Isoprene rubber polymerization reaction control method and system

Abstract

The invention relates to the technical field of data processing, and discloses an isoprene rubber polymerization reaction control method and system, and the method comprises the steps: collecting real-time reaction data and real-time image data of an isoprene rubber polymerization reaction; performing feature extraction on the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain time sequence features and image features; based on a cross-modal attention mechanism, fusing the time sequence features and the image features to obtain multi-modal joint representation; processing the multi-modal joint representation through a lightweight pre-training model embedded with Arrhenius equation constraint, and outputting a reaction prediction result; according to the reaction prediction result, generating a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm, and executing the reaction control strategy so as to control the polymerization reaction of the isoprene rubber, through deep fusion of multi-source data and a plurality of advanced algorithms, effective control of the isoprene rubber polymerization reaction is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular, to a method and system for controlling isoprene rubber polymerization reaction. Background Art

[0002] Traditional industrial production of isoprene rubber mainly relies on solution polymerization. However, during the polymerization reaction, as the viscosity of the reaction materials gradually increases, the aggregation of high-viscosity intermediate products or final products will affect the utilization rate of raw materials, resulting in a decrease in reaction efficiency. In addition, high-viscosity intermediate products or final products are likely to adhere to the inner wall of the reaction vessel. Coupled with the fact that the polymerization reaction is an exothermic reaction, it will cause the temperature in the reaction vessel to be too high, affecting the quality of the final product and the utilization rate of raw materials. Seriously, it will cause the temperature of the reaction vessel to be too high, and then dangerous accidents may occur.

[0003] In response to this phenomenon, currently, an industrial camera is usually used to collect real-time reaction images of isoprene rubber polymerization reaction. After simple processing, the images are input into a machine learning model for recognition processing to judge the reaction degree. Then, combined with the experience of professionals, a control strategy is generated, and the reaction process is adjusted in real time. However, simple image processing easily leads to low accuracy in subsequent model recognition, resulting in the generated control strategy being unable to improve the reaction efficiency of isoprene rubber polymerization reaction.

[0004] Therefore, how to effectively control the isoprene rubber polymerization reaction and then improve its reaction efficiency and safety has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0005] The present invention provides a method and system for controlling isoprene rubber polymerization reaction, which solves the problem of how to effectively control the isoprene rubber polymerization reaction and then improve its reaction efficiency and safety.

[0006] To solve the above technical problems, a first aspect of the present invention provides a method for controlling isoprene rubber polymerization reaction, including: Collecting real-time reaction data and real-time image data of isoprene rubber polymerization reaction; Based on a pre-trained deep learning model, respectively extracting features from the real-time reaction data and the real-time image data to obtain time series features and image features; Based on a cross-modal attention mechanism, fusing the time series features and the image features to obtain a multi-modal joint representation; Processing the multi-modal joint representation through a lightweight pre-trained model embedded with the Arrhenius equation constraint, and outputting a reaction prediction result; According to the reaction prediction result, generating a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm and executing it to achieve the control of the isoprene rubber polymerization reaction.

[0007] As one of the preferred solutions, the pre-trained deep learning model extracts features from the real-time reaction data and the real-time image data respectively to obtain temporal features and image features, including: Using a pre-trained long short-term memory network to extract the temperature change rate and pressure fluctuation amplitude in the real-time reaction data to generate temporal features; Processing the real-time image data through a pre-trained convolutional neural network and image segmentation technology to obtain image features.

[0008] As one of the preferred solutions, the processing of the real-time image data through a pre-trained convolutional neural network and image segmentation technology to obtain image features includes: Performing grayscale and binarization processing on the real-time image data to obtain a real-time processed image; Using the shallow convolutional kernel of the convolutional neural network to extract the color distribution features of the real-time processed image to obtain a color histogram; Extracting the texture roughness of the real-time processed image through the middle convolutional layer of the convolutional neural network and processing the texture roughness based on the deep convolutional layer of the convolutional neural network to obtain texture features; Calculating the colloidal displacement field between adjacent frames in the real-time processed image using the optical flow method to generate a dynamic viscosity change feature vector; Detecting bubbles in the real-time processed image using a pre-trained segmentation model to obtain bubble features; Integrating the color histogram, the texture features, the dynamic viscosity change feature vector and the bubble features to obtain image features.

[0009] As one of the preferred solutions, the detecting bubbles in the real-time processed image using a pre-trained segmentation model to obtain bubble features includes: Using a pre-trained MobileNetV3 model to process the real-time processed image to obtain a bubble segmentation mask; Performing hole filling, noise filtering and edge thinning processing on the bubble segmentation mask to obtain bubble connected regions; Calculating the area, circularity and centroid coordinates of each bubble connected region to construct bubble static features; Based on the centroid coordinates, using the SORT algorithm to associate bubbles in adjacent frames to obtain the continuous coordinates of each bubble to quantify the kinematic parameters of each bubble; Using Stokes' law to verify each bubble to mark abnormal bubbles; Integrate the area, the circularity, the kinematic parameters, and the abnormal bubbles to obtain bubble features.

[0010] As one of the preferred solutions, based on the cross-modal attention mechanism, fuse the temporal features and the image features to obtain a multi-modal joint representation, including: Encode the temporal features using the temporal self-attention layer of a pre-trained Transformer to obtain a temporal feature vector; Process the image features using the local blocks of a pre-trained Vision Transformer to generate an image feature vector; Based on the cross-modal attention mechanism, perform bidirectional interaction between the temporal feature vector and the image feature vector, and introduce a learning parameter matrix during the interaction to adjust the interaction intensity, generating a first enhanced representation and a second enhanced representation; Aggregate the first enhanced representation and the second enhanced representation through a gating mechanism, and introduce a conditional routing mechanism during the aggregation to dynamically select a fusion path based on the modal correlation between the first enhanced representation and the second enhanced representation, obtaining a multi-modal joint representation.

[0011] As one of the preferred solutions, process the multi-modal joint representation through a lightweight pre-trained model constrained by the Arrhenius equation, and output reaction prediction results, including: Construct an Arrhenius equation based on the isoprene polymerization reaction, and input the temperature data in the real-time reaction data as an explicit feature into the Arrhenius equation to obtain Arrhenius auxiliary features; Concatenate the multi-modal joint representation, the temperature data, and the Arrhenius auxiliary features into an input tensor, and input the input tensor into the lightweight pre-trained model for processing, outputting the predicted reaction conversion rate and the predicted viscosity change rate for a future preset time period as the reaction prediction results.

