error, Root Mean Square Error, Mean Absolute Error, and Mean Absolute Percentage Error, were used to calculate the predicted values and measured values of the trained BP and LSTM models according to the above evaluation indicators. The calculation results are shown in Table 16:

[0255] Table 16 Model Performance Evaluation Indicators of Example 2

[0256] BP LSTM MSE 0.0650907 0.03742334 RMSE 0.2551288 0.1934511 MAE 0.23254013 0.1369849 MAPE 0.39978722 0.18528755

[0257] In this embodiment, two neural network models, BP and LSTM, were established using the methods and devices disclosed in the present invention, and the glucose content during the production process was predicted using the two models respectively, obtaining two prediction results. Both prediction results can meet the actual production requirements. By selecting four indicators, MSE, RMSE, MAE, and MAPE, to compare and analyze the two prediction results, in terms of the accuracy of the prediction results, the LSTM neural network model is superior to the BP neural network model.

Claims

1. A real-time prediction method for the glucose content of a reaction solution in a production process, characterized in that: The specific steps include: Step 1, data collection; gluconate is prepared by the dual enzyme method, the reaction device is started and the data collection system is started at the same time, the detection value of each variable is collected once every 2 seconds, and the detection data of each variable is sorted and stored with the collection time as the index, and the collected multiple variable data form the input data set used by the neural network model; the glucose content value of the reaction solution is measured by titration method by regular manual sampling to form the output data set used by the neural network model; when the measured glucose content value is lower than the given value, the reaction and data collection are terminated; Repeat the above dual enzyme reaction and data collection multiple times to obtain multiple groups of input data sets and output data sets, the data collected by the first dual enzyme reaction forms the first group of input data sets and output data sets of the first stage, the data collected by the nth dual enzyme reaction forms the nth group of input data sets and output data sets of the first stage, n is the number of dual enzyme reactions, n≧2; Step 2, data storage method: the collected data automatically stored in step 1 are sorted and numbered according to the collection time as the index, and each variable data collected at the same collection time occupies one column and is arranged in a row to form a data record, and multiple data records form an input variable data set; the manual sampling titration measurement value of glucose content is stored according to the manual sampling time as the index, and the row where the stored data with the same manual sampling time as the collection time in the input variable data set is selected, and the column where the row is located is filled to form an output data set; the number of data records of glucose content values ​​in the output data set is less than the number of data records of input variables in the input data set, and the interpolation method is used to generate the missing glucose content value data to form an output data set corresponding to the input variable data records one by one, which is named as the second stage of the first to nth group of input data sets and output data sets, where n is the number of double enzyme reactions, and n≧2; Step 3, data collection of natural heat dissipation of the reactor and its accessories: glucose oxidase and catalase are removed from the raw materials, and the other raw materials are exactly the same as those in step 1. All reaction conditions and process parameters are exactly the same as those in step 1. The reaction device and the data acquisition system are started, and data collection continues for 0.5 hours after the raw material liquid temperature is the same as room temperature. The operation of the reaction device and the data acquisition system is stopped, and data collection ends. The storage method of the collected data in this step is similar to that in step 2, and the data is named as the first stage variable data set of approximate value measurement after storage. If the reaction raw materials and reaction conditions in the double enzyme reaction process and the data collection process repeated multiple times in step 1 do not change, the data collection in this step only needs to be performed once; Step 4, optimal time period data merging: according to the determined optimal time period, the data in the first to nth input data sets and output data sets of the second stage are processed to form a data result set of a longer time period, which is named the first