Online isocyanate content monitoring method and system based on machine learning

Abstract

The invention discloses an online isocyanate content monitoring method and system based on machine learning, and relates to the field of chemical reaction process analysis technology and machine learning cross technology. Comprising the following steps: S1, constructing a data set; s2, quantitative model design; s3, model training and evaluation; and S4, applying the model. The system comprises a reactor, an online viscometer, a temperature sensor and a computer. According to the method, the end point control repeatability in the polyurethane production process can be improved, the production batch stability and the product percent of pass are effectively improved, and the labor cost is reduced.

Description

Technical Field

[0001] The present invention relates to the cross-technical field of chemical reaction process analysis technology and machine learning, and in particular to an isocyanate content online monitoring method and system based on machine learning. Background Art

[0002] Polyurethane is a synthetic resin with excellent physical and chemical properties, which is widely used in many industries such as chemical industry, automobile, and construction. However, in the current polyurethane production process, there is a problem of insufficient process control precision, which leads to large fluctuations in batch quality, reduces the yield rate, and affects the market promotion of polyurethane series products. The main reason for this problem is the insufficient control precision of the polymerization reaction endpoint. The isocyanate content of the reaction material is a key parameter for the control of the reaction endpoint. At present, after manual sampling in production, the di-n-butylamine titration method is used to detect the remaining isocyanate content in the reaction material, and the reaction endpoint is judged based on this result. However, this method has the following disadvantages: first, the titration detection operation takes a certain amount of time, which causes the test results to only reflect the reaction situation at the time of sampling, and cannot reflect the reaction situation at the time of completion of the detection. The timeliness of the method is poor, which is easy to cause overreaction and product quality fluctuations. Second, performing multiple detection operations requires a lot of labor costs.

[0003] With the development of digital production process, process analysis technology (PAT) has received more and more attention. Process analysis technology monitors the progress of chemical reactions by real-time monitoring and comprehensive analysis of multiple parameters in the chemical reaction process. Process analysis technology has positive significance for improving product quality, improving the level of production digitalization, and reducing labor costs. Currently, the methods that can be applied to polyurethane reaction process monitoring include infrared spectroscopy, viscosity method, light scattering method, dielectric method, Raman spectroscopy, etc. Considering factors such as equipment investment, maintenance difficulty, and operator skill requirements, the viscosity method is currently the most valuable method for promotion and application.

[0004] In recent years, artificial intelligence technology has developed rapidly, and AI for Science has become a research hotspot. Artificial intelligence technology provides new solutions to scientific and engineering problems that have plagued the chemical industry for many years.

[0005] Therefore, those skilled in the art urgently need to propose an online monitoring method and system for isocyanate content based on machine learning to solve the difficulties existing in the prior art. Summary of the invention

[0006] In view of this, the present invention provides an isocyanate content online monitoring method and system based on machine learning. The isocyanate content quantitative model is used to calculate the isocyanate content of the reaction materials in real time and display it on the computer screen. The method can effectively improve the batch stability and product qualification rate of production and reduce labor costs by improving the repeatability of endpoint control in the polyurethane production process.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: an online monitoring method of isocyanate content based on machine learning, comprising:

[0008] S1. Constructing a data set: conducting experiments at different temperatures, collecting experimental data of reaction samples at different time points during the experiment, and obtaining a summary of multiple sets of experimental data, including: temperature, viscosity, and isocyanate titration content, to form a data set;

[0009] S2, quantitative model design: construct an isocyanate quantitative model, using the temperature and viscosity data obtained in S1 as inputs to the isocyanate quantitative model, and the model-predicted value of the isocyanate content as output;

[0010] S3, model training and evaluation: based on the data set obtained in S1, the isocyanate quantitative model in S2 is optimized to obtain an optimized isocyanate quantitative model;

[0011] S4. Model application: The temperature and viscosity collected in real time from the reactor are input into the isocyanate content quantitative model to obtain the real-time isocyanate content in the reaction material and display it on the computer screen.