[0012] As one of the preferred solutions, generate and execute a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm according to the reaction prediction results to achieve the control of the isoprene polymerization reaction, including: Obtain the reaction error corresponding to the reaction prediction results according to the transfer learning method, and decompose the reaction error into a short-term error and a long-term error using the moving window average method; Construct a dynamic weight function through the long-term error and the short-term error, and use the dynamic weight function to quantify the first contribution ratio and the second contribution ratio corresponding to the fuzzy control method and the anti-saturation integral PID control algorithm respectively; Based on the reaction error, a first control signal and a second control signal are generated by a fuzzy control method and an anti-saturation integral PID control algorithm, and the total control signal is obtained by weighting the first contribution ratio and the second contribution ratio; According to the total control signal and the reaction error, the pulse width, pulse interval and pulse amplitude are adjusted by a gradient optimization algorithm to generate and execute the reaction control strategy, so as to realize the control of the isoprene rubber polymerization reaction.

[0013] As one of the preferred solutions, after processing the multi-modal joint representation by the lightweight pre-trained model embedded with the Arrhenius equation constraint and outputting the reaction prediction result, it further includes: According to the reaction prediction result, a pre-trained hierarchical architecture model is used to generate and execute the reaction control strategy, so as to realize the control of the isoprene rubber polymerization reaction; wherein, the hierarchical architecture model includes a bottom control layer, a middle control layer and a top control layer.

[0014] As one of the preferred solutions, the generating and executing the reaction control strategy by using a pre-trained hierarchical architecture model according to the reaction prediction result to realize the control of the isoprene rubber polymerization reaction includes: Design a temperature controller based on the PID algorithm to construct the bottom control layer, and input the deviation between the current temperature and the preset temperature into the bottom control layer for processing with the preset temperature upper limit as the first constraint, and output the control adjustment sequence; Construct the middle control layer according to the pre-trained LSTM neural network, input the reaction prediction result into the middle control layer for processing, output the predicted energy consumption, and establish a multi-objective optimization model with the maximization of conversion rate, the minimization of viscosity fluctuation and the minimization of energy consumption as the ultimate goal, and use the dynamic balance of the preset upper and lower limits of multiple reaction parameters and the raw material inlet and outlet rate as the second constraint, and solve the multi-objective optimization model by a preset optimization algorithm to obtain the predicted control sequence; In the top control layer, a reward function is constructed based on the first constraint, the second constraint and the ultimate goal, and the depth reinforcement learning algorithm is used to adjust the control adjustment sequence and the predicted control sequence to generate and execute the reaction control strategy, so as to realize the control of the isoprene rubber polymerization reaction.

[0015] The second aspect of the present invention provides a control system for isoprene rubber polymerization reaction, including: A data acquisition module for collecting real-time reaction data and real-time image data of the isoprene rubber polymerization reaction; A feature extraction module for respectively extracting features from the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain time series features and image features; A feature fusion module, which is used to fuse the temporal features and the image features based on a cross-modal attention mechanism to obtain a multi-modal joint representation; A reaction prediction module, which is used to process the multi-modal joint representation through a lightweight pre-trained model constrained by the Arrhenius equation and output a reaction prediction result; A strategy execution module, which is used to generate and execute a reaction control strategy according to the reaction prediction result through a fuzzy control method and an anti-saturation integral PID control algorithm to achieve the control of the isoprene rubber polymerization reaction.

[0016] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: (1) Through the synergistic effect of real-time data and image data, the physical phenomena and dynamic changes of the isoprene rubber polymerization reaction are revealed, and deep learning models are used to extract and fuse the features in the reaction process to improve the comprehensiveness of the reaction representation; (2) The Arrhenius equation is embedded in the pre-trained model to force the model to follow physical laws during the prediction process, so that while maintaining high efficiency, the physical rationality of the prediction is improved; (3) An intelligent control strategy is adopted to improve the stability and efficiency of reaction control. Combining deep learning models and intelligent control algorithms, intelligent monitoring and effective control of the reaction process are realized, manual intervention is reduced, production efficiency and consistency are improved, and further, production safety and economy are improved. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for implementation will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of a method for controlling an isoprene rubber polymerization reaction provided by an embodiment of the present invention; Figure 2 It is a structural diagram of a system for controlling an isoprene rubber polymerization reaction provided by an embodiment of the present invention; Reference Signs: Among them, 10. Data acquisition module; 20. Feature extraction module; 30. Feature fusion module; 40. Reaction prediction module; 50. Strategy execution module. Detailed Embodiments

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0020] In the description of the present application, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first", "second", "third", etc. may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise stated, the meaning of "a plurality" is two or more.

[0021] In the description of the present application, it should be noted that unless otherwise clearly defined and limited, the terms "mounted", "connected", "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the internal communication of two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are only for the purpose of illustration and do not indicate or imply that the system or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation of the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0022] In the description of the present application, it should be noted that unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field of the present invention. The terms used in the specification of the present invention are only for describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0023] In one embodiment, as Figure 1 shown, the first aspect of the present invention provides a method for controlling isoprene rubber polymerization reaction, including: S1. Collect real-time reaction data and real-time image data of the isoprene rubber polymerization reaction; Specifically, a variety of sensors are installed on the reactor or reaction kettle for isoprene rubber polymerization reaction to collect real-time reaction data in the reaction, such as time series data of temperature, pressure, viscosity, stirring power, reaction time, etc. And real-time image data of the reaction is collected from multiple angles through a high-temperature resistant camera or an industrial camera to capture the material color, texture, transparency, bubbles, particle distribution, and wall sticking situation of the polymerization reaction, etc. And the collected data is cleaned and synchronized to ensure that the time stamps of the sensor data and the image frames are aligned. In addition, to facilitate the acquisition of real-time images and reaction data, the reactor or reaction kettle used in the reaction is colorless or transparent.

[0024] S2. Respectively perform feature extraction on the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain time series features and image features; In one embodiment, step S2 includes: Use a pre-trained long short-term memory network to extract the temperature change rate and pressure fluctuation amplitude in the real-time reaction data to generate time series features; Process the real-time image data through a pre-trained convolutional neural network and image segmentation technology to obtain image features.

[0025] Specifically, the present invention calculates the temperature change rate and pressure fluctuation amplitude based on the real-time reaction data collected in real time, performs normalization processing on them and then inputs them into a pre-trained long short-term memory network for processing to output time series features; then uses a pre-trained convolutional neural network and image segmentation technology to perform segmentation processing on the real-time image data to extract the image features in the real-time image. The present invention processes the real-time reaction data and the real-time image data through a deep learning model to capture the multi-dimensional dynamic features of the isoprene rubber polymerization reaction, reduce the deviation of a single data source, and improve the accuracy of prediction.