to nth input data sets and output data sets of the third stage (n is the number of double enzyme method reactions, n≧2); according to the same optimal time period, the data in the first stage variable data set of the approximate value measurement are processed to form a data result set of a longer time period, which is named the second stage variable data set of the approximate value measurement; Step 5, data calculation: perform data calculation on some variable data in step 4 to obtain new variables and variable data, and use the serial number of each row of data in the first to nth input data set and output data set of the third stage as the index, merge and store the original data in the first to nth input data set and output data set of the third stage together to form the first to nth input data set and output data set of the fourth stage, where n is the number of double enzyme method reactions, and n≧2; Step 6, data processing: The output data sets from the first group to the nth group in the fourth stage of step 5 are normalized, and the obtained input variable data sets and output variable data sets are named as the input data sets from the first group to the nth group and output data sets from the fifth stage, where n is the number of double enzyme method reactions, and n≧2; Step 7, screening of variables: screening the input variables in the first to nth input data sets and output data sets of the fifth stage, selecting the variables and data with high correlation with the glucose content value to form the input variable data set, and the original glucose content value data is the output variable data set, forming the first to nth input data sets and output data sets of the sixth stage, where n is the number of double enzyme method reactions, and n≧2; Step 8, model training: using the data in the first to nth input data sets and output data sets in the sixth stage as input variables and output variables, the neural network model is trained to obtain a trained model; Step 9, prediction of glucose content: Use the model trained in step 8 to predict the glucose content in the reaction liquid for the new reaction process. For the reaction process in which the glucose content in the reaction liquid needs to be predicted, select the variables that need to collect data according to the types of variables with high correlation obtained by screening in step 7, collect the real-time data of these variables in the new reaction process, and then detect, collect, store, calculate, merge and process the data of these variables according to steps 1 to 6, and normalize the processed data to form the input variable data of the prediction stage. Input the input variable data of the prediction stage into the model trained in step 8, and obtain the output value after calculation. The output value is reversely processed according to the normalization processing method to obtain the predicted value of the glucose content corresponding to the input variable data of the prediction stage; The calculation method for the natural heat dissipation QR of the reactor and its accessories in step 5 is: Step a: Calculate the variable data in the variable data set of the second stage of approximate value measurement in step 4 to obtain new variables and variable data, and use the serial number of each row of data in the variable data set of the second stage of approximate value measurement as an index, merge and store them together with the original data in the data set to form the variable data set of the third stage of approximate value measurement. The calculation method of variables and data is as follows: (1) Calculate the approximate value of the heat carried away by the gas during the measurement phase TQG <h2 style=";text-align:left;direction:ltr">TQG<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (TCFG)<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> -TCFG<h2 style=";text-align:left;direction:ltr"> i-1 <h2 style=";text-align:left;direction:ltr"> )×ρG×CG×(TTGO<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> -TTGI<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ) Where i is the current data record row number of the second phase of the approximate measurement data set, i≧2, and the value of TQG1 is null; ρG is the density of air, and the value is ρG=1.29kg / m 3 , CG is the specific heat capacity of air, and its value is CG = 1.005 kJ / (kg*K). (2) Calculate the approximate value of the heat carried away by the circulating water during the measurement phase TQL TQL i =(TLW i -TLW i-1 )×ρL×CL×(TWTO i -TWTI i ) Where i is the current data record row number of the second phase of the approximate measurement data set, i≧2, and the value of TQL1 is null; ρL is the density of water, and the value is ρL=1000kg / m 3 , CL is the specific heat capacity of water, and the value is CL = 