[0012] In the above method, optionally, the isocyanate quantitative model in S2 includes:

[0013] Data preprocessing module: including outlier filtering and filtering process;

[0014] Normalization module: Map the data to the [min, max] interval, using the maximum-minimum normalization method to obtain normalized temperature data and normalized viscosity data;

[0015] Multilayer Perceptron: inputs normalized temperature data and normalized viscosity data, and outputs normalized isocyanate content prediction data;

[0016] Denormalization module: converts the normalized isocyanate content prediction data output by the multi-layer perceptron into isocyanate content prediction data.

[0017] The above method optionally includes a normalization module: mapping the temperature, viscosity or isocyanate content data to the [min, max] interval, using the maximum-minimum normalization method, and calculating using the following formula:

[0018]

[0019] Where: X i is the i-th data to be normalized, X min is the minimum value of the data to be normalized, X max is the maximum value of the data to be normalized, min is the lower limit of the data to be normalized mapped to the target interval, max is the upper limit of the data to be normalized mapped to the target interval, N xi is the normalized value of the i-th data to be normalized.

[0020] In the above method, optionally, a multilayer perceptron is the main body of the quantitative model, which is composed of an input layer, a hidden layer, and an output layer;

[0021] Among them, the input layer is the normalized temperature and viscosity;

[0022] There are k hidden layers in total, k ranges from 1 to 50;

[0023] Each hidden layer contains 2 to 100 neurons;

[0024] The relationship between neurons in the hidden layers is a fully connected relationship;

[0025] The activation function of each hidden layer is GeLU;

[0026] The output layer is the normalized isocyanate content.

[0027] The above method, optionally, performs inverse normalization: inverse normalization of the data from the multilayer perceptron is performed using the inverse operation of maximum-minimum normalization, calculated using the following formula:

[0028]

[0029] Where: X nco,min is the minimum value of the isocyanate titration content data in the data set, X nco,max is the maximum value of the isocyanate titration content data in the data set, N nco,i is the normalized isocyanate content data, min is the lower limit of the normalized mapping to the target interval, and max is the upper limit of the normalized mapping to the target interval. is the inverse normalized value, that is, the model-predicted value of isocyanate content.

[0030] In the above method, optionally, constructing a data set in S1 includes:

[0031] For a process with a specified temperature of T d ℃ reaction process, respectively, T d±δ℃ is the process control temperature for the experiment, where δ is any value between 0 and 50; no less than 3 groups of experiments are carried out at different temperatures;

[0032] During the experiment, reaction material samples were collected at different reaction time points, and the isocyanate content was determined by di-n-butylamine titration. At the same time, the temperature and viscosity data that had been filtered and filtered for abnormal values ​​at the sampling time were recorded. 2 to 50 samples were taken in each experimental process, and multiple groups of experimental data were summarized to obtain several data combinations, including temperature, viscosity and isocyanate titration content, to form a data set.

[0033] In the above method, optional optimization contents in S3 include:

[0034] The temperature, viscosity, and isocyanate titration content in the data set S1 are normalized to the maximum and minimum values, and the normalized results are used as the training set for optimizing the hidden layer parameters of the multi-layer perceptron: the normalized temperature and viscosity are used as the input data features of the multi-layer perceptron; the normalized isocyanate titration content is used as the output data label of the multi-layer perceptron;

[0035] The mean square error is used as the loss function, and the hidden layer parameters are optimized using an optimization algorithm; the optimization algorithm is the stochastic gradient descent algorithm; the mean absolute error indicator is used to evaluate the multilayer perceptron.

[0036] An online monitoring system for isocyanate content based on machine learning, which performs any of the above-mentioned online monitoring methods for isocyanate content based on machine learning, comprising: a reactor, an online viscometer, a temperature sensor, and a computer;

[0037] The computer is connected to an online viscometer and a temperature sensor through a communication line. The temperature sensor and the online viscometer are arranged inside the reactor. The reactor is a container for the polymerization reaction of polyurethane, which is a kettle reactor.

[0038] Online viscometers are used to measure the real-time viscosity of materials and are either vibrating or rotating.