[0026] In one embodiment, the process of processing the real-time image data through a pre-trained convolutional neural network and image segmentation technology to obtain image features includes: Perform grayscale and binarization processing on the real-time image data to obtain a real-time processed image; Use the shallow convolutional kernels of the convolutional neural network to extract the color distribution features of the real-time processed image to obtain a color histogram; Extract the texture roughness of the real-time processed image through the middle convolutional layer of the convolutional neural network, and process the texture roughness based on the deep convolutional layer of the convolutional neural network to obtain texture features; Use the optical flow method to calculate the colloid displacement field between adjacent frames in the real-time processed image to generate a dynamic viscosity change feature vector; Detect the bubbles in the real-time processed image using a pre-trained segmentation model to obtain bubble features; Integrate the color histogram, the texture features, the dynamic viscosity change feature vector, and the bubble features to obtain image features.

[0027] Specifically, features such as the transparency, color change, presence or absence of particles, and foam formation of the reaction materials during the isoprene rubber polymerization reaction are all related to the reaction degree. For example, as the reaction progresses, the viscosity increases, and different textures or flow characteristics will appear on the surface of the material, so the images captured by the camera will change. Based on this, the present invention uses machine learning technology and image segmentation technology to extract image features in real-time image data, which includes the following steps: The present invention first grayscales the real-time image data to reduce the computational amount and enhance the contrast, thereby retaining the key brightness information. Then, the median filtering method is used to eliminate the salt-and-pepper noise in the image to retain the clarity of the bubble edges. Then, Gaussian filtering is used to smooth the light scattering noise in the image to improve the accuracy of texture analysis. Subsequently, the denoised real-time image data is binarized to segment the bubbles from the background, reduce the interference of uneven illumination, and obtain a binary mask image, that is, the real-time processed image.

[0028] Use the shallow convolutional kernel of the pre-trained CNN (such as the first convolution layer of VGG) to extract the color distribution features of the real-time processed image, and statistically analyze the pixel value distribution of the RGB channels to generate a color histogram to reflect the overall hue change of the reaction solution. And calculate the variance of each channel and take the mean value through the middle convolutional layer of the pre-trained CNN (such as Conv3 of VGG) to capture the scalar texture roughness value of the real-time processed image. Then use the deep convolutional layer of the pre-trained CNN (such as res5c of VGG) to process the scalar texture roughness value of the image, and apply global average pooling to generate texture features to capture the microstructure of the generation or phase separation of intermediate products and precipitates.

[0029] Taking the real-time processed image as the input, use the Farneback algorithm to calculate the displacement field, then calculate the amplitude of the displacement field, and statistically analyze the mean value, variance, and main direction histogram of the amplitude as the dynamic viscosity change feature vector, which can be used to quantify the macroscopic characteristics of fluid motion; in addition, a viscosity sensor such as a rotational viscometer can also be used to measure the material viscosity change data during the reaction as the dynamic viscosity change feature vector.

[0030] Then use the pre-trained segmentation model to segment the real-time processed image, and use the bubble kinematics model to detect the bubbles in each region, and then obtain the bubble features to indicate the severity of the reaction; finally, perform normalization processing and feature splicing processing on the color histogram, texture features, dynamic viscosity change feature vector, and bubble features to obtain the image features.

[0031] The present invention combines multi-dimensional features such as color, texture, dynamic viscosity, and bubbles, which can comprehensively describe the physical and chemical state of the reaction system, avoid the one-sidedness of single features, and improve the robustness of state judgment; by using grayscale and binarization processing, the complexity of the image can be quickly simplified, and the subsequent calculation amount can be reduced; by utilizing the hierarchical feature extraction ability of the pre-trained model, both low-order color texture and high-order semantic information are taken into account; by combining the physical model-driven (optical flow method) and data-driven (image segmentation technology) methods, the interpretability of dynamic features is enhanced.

[0032] In one embodiment, detecting bubbles in the real-time processed image by using the pre-trained segmentation model to obtain bubble features includes: Processing the real-time processed image by using the pre-trained MobileNetV3 model to obtain a bubble segmentation mask; Performing hole filling, noise filtering, and edge refinement on the bubble segmentation mask to obtain bubble connected components; Calculating the area, circularity, and centroid coordinates of each bubble connected component to construct bubble static features; Based on the centroid coordinates, using the SORT algorithm to associate bubbles in adjacent frames to obtain the continuous coordinates of each bubble to quantify the kinematic parameters of each bubble; Verifying each bubble by using Stokes' law to mark abnormal bubbles; Integrating the area, the circularity, the kinematic parameters, and the abnormal bubbles to obtain bubble features.

[0033] Specifically, the present invention uses MobileNetV3 as the encoder, which is pre-trained on a synthetic bubble dataset (Blender simulates different viscosities and lighting conditions), and uses it to process the real-time processed image to output a per-pixel bubble mask (0 / 1 binary image); then, morphological post-processing of the obtained bubble segmentation mask is performed, including: hole filling: performing morphological closing operation (5×5 circular kernel) on the segmentation mask to close the internal holes of the bubbles; noise filtering: removing small artifacts based on the connected component area threshold (such as particles with a diameter < 3 pixels are determined as noise); edge refinement: extracting the sub-pixel-level bubble contour through the Sobel operator to improve the positioning accuracy, and further obtaining multiple bubble connected components.

[0034] Calculate the area (which can be counted by pixels), circularity (4π * area / square of the perimeter), and centroid coordinates of each bubble connected region, and use them as static features of the bubbles; then, based on the centroid coordinates of each bubble connected region, use the SORT (Simple Online and Realtime Tracking) algorithm to associate the bubbles in adjacent frames, use the Kalman filter to predict the position in the next frame, use the Hungarian algorithm to match the detection box with the prediction box (IoU threshold > 0.7), and record the continuous trajectory coordinates (time series coordinates) of each bubble; calculate kinematic parameters such as instantaneous velocity, acceleration, and trajectory curvature based on the calculated coordinates, and use Stokes' law to verify each bubble to mark the unqualified ones as abnormal bubbles; summarize the bubble density (number of bubbles per unit area), average velocity variance (reflecting flow uniformity), and proportion of abnormal bubbles (indicating reaction abnormal risk) in each frame of the image, and integrate them with their area, circularity, kinematic parameters, and abnormal bubbles to obtain bubble characteristics.

[0035] The present invention quantifies the morphological characteristics of bubbles through static features (area, circularity) to distinguish bubble types (such as microbubbles, large bubbles); constructs dynamic features (movement speed, trajectory) and combines them with Stokes' law for verification to identify non-ideal moving bubbles caused by abnormal fluid viscosity or impurities; combines precise bubble characterization with dynamic tracking, uses abnormal marking to improve monitoring reliability, avoids misjudgment (such as mistaking impurity particles for bubbles), improves the accuracy of later prediction, and enhances the credibility of the physical model.