4.1829 kJ / (kg*K); (3) Calculate the approximate value of the natural heat dissipation TQR of the reactor and its accessories during the measurement phase TQR i =mc(TTRL i-1 -TTRL i )-TQG i -TQL i Where, i is the current data record row number of the second stage data set of the approximate measurement, i≧2, and the value of TQR1 is null; m is the mass of the reaction liquid, and c is the specific heat capacity of the reaction liquid; Step b: Get the function formula for calculating QR: The temperature of the reaction liquid in the third stage variable data set TTRL is measured with an approximate value j is the input variable, TQR j The second-order polynomial function obtained by curve fitting for the output variable is: TQR j N 0.4477×TTRL j 2 -28.82×TTRL j +465.5(1) Where j is the serial number of the current data record row of the variable data set in the third stage of approximate value measurement, j ≧ 2, (the value of TQR1 is null and not used), TTRL j is the temperature value of the reaction liquid in the jth row of the variable data set of the third stage of the approximate value measurement phase; Function (1) is converted into a function for calculating QR as follows: QR x =0.4477×TRL x 2 -28.82×TRL x +465.5(2) Where x is the sequence number of the current data record row in the first to nth input data sets and output data sets of the third stage, x ≧ 1, TRL x is the temperature value of the reaction liquid in the xth data record row in the data set; The calculation method of the G-type variable data in step 5 is specifically as follows: (1) The gas takes away the heat QG QG x =(CFG x -CFG x-1 )×ρG×CG×(TGO x -TGI x ) Where x is the serial number of the current data record row in the input data set and output data set of the third stage from the first to the nth group, x ≧ 2, the value of QG1 is the same as that of QG2, ρG is the density of air, and the value is ρG = 1.29 kg / m 3 , CG is the specific heat capacity of air, and its value is CG=1.005kJ / (kg*K); (2) Circulating water takes away heat QL QL x =(LW x -LW x-1 )×ρL×CL×(WTO x -WTI x ) Where x is the serial number of the current data record row in the input data set and output data set of the third stage from the first to the nth group, x ≧ 2, the value of QL1 is the same as that of QL2, ρL is the density of water, and the value is ρL = 1000 kg / m 3 , CL is the specific heat capacity of water, and its value is CL = 4.1829 kJ / (kg*K); (3) Reaction heat release QA QA x =mc(TRL x -TRL x-1 )+QR x +QG x +QL x Where x is the serial number of the current data record row in the third stage, the first to the nth input data set and the output data set, x ≧ 2, the value of QA1 is the same as QA2, m is the mass of the reaction liquid, c is the specific heat capacity of the reaction liquid, and the value is c = 6.8694 kJ / (kg*K). QR x The value is based on the reaction liquid temperature TRL of the current data record line. x The value is calculated by function (2) in step b, expressed in TRL x The value is the input variable of the function, substitute it into function (2), and calculate the corresponding QR x value; (4) Cumulative reaction heat release AQA <h2 style=";text-align:left;direction:ltr">AQA<h2 style=";text-align:left;direction:ltr"> x <h2 style=";text-align:left;direction:ltr"> =QA1+QA2+QA3+…+QA<h2 style=";text-align:left;direction:ltr"> x-1 <h2 style=";text-align:left;direction:ltr"> +QA<h2 style=";text-align:left;direction:ltr"> x Where x is the sequence number of the current data record row in the first to the nth input data set and output data set of the third stage, x≧1; (5) Accumulated heat removed by gas AQG AQG x =QG1+QG2+QG3+…+QG x-1 +QG x Where x is the sequence number of the current data record row in the first to the nth input data set and output data set of the third stage, x≧1; (6) Accumulated circulating water heat removal AQL IQ x =QL1+QL2+QL3+…+QL x-1 +QL x Where x is the sequence number of the current data record row in the first to the nth input data set and output data set of the third stage, x ≧ 1; (7) Cumulative reaction oxygen consumption AOG Since the double enzyme reaction is an oxygen-consuming reaction, the oxygen content OG detection value of the exhaust gas is less than 21%. The oxygen consumption is 21% minus the detection value OG. x The accumulated exhaust gas flow rate of the current data record line, OCG x Reaction oxygen consumption for the current data record line. OCG x =(21-AND x )×(CFG x -CFG x-1 ) Where x is the sequence number of the current data record row in the first to nth input data sets and output data sets of the third stage, x ≧ 2, and the value of OCG1 is the same as that of OCG2; AOG x =OCG1+OCG2+OCG3+…+OCG x-1 +OCG x Where x is the sequence number of the current data record row in the first to nth input data sets and output data sets of the third stage, and x≧1.