[0039] The temperature sensor is used to measure the material temperature and should be located near the online viscometer or integrated inside the online viscometer to minimize the impact of material temperature non-uniformity on the viscosity measurement value;

[0040] Computers are used for data acquisition and data analysis.

[0041] It can be seen from the above technical solution that, compared with the prior art, the online monitoring method and system of isocyanate content based on machine learning of the present invention has the following beneficial effects:

[0042] During the polyurethane polymerization process, the temperature and viscosity of the reaction materials are collected in real time and transmitted to the computer. The isocyanate content of the reaction materials is calculated in real time using the isocyanate content quantitative model and displayed on the computer screen; the present invention can effectively improve the batch stability and product qualification rate of production and reduce labor costs by improving the repeatability of endpoint control in the polyurethane production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0044] Figure 1 A flow chart of an online monitoring method for isocyanate content based on machine learning provided by the present invention;

[0045] Figure 2 A schematic diagram of a quantitative model for isocyanate content provided by the present invention;

[0046] Figure 3 A visualization diagram of the isocyanate content quantitative model provided by the present invention;

[0047] Figure 4 A real-time isocyanate model predicted content-reaction time curve provided by the present invention;

[0048] Figure 5 This is a structural diagram of an isocyanate content online monitoring system based on machine learning provided by the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, the elements defined by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0051] The present invention can be used in many general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, distributed computing environments including any of the above devices or devices, etc.

[0052] Reference Figure 1 As shown, the present invention discloses an online monitoring method for isocyanate content based on machine learning, comprising:

[0053] S1. Constructing a data set: conducting experiments at different temperatures, collecting experimental data of reaction samples at different time points during the experiment, and obtaining a summary of multiple sets of experimental data, including: temperature, viscosity, and isocyanate titration content, to form a data set;

[0054] S2, quantitative model design: construct an isocyanate quantitative model, using the temperature and viscosity data obtained in S1 as inputs to the isocyanate quantitative model, and the model-predicted value of the isocyanate content as output;

[0055] S3, model training and evaluation: based on the data set obtained in S1, the isocyanate quantitative model in S2 is optimized to obtain an optimized isocyanate quantitative model;

[0056] S4. Model application: The temperature and viscosity collected in real time from the reactor are input into the isocyanate content quantitative model to obtain the real-time isocyanate content in the reaction material and display it on the computer screen.

[0057] Further, such as Figure 2 As shown, the isocyanate quantitative model in S2 includes:

[0058] Data preprocessing module: including outlier filtering and filtering process;

[0059] Normalization module: Map the data to the [min, max] interval, using the maximum-minimum normalization method to obtain normalized temperature data and normalized viscosity data;

[0060] Multilayer Perceptron: inputs normalized temperature data and normalized viscosity data, and outputs normalized isocyanate content prediction data;

[0061] Denormalization module: converts the normalized isocyanate content prediction data output by the multi-layer perceptron into isocyanate content prediction data.

[0062] Specifically, outlier filtering: two thresholds HH and LL are pre-configured for the temperature and viscosity inputs respectively. If the model input value is higher than HH or lower than LL, the data cannot enter the subsequent processing.

[0063] Filtering: Use a filtering algorithm to process the signal to reduce the noise of the input signal. The filtering algorithm is one of the following: limiting filtering method, median filtering method, arithmetic mean filtering method, recursive mean filtering method, median mean filtering method, limiting mean filtering method, and weighted recursive mean filtering method.

[0064] In a specific embodiment, the quantitative model design

[0065] Abnormal value filtering: The temperature input threshold HH is 90°C, and the threshold LL is 50°C. The viscosity input threshold HH is 5000 centipoise, and the threshold LL is 0 centipoise.

[0066] Filtering: The input temperature and viscosity signals are filtered using the arithmetic mean filtering method, and the average value of the measured values ​​within the last 10 seconds is calculated as the filter output.

[0067] Normalization: Map the temperature and viscosity inputs to the [-1,1] interval using the maximum-minimum normalization method.

[0068] The input layer is preprocessed data, which is normalized temperature and viscosity. There are 4 hidden layers, each containing 10, 20, 30, and 20 neurons respectively. The relationship between neurons in the hidden layers is fully connected. The activation function of each hidden layer is GeLU.