[0036] S3. Based on the cross-modal attention mechanism, fuse the temporal features and the image features to obtain a multi-modal joint representation; In one embodiment, step S3 includes: Encode the temporal features using the temporal self-attention layer of a pre-trained Transformer to obtain a temporal feature vector; Process the image features using the local blocks of a pre-trained Vision Transformer to generate an image feature vector; Based on the cross-modal attention mechanism, perform bidirectional interaction between the temporal feature vector and the image feature vector, and introduce a learning parameter matrix during the interaction to adjust the interaction intensity, generating a first enhanced representation and a second enhanced representation; Aggregate the first enhanced representation and the second enhanced representation through a gating mechanism, and introduce a conditional routing mechanism during the aggregation to dynamically select the fusion path based on the modal correlation between the first enhanced representation and the second enhanced representation to obtain a multi-modal joint representation.

[0037] Specifically, the present invention uses a variant of LSTM or the temporal self-attention layer of Transformer to encode temporal features to extract global time dependencies and generate temporal feature vectors; and uses the high-level features of a convolutional neural network or the local blocks of Vision Transformer to extract the spatial structure information of image features and generate image feature vectors.

[0038] Based on the cross-modal attention mechanism, a shared attention module is designed to allow the temporal feature vector and the image feature vector to interact bidirectionally as Query, Key, and Value respectively, i.e., temporal-to-image interaction: using the temporal feature vector as the query, and the image feature vector as the key and value to generate a temporally enhanced representation of the image, and image-to-temporal interaction: using the image feature vector as the query, and the temporal feature vector as the key and value to generate an image-enhanced representation of the time series; then a learning parameter matrix is introduced during the interaction process to dynamically adjust the interaction intensity between different modalities, thereby avoiding a single modality from dominating the fusion process, and taking the temporally enhanced representation of the image after the interaction as the first enhanced representation, and taking the image-enhanced representation of the time series after the interaction as the second enhanced representation.

[0039] Use a gating mechanism (such as the gating unit of GRU or gated convolution) to calculate the modality correlation and its corresponding weights between the first enhanced representation and the second enhanced representation, and introduce a conditional routing mechanism (such as soft routing based on feature similarity) to quantify the relationship between the difference in weights of the two enhanced representations and a preset difference. When the calculated difference is less than the preset difference, weighted fusion is performed on the corresponding enhanced representation using the calculated weights to obtain a multi-modal joint representation; when the calculated difference is not less than the preset difference, gated splicing is performed on the corresponding enhanced representation using the Sigmoid function and the calculated weights to obtain a multi-modal joint representation.

[0040] The present invention captures the potential correlation between temporal features and image features through a cross-modal attention mechanism for in-depth fusion of complementary information; adopts a gating mechanism and a conditional routing to dynamically adjust the fusion weights according to the modality correlation, avoiding the modality dominance deviation caused by fixed weights, and improving the robustness of the model in complex scenarios; has high flexibility and can adapt to different modality data distributions and task requirements; has strong robustness and has good fault tolerance for modality missing or noisy data.

[0041] S4. Process the multi-modal joint representation through a lightweight pre-trained model constrained by the Arrhenius equation to output a reaction prediction result; In one embodiment, step S4 includes: Construct the Arrhenius equation based on the isoprene rubber polymerization reaction, and use the temperature data in the real-time reaction data as an explicit feature to input into the Arrhenius equation to obtain the Arrhenius auxiliary feature; Concatenate the multi-modal joint representation, the temperature data, and the Arrhenius auxiliary feature into an input tensor, and input the input tensor into the lightweight pre-trained model for processing, and output the predicted reaction conversion rate and the predicted viscosity change rate for a future preset time period as the reaction prediction result.

[0042] Specifically, the present invention constructs the Arrhenius equation based on the activation energy Ea and the pre-exponential factor A of the isoprene rubber polymerization reaction: k(T)=A*e −Ea / (R*T) ; where k(T) is the temperature-dependent reaction rate constant; T is the real-time temperature; R is the gas constant; the activation energy can be determined by fitting the slope -Ea / R of the historical reaction data (temperature-conversion curve). Then, the real-time temperature data in the real-time reaction data is used as an explicit feature to input into the Arrhenius equation to calculate the linear combination of the logarithm of the rate constant and the reciprocal of the temperature, and the Arrhenius auxiliary feature FArr = [lnk(T), 1 / T] is obtained, which is used to characterize the reaction kinetic characteristics.

[0043] Concatenate the multi-modal joint representation, the real-time temperature data, and the Arrhenius auxiliary feature into an input tensor according to the time step or the feature dimension to input into the lightweight pre-trained model (such as TCN, GNN, etc.) for processing, and output the predicted reaction conversion rate and the predicted viscosity change rate for a future preset time period (such as the next 5 minutes) and use them as the reaction prediction result. In addition, the lightweight pre-trained model can also be a dual-branch network; the physical constraint branch layer in this network embeds an Arrhenius equation calculation layer to enforce the output to be strongly related to the reaction rate constant, and the data-driven branch layer in this network uses a 1D convolutional layer to extract temporal-spatial features, combines a Transformer encoder to capture long-range dependencies, then maps the obtained features to the conversion rate through a fully connected layer, corrects the deviation in combination with the Arrhenius equation, and then outputs the dynamic viscosity based on the temperature-viscosity-concentration multi-variable model constructed using historical data, and compensates for the non-linear effect through a residual connection, so as to obtain the reaction prediction result.

[0044] By explicitly embedding the Arrhenius equation, the present invention injects the prior knowledge of reaction kinetics (the exponential effect of temperature on reaction rate) into the pre-trained model to avoid pure data-driven black-box prediction and improve the physical rationality and interpretability of the prediction results. The pre-trained model compresses the number of parameters through knowledge distillation while retaining the key information of multi-modal features to meet the real-time inference requirements of industrial scenarios. By integrating physical equation features (Arrhenius auxiliary features), multi-modal joint representation (image + time series), and original temperature data, the coupling effect of macroscopic physical property changes and microscopic reaction dynamics can be covered, that is, the temperature fluctuation quantifies the reaction rate adjustment through the Arrhenius equation. The implicit influence of by-product accumulation is captured through multi-modal features, thereby enhancing the prediction ability.

[0045] In another embodiment, the input layer of the lightweight pre-trained model is a high-dimensional input tensor spliced by multi-modal joint representation, temperature data, and Arrhenius auxiliary features. A depthwise separable convolution (MobileNet Block) and a gated linear unit (GLU) are used to construct the encoder, and a double-branch fully connected layer is used as the output layer to predict the conversion rate and the viscosity change rate respectively. An explicit Arrhenius constraint layer (i.e., a physics-guided neural network layer) is added to the model for loss calculation.

[0046] The present invention optimizes the model parameters in stages to gradually enhance the influence of physical constraints, specifically: In the pre-training stage (without physical constraints), historical data or ImageNet is used as the sample set, enabling the model to initially learn the mapping relationship between multi-modal features and reaction results. During the training process, only the task loss (MSE) function is used, and the Adam optimizer is used to optimize the model parameters to obtain the basic model parameters.