2. The real-time prediction method according to claim 1, characterized in that: In step 1, manual sampling is performed in the laboratory to detect the glucose content using a chemical titration method and record the results. The manual sampling period is once every 30 minutes or once every 60 minutes.

3. The real-time prediction method according to claim 1, characterized in that: In step 1, the reaction device includes a reactor and an instrument sensor, specifically including a reactor 1; a jacket 2 is provided outside the reactor 1; a reaction liquid 3 is contained inside the reactor 1; a motor 7 is provided on the top of the reactor 1; a stirring paddle 4 is connected to the motor 7, and the stirring paddle 4 extends into the reactor 1; an air outlet pipeline 10 is provided above one side of the reactor 1, and a water inlet pipeline 9 is provided below the same side of the air outlet pipeline 10; a water outlet pipeline 11 is provided above the other side of the reactor 1, and an air inlet pipeline 5 is provided below the same side of the water outlet pipeline 11; a feed inlet 8 is provided above the water outlet pipeline 11; a discharge port 6 is provided below the reactor 1; The outlet pipe 10 is provided with an outlet temperature sensor 12, a gas flow meter 13 and an oxygen analyzer 14; the top of the reactor 1 is provided with an in-reactor temperature sensor 15 extending into the reaction liquid 3, a pH meter 16, a dissolved oxygen meter 17, a turbidity meter 18 and a conductivity sensor 19; a separate in-reactor pressure sensor 20 is provided on the top of the reactor 1; the water inlet pipe 9 is provided with a calorimeter inlet water temperature sensor 21 and a calorimeter 22; the water outlet pipe 11 is provided with a calorimeter outlet water temperature sensor 23; the air inlet pipe 5 is provided with an air inlet temperature sensor 24.

4. The real-time prediction method according to claim 1, characterized in that: In step 1, the variable data is: A, acquisition time TM, pressure in the reactor PR, reaction liquid temperature TRL; B. The air inlet temperature TGI, the exhaust gas temperature TGO, the exhaust gas cumulative flow rate CFG, and the exhaust gas oxygen content OG of the air entering the reactor; C. The cumulative inlet flow rate LW, inlet temperature WTI and outlet temperature WTO of the circulating water in the jacket of the reactor; D. pH value RPH, dissolved oxygen rate RDO, oxygen saturation value ROS, oxygen partial pressure value ROPP, conductivity RC, resistivity RER, total dissolved matter RTDS, salinity RS, turbidity value RT of the reaction liquid in the reactor; E, glucose content in the reaction solution GC; F. Natural heat dissipation QR of the reactor and its accessories; G, heat carried away by gas QG, heat carried away by circulating water QL, heat released by reaction QA, cumulative heat released by reaction AQA, cumulative heat carried away by gas AQG, cumulative heat carried away by circulating water AQL, cumulative oxygen consumption by reaction AOG; Among them, Class A variables are conventional variables, which are fully collected and recorded; Class B and C variables are partially or fully collected and recorded according to the heating and cooling conditions of the reaction and the oxygen consumption; Class D variables are partially or fully collected and recorded according to the reactants and the requirements for monitoring the reaction process; Class E variables are measured by manual titration in the laboratory after manual sampling; Category F variables cannot be directly detected and obtained, and require the special detection method in step 3 to collect and record data and then further calculate to obtain approximate values; Category G variables are further calculated from Category A, B, C, F, and G variables.

5. The real-time prediction method according to claim 1, characterized in that: In step 3, the variables collected for the natural heat dissipation of the reactor and its accessories are: H, acquisition time TTM, pressure in the reactor TPR, reaction liquid temperature TTRL; I. The air inlet temperature TTGI, the exhaust gas temperature TTGO, and the exhaust gas cumulative flow rate TCFG of the air entering the reactor; J. The cumulative inlet flow rate TLW, inlet temperature TWTI and outlet temperature TWTO of the circulating water in the jacket of the reactor.

6. The real-time prediction method according to claim 1, characterized in that: In step 4, the optimal time period is any one of 1 minute, 3 minutes, 5 minutes, 10 minutes or 30 minutes.

7. The real-time prediction method according to claim 1, characterized in that: In step 7, the screening method is grey correlation analysis.

8. The real-time prediction method according to claim 1, characterized in that: In step 8, the neural network model is BP or LSTM.

Images