[0069] Furthermore, the normalization module maps the temperature, viscosity or isocyanate content data to the [min, max] interval, using the maximum-minimum normalization method, calculated using the following formula:

[0070]

[0071] Where: X i is the i-th data to be normalized, X minis the minimum value of the data to be normalized, X max is the maximum value of the data to be normalized, min is the lower limit of the data to be normalized mapped to the target interval, max is the upper limit of the data to be normalized mapped to the target interval, N xi is the normalized value of the i-th data to be normalized.

[0072] Furthermore, the multilayer perceptron is the main body of the quantitative model, which is composed of an input layer, a hidden layer, and an output layer; the input layer is normalized temperature and viscosity. There are k hidden layers in total, k is 1 to 50, preferably 2 to 10. Each hidden layer contains 2 to 100 neurons, preferably 20 to 50. The relationship between neurons in the hidden layers is a fully connected relationship. The activation function of each hidden layer can be selected from Sigmoid, ReLU, GeLU, Tanh, preferably GeLU. The output layer is the normalized isocyanate content.

[0073] Furthermore, inverse normalization is performed on the data from the multilayer perceptron, using the inverse operation of maximum-minimum normalization, calculated using the following formula:

[0074]

[0075] Where: X nco,min is the minimum value of the isocyanate titration content data in the data set, X nco,max is the maximum value of the isocyanate titration content data in the data set, N nco,i is the normalized isocyanate content data, min is the lower limit of the normalized mapping to the target interval, and max is the upper limit of the normalized mapping to the target interval. is the inverse normalized value, that is, the model-predicted value of isocyanate content.

[0076] Furthermore, the dataset constructed in S1 includes:

[0077] For a process with a specified temperature of T d ℃ reaction process, respectively, T d ±δ℃ is the process control temperature for the experiment, where δ is any value between 0 and 50; no less than 3 groups of experiments are carried out at different temperatures;

[0078] During the experiment, reaction material samples were collected at different reaction time points, and the isocyanate content was determined by di-n-butylamine titration. At the same time, the temperature and viscosity data that had been filtered and filtered for abnormal values ​​at the sampling time were recorded. 2 to 50 samples were taken in each experimental process, and multiple groups of experimental data were summarized to obtain several data combinations, including temperature, viscosity and isocyanate titration content, to form a data set.

[0079] In a specific embodiment, the process of this embodiment is: 1541.7g of isophorone diisocyanate and 3472.3g of poly(1,4-butylene adipate) (molecular weight 1000) are mixed and reacted at 65°C, 68°C, 74°C, and 77°C, respectively. During the experiment, samples were taken at different reaction time points, and the isocyanate content was determined by di-n-butylamine titration. The temperature and viscosity of the reaction at the time of sampling were recorded at the same time. The experimental data are summarized in Tables 1 to 4 to form a data set.

[0080] Table 1. Experimental data at 65°C

[0081]

[0082]

[0083] Table 2. Experimental data at 68°C

[0084]

[0085]

[0086] Table 3. Experimental data at 74°C

[0087]

[0088]

[0089] Table 4. Experimental data at 77°C

[0090]

[0091]

[0092] Furthermore, the optimization content in S3 includes: normalizing the temperature, viscosity, and isocyanate titration content in the data set of step S1 to the maximum-minimum value respectively, and using the normalized results as the training set for optimizing the hidden layer parameters of the multi-layer perceptron: using the normalized temperature and viscosity as the input data features of the multi-layer perceptron; and using the normalized isocyanate titration content as the output data label of the multi-layer perceptron;

[0093] The mean square error is used as the loss function, and the hidden layer parameters are optimized using an optimization algorithm; the optimization algorithm is the stochastic gradient descent algorithm; the mean absolute error indicator is used to evaluate the multilayer perceptron.

[0094] Specifically, L1 norm or mean square error is used as the loss function, and an optimization algorithm is used to optimize the hidden layer parameters. The optimization algorithm is a variety of gradient descent algorithms, Adam algorithm and other commonly used algorithms for neural network training in the industry, preferably a stochastic gradient descent algorithm.