[0047] In the fine-tuning stage (introducing physical constraints), the MSE function is combined with the physical constraint loss function designed based on the Arrhenius equation and the differential equation of the conversion rate constructed therefrom to obtain the total loss function to optimize the basic model parameters during the training process. Among them, the physical constraint loss function is represented by the following formula:

[0048]

[0049] In the formula, L phy is the physical constraint loss function; N is the total number of iterations; i is the number of iterations; is the predicted value of the reaction conversion rate; t is the time; n is the kinetic order to which the isoprene rubber polymerization reaction belongs; CIt is the reaction conversion rate.

[0050] In the total loss function, the initial weight of the physical constraint loss function is set to 0.1, and it increases by 0.1 in each subsequent round. However, if the decline rate of the physical constraint loss function is less than 5%, the weight corresponding to the physical constraint loss function is frozen until the next stage. If the MSE loss function rises, early stopping is triggered (Rollback to the previous round of parameters), and the RAdam optimizer is used to fine-tune the model parameters during training to obtain the adjusted parameters.

[0051] In the stage of strengthening physical constraints (high-weight constraints), the model is forced to strictly follow the Arrhenius equation, and the weight of the physical constraint loss function is fixed at 1.0, and higher-order physical constraints are added to form a higher-order loss function L phy-high , which is represented by the following formula:

[0052] In the formula, γ is a parameter, and its value is 0.5, which is only enabled when the physical constraint loss function is less than 0.01.

[0053] During the training process, a second-order optimizer (L-BFGS) is used to improve the convergence accuracy, and the input layer is frozen, and only the output head and the Arrhenius constraint layer are fine-tuned to obtain the updated parameters and verify the accuracy of the model after configuring the updated parameters. When its accuracy does not meet the requirements, the fine-tuning stage and the stage of strengthening physical constraints are iteratively executed until the final model accuracy meets the preset requirements, and then the trained lightweight pre-trained model can be obtained.

[0054] The present invention uses a multi-round training mechanism to train the lightweight pre-trained model. Among them, there is no constraint in the pre-training stage, and multi-modal feature learning is introduced; in the fine-tuning stage, the weight of the physical constraint loss function is gradually increased to balance physics and data; in the stage of strengthening constraints, higher-order physical terms are used to ensure strict kinetic consistency; by injecting physical constraints in stages, the lightweight pre-trained model can not only learn complex patterns from data during training, but also strictly follow the reaction kinetics law, greatly improving the prediction accuracy and achieving high-precision and interpretable reaction prediction in industrial scenarios.

[0055] S5. According to the reaction prediction result, a reaction control strategy is generated and executed through a fuzzy control method and an anti-saturation integral PID control algorithm to achieve the control of the isoprene rubber polymerization reaction; In one embodiment, step S5 includes: Obtain the reaction error corresponding to the reaction prediction result according to the transfer learning method, and decompose the reaction error into a short-term error and a long-term error by using the moving window average method; Construct a dynamic weight function based on the long-term error and the short-term error, and use the dynamic weight function to quantify the first contribution ratio and the second contribution ratio corresponding to the fuzzy control method and the anti-saturation integral PID control algorithm respectively; Based on the reaction error, generate a first control signal and a second control signal through the fuzzy control method and the anti-saturation integral PID control algorithm, and use the first contribution ratio and the second contribution ratio for weighting to obtain a total control signal; According to the total control signal and the reaction error, adjust the pulse width, pulse interval and pulse amplitude through the gradient optimization algorithm, generate the reaction control strategy and execute it to realize the control of the isoprene rubber polymerization reaction.

[0056] Specifically, the present invention collects the reaction conversion rate and the viscosity change rate corresponding to different control inputs (such as temperature, catalyst flow rate, etc.) in historical reactions or historical simulations, and performs dimensionality reduction processing on these historical data to extract key features such as the conversion rate peak and the viscosity change slope. Then, according to the key features and the collected historical data, a mapping relationship table between the control input and the response feature is established to construct an input-response database; then, the Euclidean distance or cosine similarity is used to perform similarity matching on the current reaction conversion rate prediction value and the viscosity change rate prediction value with the historical response features, and the historical response with the highest similarity is selected, and the corresponding control input is extracted as the initial control strategy; subsequently, according to the process requirements, the target conversion rate and the target viscosity change rate are defined, and the error data of the similar historical response is used as the reaction error corresponding to the current reaction prediction result according to the transfer learning method, and the reaction error is decomposed into a short-term error (reflecting the rapid dynamic response demand) and a long-term error (reflecting the steady-state accuracy demand) by using the moving window average method.

[0057] Construct a dynamic weight function through the long-term error and the short-term error, which is represented by the following formula:

[0058] In the formula, is the dynamic weight function; t is the time; k 1, k 2 are adjustable parameters used to control the relative weights of the short-term error and the long-term error; e 1, e2 are the short-term error and the long-term error respectively.

[0059] Next, a fuzzy controller is designed. It takes the long-term error and short-term error as input variables and outputs a control signal. The fuzzy rule base can be designed based on process experience. For example, if the short-term conversion rate error is large and the short-term viscosity error is small, increase the reaction temperature; if the long-term viscosity error is large and the long-term conversion rate error is small, adjust the catalyst concentration, etc. An anti-windup integral PID controller is also designed. It takes the reaction error as input and outputs a compensation signal, adopting an anti-windup strategy. When the output control signal approaches the actuator limit, the integral term accumulation is dynamically frozen to prevent integral windup. Then, the control mode is dynamically selected according to the error scale characteristics: Fuzzy dominant mode: When the short-term error is significant, the signal output by the fuzzy controller is the main one; PID dominant mode: When the long-term error is significant, the signal output by the anti-windup integral PID controller is the main one; Hybrid mode: When both the short-term and long-term errors cannot be ignored, after weighted summation of their output signals according to their weights, the first contribution ratio and the second contribution ratio corresponding to the fuzzy control method and the anti-windup integral PID control algorithm are obtained respectively.

[0060] The reaction error and its corresponding long-term and short-term errors are respectively input into the anti-windup integral PID controller and the fuzzy controller for processing, generating the second control signal corresponding to the anti-windup integral PID controller and the first control signal corresponding to the fuzzy controller respectively. Then, the corresponding control signals are weighted by the first contribution ratio and the second contribution ratio respectively, and combined with the switching logic to generate the total control signal.

[0061] Finally, the total control signal is parameterized into three core variables: Pulse width (PWM): Controls the action time of the actuator; Pulse interval (PI): Controls the dormancy time of the actuator; Pulse amplitude (PA): Controls the output intensity of the actuator. The objective function of pulse control is defined by comprehensively considering the adjustment time (system response speed), overshoot (system stability), and energy consumption (actuator energy consumption), and the pulse width, pulse interval, and pulse amplitude are optimized and adjusted through the gradient optimization algorithm. Then, combined with the state of the reaction error and the optimization results, the optimal pulse mode is dynamically selected as the reaction control strategy and sent to the actuator for execution to achieve the control of the isoprene rubber polymerization reaction. Among them, for the optimal pulse mode, if the error is large and fast response is required, the "high PWM + short PI" mode is selected; if the error is small and energy saving is required, the "low PWM + long PI" mode is selected, etc.

[0062] The present invention combines the non - linear processing ability of fuzzy control and the steady - state accuracy of anti - saturation PID. Through the dynamic weight real - time adjustment strategy, it can not only respond quickly under severe disturbances, but also eliminate long - term steady - state errors and adapt to complex and changeable reaction environments. The sliding window averaging method is used to separate short - term fluctuations and long - term trends, avoiding control oscillations or lags caused by single - time - scale errors. The gradient optimization algorithm dynamically adjusts pulse parameters, enabling the actuator (such as heaters, stirring motors, etc.) to achieve a balance between energy consumption and response speed, reducing equipment losses while effectively controlling the isoprene rubber polymerization reaction.

[0063] In one embodiment, after step S4, it further includes: According to the reaction prediction result, a pre - trained hierarchical architecture model is used to generate a reaction control strategy and execute it to control the isoprene rubber polymerization reaction; wherein, the hierarchical architecture model includes a bottom - layer control layer, a middle - layer control layer, and a top - layer control layer. In one embodiment, the step of using a pre - trained hierarchical architecture model to generate a reaction control strategy and execute it to control the isoprene rubber polymerization reaction according to the reaction prediction result includes: Based on the PID algorithm, a temperature controller is designed to construct the bottom - layer control layer, and with the preset temperature upper limit as the first constraint, the deviation between the current temperature and the preset temperature is input into the bottom - layer control layer for processing, and a control adjustment sequence is output. The middle - layer control layer is constructed according to a pre - trained LSTM neural network. The reaction prediction result is input into the middle - layer control layer for processing, and the predicted energy consumption is output. A multi - objective optimization model is established with the maximization of conversion rate, the minimization of viscosity fluctuation, and the minimization of energy consumption as the final goals. With the dynamic balance of the preset upper and lower limits of multiple reaction parameters and the raw material inlet and outlet rates as the second constraint, the multi - objective optimization model is solved through a preset optimization algorithm to obtain a predicted control sequence. In the top - layer control layer, a reward function is constructed based on the first constraint, the second constraint, and the final goal. The deep reinforcement learning algorithm is used to adjust the control adjustment sequence and the predicted control sequence, generate a reaction control strategy and execute it to control the isoprene rubber polymerization reaction.

[0064] Specifically, the present invention can also use a pre - trained hierarchical architecture model to generate a reaction control strategy, including: First, in the underlying control layer, the present invention designs a temperature controller based on the PID (Proportional-Integral-Derivative) algorithm. By adjusting the three parameters of proportional, integral, and derivative, it can achieve precise control of the temperature, and set a preset temperature upper limit as the first constraint to ensure that the reaction temperature will not be too high and avoid adverse effects on the reaction. Then, the deviation between the current temperature in the real-time reaction data and the preset temperature is input into the underlying control layer to be processed by the PID algorithm, and then a control adjustment sequence is output. This sequence contains specific instructions for temperature adjustment, such as the intensity and time of heating or cooling, and the adjustment amount of the cooling water flow rate, etc.

[0065] Next, the present invention constructs a middle control layer based on a pre-trained LSTM neural network. It can process sequence data and is suitable for processing the reaction prediction results. The reaction prediction results are input into the middle control layer to be processed by the LSTM neural network, and the predicted energy consumption of the polymerization reaction in a preset future period (such as the next 5 minutes) is output. Then, a multi-objective optimization model is established with the maximization of conversion rate, the minimization of viscosity fluctuation, and the minimization of energy consumption as the ultimate goals, and the preset upper and lower limits of temperature, the preset upper and lower limits of pressure, the preset upper and lower limits of stirring speed, and the dynamic balance of the raw material feeding and discharging rate are set as the second constraint to ensure the safety and stability of the reaction process. By using a preset optimization algorithm (such as genetic algorithm, particle swarm algorithm, etc.) to solve the multi-objective optimization model, a predicted control sequence can be obtained, which contains specific control strategies for achieving multi-objective optimization under different constraint conditions, such as the adjustment amount of stirring speed and the adjustment amount of raw material feeding rate, etc.

[0066] Finally, in the upper control layer, the current conversion rate, viscosity, temperature, pressure, energy consumption, safety risk level, etc. are used as the state space, and the adjustment of the weights of the objective function (such as increasing the priority of conversion rate or decreasing the priority of energy consumption), the fine-tuning of PID parameters, etc. are used as the action space to conduct environmental modeling. The first constraint and the second constraint and other hard constraints are encoded as penalty terms of the reward function, and a reward function is constructed based on the conversion rate, energy consumption, and safety risk in the ultimate goal. The control adjustment sequence and the predicted control sequence are adjusted by using a deep reinforcement learning algorithm (such as PPO, SAC) combined with a safe reinforcement learning (Safe RL) mechanism to generate a reaction control strategy and send it to the actuator for execution to achieve effective control of the isoprene rubber polymerization reaction.

[0067] Through a hierarchical architecture model, combined with PID algorithm, LSTM neural network and deep reinforcement learning algorithm, the present invention can achieve precise control of isoprene rubber polymerization reaction, improve product quality and production efficiency; when constructing the middle control layer, multiple objectives such as maximizing conversion rate, minimizing viscosity fluctuation and minimizing energy consumption are considered, and a multi-objective optimization model is used for solution, making the control strategy more comprehensive and optimized; the application of the deep reinforcement learning algorithm in the high-level control layer enables the control strategy to be dynamically adjusted according to the real-time reaction situation, improving the adaptability and robustness of the system; the design of the hierarchical architecture model of the present invention makes each control layer relatively independent, facilitating maintenance and upgrade, and also improving the scalability of the system.

[0068] In addition, the solution of the multi-objective optimization model can also be carried out by using a hybrid algorithm of differential evolution and dynamic programming. Among them, the differential evolution algorithm can generate new solutions through the population difference vector, quickly explore the global solution space, avoid falling into local optimum, and the dynamic programming algorithm can screen the optimal solutions that meet the constraints through the state transition equation among the candidate solutions generated by the differential evolution algorithm, realizing local fine-tuning. The differential evolution provides diversity, and the dynamic programming algorithm improves the solution quality, thereby reducing invalid calculations. The hybrid algorithm includes: Based on the predicted energy consumption, a multi-objective optimization model is constructed with the maximization of conversion rate, the minimization of viscosity fluctuation and the minimization of energy consumption as the ultimate goals, and the upper and lower limits of the preset temperature, the upper and lower limits of the preset pressure, the upper and lower limits of the preset stirring speed, and the dynamic balance of the raw material feeding and discharging rate are set as the second constraint.