[0095] The evaluation of the multilayer perceptron uses one or more of the following indicators: mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and coefficient of determination (R2).

[0096] In a specific embodiment, the temperature, viscosity, and isocyanate titration content in the data set of step S3 are respectively normalized to the maximum-minimum value. The normalized temperature and viscosity are used as the input data features of the multi-layer perceptron, and the normalized isocyanate titration content is used as the output data label of the multi-layer perceptron. The mean square error is used as the loss function, and the hidden layer parameters are optimized using the stochastic gradient descent algorithm to complete the model training.

[0097] Calculate the evaluation indicators of the multilayer perceptron: MAE: 0.0138424, MSE: 0.000514020, RMSE: 0.022672, R2: 0.998522.

[0098] The isocyanate content quantitative model can be visualized by using the isocyanate content output by the model as the Z coordinate and the temperature and viscosity input by the model as the X and Y coordinates, respectively. Figure 3 .

[0099] In a specific embodiment, the optimized model is applied, and the collected temperature and viscosity data are input into the trained model to solve the isocyanate content of the current material in real time. The change of the isocyanate model predicted by the real-time solution with the reaction time is shown in Figure 4 .

[0100] Table 5 lists the isocyanate content measured by titration when sampling at different reaction times. The difference between the model predicted content and the titrated content is the absolute error of the model prediction, and the result is better than ±0.1%, indicating that the isocyanate content online monitoring technology of the present invention can measure the isocyanate content of the reaction material in real time with high accuracy.

[0101] Table 5. Actual production data of polyurethane

[0102]

[0103]

[0104] and Figure 1 Correspondingly, this embodiment also discloses an isocyanate content online monitoring system based on machine learning, such as Figure 5 As shown, by applying any of the above-mentioned online monitoring methods for isocyanate content based on machine learning,

[0105] Includes: reactor, online viscometer, temperature sensor, computer;

[0106] The computer is connected to an online viscometer and a temperature sensor through a communication line. The temperature sensor and the online viscometer are arranged inside the reactor. The reactor is a container for the polymerization reaction of polyurethane, which is a kettle reactor.

[0107] Online viscometers are used to measure the real-time viscosity of materials and are either vibrating or rotating.

[0108] The temperature sensor is used to measure the material temperature and should be located near the online viscometer or integrated inside the online viscometer to minimize the impact of material temperature non-uniformity on the viscosity measurement value;

[0109] Computers are used for data acquisition and data analysis.

[0110] Specifically, the equipment involved in the system of the present invention includes: a reactor, an online viscometer, a temperature sensor, and a computer.

[0111] An online viscometer is installed on the kettle reactor. The online viscometer adopts the vibration principle, and the temperature sensor is integrated inside the online viscometer. The online viscometer and the temperature sensor are connected to the desktop computer through a communication line.

[0112] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The system embodiment described above is only schematic, wherein the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0113] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for online monitoring of isocyanate content based on machine learning, characterized in that: include: S1. Constructing a data set: conducting experiments at different temperatures, collecting experimental data of reaction samples at different time points during the experiment, and obtaining a summary of multiple sets of experimental data, including: temperature, viscosity, and isocyanate titration content, to form a data set; S2, quantitative model design: construct an isocyanate quantitative model, using the temperature and viscosity data obtained in S1 as inputs to the isocyanate quantitative model, and the model-predicted value of the isocyanate content as output; S3, model training and evaluation: based on the data set obtained in S1, the isocyanate quantitative model in S2 is optimized to obtain an optimized isocyanate quantitative model; S4. Model application: The temperature and viscosity collected in real time from the reactor are input into the isocyanate content quantitative model to obtain the real-time isocyanate content in the reaction material and display it on the computer screen.

2. The method for online monitoring of isocyanate content based on machine learning according to claim 1, characterized in that: The isocyanate quantitative model in S2 includes: Data preprocessing module: including outlier filtering and filtering process; Normalization module: Map the data to the [min, max] interval, using the maximum-minimum normalization method to obtain normalized temperature data and normalized viscosity data; Multilayer Perceptron: inputs normalized temperature data and normalized viscosity data, and outputs normalized isocyanate content prediction data; Denormalization module: converts the normalized isocyanate content prediction data output by the multi-layer perceptron into isocyanate content prediction data.