[0069] In the differential optimization algorithm stage, an initial population is randomly generated, and each individual in the population satisfies the parameter upper and lower limit constraints; then for each individual, three different individuals are randomly selected to generate a mutation vector; a crossover operation is performed on each individual and the mutation vector to generate a trial vector; the objective function values of the initial individual and the trial vector are compared, and the better individual is selected to enter the next generation.

[0070] In the dynamic programming algorithm stage, the state is defined as the current reaction parameter combination, that is, the individual combination in the population, and the action is the parameter adjustment amount; according to the candidate solutions generated by the differential optimization algorithm, the objective function value of each state is calculated, and then the optimal transfer path is selected according to the calculation result of the objective function, and the solutions that do not meet the feeding and discharging rate balance are eliminated through the pruning strategy.

[0071] Finally, from the Pareto front solution set generated by DE-DP, the optimal solution is selected according to the process requirements (such as giving priority to conversion rate while restricting viscosity fluctuation and energy consumption), and the optimal solution is mapped to a specific reaction parameter control sequence, such as temperature, pressure, catalyst concentration, etc., to obtain the predictive control sequence.

[0072] The DE-DP algorithm adopted by the present invention provides an efficient and reliable solution for the precise control of isoprene rubber polymerization reaction through the collaboration of global and local optimization and the handling of dynamic constraints.

[0073] In addition, it should be noted that all the models adopted by the present invention are trained models, which can obtain historical reaction data and historical image data during the isoprene rubber polymerization reaction, and various historical training sets formed after annotating these historical data by measuring historical raw material conversion rates using an on-line analyzer (such as near-infrared spectroscopy), annotating the reaction degrees at each key time point in the historical reaction through laboratory sampling, historical control strategies, etc. They can also be pre-trained on large data sets such as similar time-series tasks (such as industrial sensor data) and ImageNet.

[0074] In the embodiments of the present application, based on the problem of how to effectively control the isoprene rubber polymerization reaction and thereby improve its reaction efficiency and safety, a control method for isoprene rubber polymerization reaction is designed. It extracts the dynamic change rules of key parameters such as temperature, pressure, and conversion rate through a deep learning model to capture the reaction kinetic characteristics, analyzes the spatial characteristics such as the fluid morphology and bubble distribution in the reactor using a pre-trained model to reveal local reaction heterogeneity, and dynamically correlates the time-series features and image features based on a cross-modal attention mechanism to enhance the comprehensiveness of feature expression; embeds the Arrhenius equation as prior knowledge through a lightweight model to reduce the dependence on a large amount of data and improve the generalization ability in small-sample scenarios. Under abnormal working conditions such as temperature fluctuations or catalyst deactivation, physical constraints can correct the model prediction deviation and avoid the "black box" problem of pure data-driven methods; adjusts the control parameters dynamically according to the real-time input of multi-modal joint representation using a fuzzy control method, and uses an anti-saturation integral PID control method to supplement the performance degradation problem of traditional PID under integral saturation to ensure that the control quantity is smoothly adjusted within a safe range; through the deep integration of data-driven (deep learning) and knowledge-driven (physical equations), combined with intelligent control (fuzzy + PID), the present invention realizes a precise perception-prediction-control closed loop for isoprene rubber polymerization reaction, has significant advantages in improving product quality, process safety and economic benefits, and provides an innovative paradigm for the intelligent upgrading of complex chemical processes.

[0075] It should be noted that although the steps in the above flowcharts are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders.

[0076] In another embodiment, as Figure 2 shown, the second aspect of the present invention provides a control system for isoprene rubber polymerization reaction, including: A data acquisition module 10 for collecting real-time reaction data and real-time image data of isoprene rubber polymerization reaction; A feature extraction module 20 for respectively extracting features from the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain temporal features and image features; A feature fusion module 30 for fusing the temporal features and the image features based on a cross-modal attention mechanism to obtain a multi-modal joint representation; A reaction prediction module 40 for processing the multi-modal joint representation through a lightweight pre-trained model embedded with the Arrhenius equation constraint and outputting a reaction prediction result; A strategy execution module 50 for generating and executing a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm according to the reaction prediction result to achieve the control of the isoprene rubber polymerization reaction. [[ID=lo]]

[0077] It should be noted that each module in the above isoprene rubber polymerization reaction control system can be implemented in whole or in part by software, hardware and their combination. The above modules can be embedded in the processor in the computer device in the form of hardware or independent of it, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific limitations of an isoprene rubber polymerization reaction control system, refer to the limitations of an isoprene rubber polymerization reaction control method in the above text. The two have the same functions and effects and will not be elaborated here.

[0078] In summary, the present invention relates to the technical field of data processing, and discloses an isoprene rubber polymerization reaction control method and system, which includes collecting real-time reaction data and real-time image data of isoprene rubber polymerization reaction; respectively extracting features from the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain temporal features and image features; fusing the temporal features and the image features based on a cross-modal attention mechanism to obtain a multi-modal joint representation; processing the multi-modal joint representation through a lightweight pre-trained model embedded with the Arrhenius equation constraint and outputting a reaction prediction result; generating and executing a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm according to the reaction prediction result to achieve the control of the isoprene rubber polymerization reaction; and realizing the effective control of the isoprene rubber polymerization reaction by deeply fusing multi-source data and various advanced algorithms.

[0079] Each embodiment in this specification is described in a progressive manner. For parts that are the same or similar in each embodiment, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, reference can be made to the corresponding description in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0080] The above-described embodiments merely represent several preferred embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the technical principle of the present invention, several improvements and substitutions can be made, and these improvements and substitutions should also be regarded as within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the protection scope of the claimed rights.

Claims

1. A method for controlling the polymerization reaction of isoprene rubber, characterized in that, Including: Collecting real-time reaction data and real-time image data of isoprene rubber polymerization reaction; Respectively extracting features from the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain temporal features and image features; Fusing the temporal features and the image features based on a cross-modal attention mechanism to obtain a multi-modal joint representation; Processing the multi-modal joint representation through a lightweight pre-trained model constrained by the Arrhenius equation to output a reaction prediction result; According to the reaction prediction result, generating and executing a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm to achieve the control of the isoprene rubber polymerization reaction.

2. The isoprene rubber polymerization reaction control method according to claim 1, characterized in that, The respectively extracting features from the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain temporal features and image features includes: Using a pre-trained long short-term memory network to extract the temperature change rate and pressure fluctuation amplitude in the real-time reaction data to generate temporal features; Processing the real-time image data through a pre-trained convolutional neural network and image segmentation technology to obtain image features.

3. The method for controlling the isoprene rubber polymerization reaction according to claim 2, wherein, The processing the real-time image data through a pre-trained convolutional neural network and image segmentation technology to obtain image features includes: Performing grayscale and binarization processing on the real-time image data to obtain a real-time processed image; Using a shallow convolutional kernel of the convolutional neural network to extract the color distribution features of the real-time processed image to obtain a color histogram; Extracting the texture roughness of the real-time processed image through a middle convolutional layer of the convolutional neural network and processing the texture roughness based on a deep convolutional layer of the convolutional neural network to obtain texture features; Calculating the colloidal displacement field between adjacent frames in the real-time processed image by using an optical flow method to generate a dynamic viscosity change feature vector; Detecting bubbles in the real-time processed image by using a pre-trained segmentation model to obtain bubble features; Integrating the color histogram, the texture features, the dynamic viscosity change feature vector and the bubble features to obtain image features.

4. The method for controlling the isoprene rubber polymerization reaction according to claim 3, wherein The detecting bubbles in the real-time processed image by using a pre-trained segmentation model to obtain bubble features includes: Processing the real-time processed image by using a pre-trained MobileNetV3 model to obtain a bubble segmentation mask; Performing hole filling, noise filtering and edge thinning processing on the bubble segmentation mask to obtain a bubble connected domain; Calculating the area, circularity and centroid coordinates of each bubble connected domain to construct bubble static features; Based on the centroid coordinates, using the SORT algorithm to associate bubbles in adjacent frames to obtain the continuous coordinates of each bubble to quantify the kinematic parameters of each bubble; Verifying each bubble by using Stokes' law to mark abnormal bubbles; Integrating the area, the circularity, the kinematic parameters and the abnormal bubbles to obtain bubble features.

5. A method for controlling isoprene rubber polymerization reaction according to claim 1, characterized in that, Based on the cross-modal attention mechanism, the temporal features and the image features are fused to obtain a multi-modal joint representation, including: Encoding the temporal features using the temporal self-attention layer of a pre-trained Transformer to obtain a temporal feature vector; Processing the image features using the local blocks of a pre-trained Vision Transformer to generate an image feature vector; Based on the cross-modal attention mechanism, performing bidirectional interaction between the temporal feature vector and the image feature vector, and introducing a learning parameter matrix during the interaction to adjust the interaction intensity, generating a first enhanced representation and a second enhanced representation; Aggregating the first enhanced representation and the second enhanced representation through a gating mechanism, and introducing a conditional routing mechanism during the aggregation to dynamically select a fusion path based on the modal correlation between the first enhanced representation and the second enhanced representation, obtaining a multi-modal joint representation.

6. The method for controlling isoprene rubber polymerization reaction according to claim 1, wherein Processing the multi-modal joint representation through a lightweight pre-trained model embedded with the Arrhenius equation constraint, and outputting a reaction prediction result, including: Constructing an Arrhenius equation based on the isoprene rubber polymerization reaction, and inputting the temperature data in the real-time reaction data as an explicit feature into the Arrhenius equation to obtain an Arrhenius auxiliary feature; Concatenating the multi-modal joint representation, the temperature data, and the Arrhenius auxiliary feature into an input tensor, and inputting the input tensor into the lightweight pre-trained model for processing, outputting the predicted reaction conversion rate and the predicted viscosity change rate for a future preset period as the reaction prediction result.

7. A method for controlling the isoprene rubber polymerization reaction according to claim 1, characterized in that, According to the reaction prediction result, generating and executing a reaction control strategy through a fuzzy control method and an anti-saturation integral PID control algorithm to achieve the control of the isoprene rubber polymerization reaction, including: Obtaining the reaction error corresponding to the reaction prediction result according to the transfer learning method, and decomposing the reaction error into a short-term error and a long-term error using the sliding window averaging method; Constructing a dynamic weight function through the long-term error and the short-term error, and quantifying the first contribution ratio and the second contribution ratio corresponding to the fuzzy control method and the anti-saturation integral PID control algorithm respectively using the dynamic weight function; Based on the reaction error, generating a first control signal and a second control signal through the fuzzy control method and the anti-saturation integral PID control algorithm, and weighting them using the first contribution ratio and the second contribution ratio to obtain a total control signal; According to the total control signal and the reaction error, adjusting the pulse width, the pulse interval, and the pulse amplitude through a gradient optimization algorithm, generating and executing the reaction control strategy to achieve the control of the isoprene rubber polymerization reaction.

8. A method for controlling isoprene rubber polymerization reaction according to claim 1, characterized in that, After processing the multi-modal joint representation through a lightweight pre-trained model embedded with the Arrhenius equation constraint and outputting the reaction prediction result, it further includes: According to the reaction prediction result, a pre-trained hierarchical architecture model is used to generate a reaction control strategy and execute it to achieve the control of the isoprene rubber polymerization reaction; wherein, the hierarchical architecture model includes a bottom control layer, a middle control layer, and a top control layer.

9. A method for controlling the isoprene rubber polymerization reaction according to claim 8, wherein The step of using a pre-trained hierarchical architecture model to generate a reaction control strategy and execute it according to the reaction prediction result to achieve the control of the isoprene rubber polymerization reaction includes: Designing a temperature controller based on the PID algorithm to construct the bottom control layer, taking the preset temperature upper limit as the first constraint, inputting the deviation between the current temperature and the preset temperature into the bottom control layer for processing, and outputting a control adjustment sequence; Constructing the middle control layer according to a pre-trained LSTM neural network, inputting the reaction prediction result into the middle control layer for processing, outputting the predicted energy consumption, and establishing a multi-objective optimization model with the maximization of conversion rate, the minimization of viscosity fluctuation, and the minimization of energy consumption as the ultimate goals, taking the dynamic balance of the preset upper and lower limits of multiple reaction parameters and the raw material inlet and outlet rates as the second constraint, and solving the multi-objective optimization model through a preset optimization algorithm to obtain a predictive control sequence; In the top control layer, a reward function is constructed based on the first constraint, the second constraint, and the ultimate goal, and a deep reinforcement learning algorithm is used to adjust the control adjustment sequence and the predictive control sequence, generate a reaction control strategy and execute it to achieve the control of the isoprene rubber polymerization reaction.

10. An isoprene rubber polymerization reaction control system, characterized in that, It includes: A data acquisition module for collecting real-time reaction data and real-time image data of the isoprene rubber polymerization reaction; A feature extraction module for respectively extracting features from the real-time reaction data and the real-time image data based on a pre-trained deep learning model to obtain temporal features and image features; A feature fusion module for fusing the temporal features and the image features based on a cross-modal attention mechanism to obtain a multi-modal joint representation; A reaction prediction module for processing the multi-modal joint representation through a lightweight pre-trained model embedded with the Arrhenius equation constraint and outputting a reaction prediction result; A strategy execution module for generating a reaction control strategy and executing it according to the reaction prediction result through a fuzzy control method and an anti-saturation integral PID control algorithm to achieve the control of the isoprene rubber polymerization reaction.

Images