3. The method for online monitoring of isocyanate content based on machine learning according to claim 2, characterized in that: Normalization module: Map the temperature, viscosity or isocyanate content data to the [min, max] interval, using the maximum-minimum normalization method, and using the following formula: Where: X i is the i-th data to be normalized, X min is the minimum value of the data to be normalized, X max is the maximum value of the data to be normalized, min is the lower limit of the data to be normalized mapped to the target interval, max is the upper limit of the data to be normalized mapped to the target interval, N xi is the normalized value of the i-th data to be normalized.

4. The method for online monitoring of isocyanate content based on machine learning according to claim 2, characterized in that: The multi-layer perceptron is the main body of the quantitative model, which consists of an input layer, a hidden layer, and an output layer; Among them, the input layer is the normalized temperature and viscosity; There are k hidden layers in total, k ranges from 1 to 50; Each hidden layer contains 2 to 100 neurons; The relationship between neurons in the hidden layers is a fully connected relationship; The activation function of each hidden layer is GeLU; The output layer is the normalized isocyanate content.

5. The method for online monitoring of isocyanate content based on machine learning according to claim 2, characterized in that: Inverse Normalization: Inverse normalization is performed on the data from the multilayer perceptron using the inverse operation of the maximum-minimum normalization, calculated using the following formula: Where: X nco,min is the minimum value of the isocyanate titration content data in the data set, X nco,max is the maximum value of the isocyanate titration content data in the data set, N nco,i is the normalized isocyanate content data, min is the lower limit of the normalized mapping to the target interval, and max is the upper limit of the normalized mapping to the target interval. is the inverse normalized value, that is, the model-predicted value of isocyanate content.

6. The method for online monitoring of isocyanate content based on machine learning according to claim 1, characterized in that: The dataset constructed in S1 includes: For a process with a specified temperature of T d ℃ reaction process, respectively, T d ±δ℃ is the process control temperature for the experiment, where δ is any value between 0 and 50; no less than 3 groups of experiments are carried out at different temperatures; During the experiment, reaction material samples were collected at different reaction time points, and the isocyanate content was determined by di-n-butylamine titration. At the same time, the temperature and viscosity data that had been filtered and filtered for abnormal values ​​at the sampling time were recorded. 2 to 50 samples were taken in each experimental process, and multiple groups of experimental data were summarized to obtain several data combinations, including temperature, viscosity and isocyanate titration content, to form a data set.

7. The method for online monitoring of isocyanate content based on machine learning according to claim 1, characterized in that: The optimizations in S3 include: The temperature, viscosity, and isocyanate titration content in the data set S1 are normalized to the maximum and minimum values, and the normalized results are used as the training set for optimizing the hidden layer parameters of the multi-layer perceptron: the normalized temperature and viscosity are used as the input data features of the multi-layer perceptron; the normalized isocyanate titration content is used as the output data label of the multi-layer perceptron; The mean square error is used as the loss function, and the hidden layer parameters are optimized using an optimization algorithm; the optimization algorithm is the stochastic gradient descent algorithm; the mean absolute error indicator is used to evaluate the multilayer perceptron.

8. An online monitoring system for isocyanate content based on machine learning, characterized in that: An online monitoring method for isocyanate content based on machine learning according to any one of claims 1 to 7 is implemented, comprising: a reactor, an online viscometer, a temperature sensor, and a computer; The computer is connected to an online viscometer and a temperature sensor through a communication line. The temperature sensor and the online viscometer are arranged inside the reactor. The reactor is a container for the polymerization reaction of polyurethane, which is a kettle reactor. Online viscometers are used to measure the real-time viscosity of materials and are either vibrating or rotating. The temperature sensor is used to measure the material temperature and should be located near the online viscometer or integrated inside the online viscometer to minimize the impact of material temperature non-uniformity on the viscosity measurement value; Computers are used for data acquisition and data analysis.

Images