Reaction kettle running state monitoring method, system and electronic equipment

Abstract

The application provides a reaction kettle operation state monitoring method, system and electronic equipment, and relates to the field of reaction kettle pre-operation state monitoring. The reaction kettle operation state monitoring method comprises the following steps: acquiring visual features and process parameter features of the operation of a reaction kettle; spatializing the process parameter features based on the visual features to obtain process parameter spatialized features; obtaining enhanced features according to the visual features and the process parameter spatialized features; and performing dynamic semantic segmentation based on the visual features and the enhanced features to monitor the operation state of the reaction kettle. The reaction kettle operation state monitoring method provided in the application embodiment can fuse image and sensor information, and can monitor the operation state of the reaction kettle in real time (leakage, local overheating, material wall sticking, mechanical failure and the like), thereby improving the accuracy of the reaction kettle operation state monitoring and the safety of the reaction kettle operation process.

Description

Technical Field

[0001] This application relates to the field of reactor operation status monitoring, and more specifically, to a reactor operation status monitoring method, system, and electronic device. Background Technology

[0002] Reactors are key equipment in modern chemical production, integrating high temperature, high pressure, stirring, and closed control. They are widely used in chemical, pharmaceutical, and energy fields. By precisely controlling parameters such as temperature, pressure, and material ratios, they achieve efficient and controllable material conversion and synthesis. Typical applications include organic / inorganic synthesis, catalytic reactions, drug synthesis, bio-fermentation, petroleum cracking, battery material preparation, and food additive production. Their versatility makes them a core device in industries such as chemical, pharmaceutical, and new energy.

[0003] However, the safe production of reactors faces many challenges, including the risk of leakage or explosion caused by extreme operating conditions (high temperature and high pressure), long-term corrosion of equipment by corrosive media, chain side reactions caused by uncontrolled reactions, and safety hazards such as response delays caused by human operation errors. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, and electronic device for monitoring the operating status of a reactor. This method integrates visual information and process information to achieve multi-dimensional real-time monitoring of the reactor's operating status, significantly improving the early warning capability of potential safety hazards during the safe operation of the reactor.

[0005] In a first aspect, embodiments of this application provide a method for monitoring the operating status of a reactor. The method includes: acquiring visual features and process parameter features of the reactor operation; spatializing the process parameter features based on the visual features to obtain spatialized process parameter features; obtaining enhanced features based on the visual features and the spatialized process parameter features; and performing dynamic semantic segmentation based on the visual features and the enhanced features to monitor the operating status of the reactor.

[0006] In the above implementation process, the embodiments of this application achieve high-precision real-time monitoring of the reactor's operating status through multimodal feature fusion and dynamic attention mechanism. By spatially aligning and adaptively weighting visual features with process parameters (such as temperature and pressure), the segmentation accuracy of key areas (such as liquid surface and bubbles) is significantly improved, thereby significantly enhancing the accuracy of anomaly detection and reducing the false alarm rate. This provides a reliable intelligent monitoring method for chemical production processes using reactors.

[0007] Optionally, in this embodiment of the application, obtaining the enhanced feature based on the visual feature and the spatialized feature of the process parameter includes: concatenating the visual feature and the spatialized feature of the process parameter to obtain a concatenated feature; generating an attention weight based on the concatenated feature; and using the attention weight to spatially weight the visual feature to obtain the enhanced feature.

[0008] In the above implementation process, the reactor operation status monitoring method provided in this application embodiment achieves dynamic fusion of process parameters and visual features through sensor feature spatialization and cross-modal attention mechanism. It maps sensor signals (such as temperature) to local image regions, establishing an explicit connection between physical state and visual features; and automatically focuses on key regions (such as high-temperature deformation zones) through attention weights, improving the model's sensitivity to abnormal reactor operation status.

[0009] Optionally, in this embodiment, the step of performing dynamic semantic segmentation based on the visual features and enhanced features to monitor the operating status of the reactor includes: downsampling the enhanced features using an encoder to obtain detailed features; inputting the visual features into a multi-layer bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of the visual features; inputting the detailed features and multi-scale fused features into a skip connection layer to obtain dynamic fused features; upsampling the dynamic fused features using a decoder to obtain an initial semantic segmentation result; and optimizing the initial semantic segmentation result to monitor the operating status of the reactor.

[0010] In the above implementation process, in the embodiments of this application, the encoder retains key detailed features through adaptive 3D dynamic convolution of process parameters, constructs cross-level feature associations through multi-layer bidirectional pyramids, and dynamically balances local and global information through gated jump connections. While ensuring real-time performance, it achieves multi-scale accurate perception of the reactor status, reliably identifying risk states such as abnormal liquid levels and bubble accumulation, providing intelligent protection for the safe operation of industrial reactors.

[0011] Optionally, in this embodiment of the application, the step of inputting the visual features into a multi-level bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of the visual features includes: dividing the multi-level visual features into multiple groups; wherein the number of groups of visual features is the same as the number of layers of the bidirectional pyramid, and adjacent groups of visual features include at least two overlapping visual features; performing bidirectional fusion according to the levels to update the intermediate visual features of each level, until the bidirectional fusion of the last layer of the bidirectional feature pyramid is completed to obtain the multi-scale fused features.

[0012] In the above implementation process, the reactor operation status monitoring method provided in this application establishes cross-scale feature association, which not only retains the detailed information of low-level features (such as bubble edges) but also integrates the semantic context of high-level features (such as the overall shape of the liquid surface); the multi-level bidirectional feature pyramid structure supports repeated stacking, enhances the feature fusion capability, and significantly improves the detection sensitivity of the segmentation network for minor anomalies.

[0013] Optionally, in this embodiment of the application, the step of inputting the detailed features and multi-scale fusion features into the skip connection layer to obtain dynamic fusion features includes: generating dynamic gating weights based on the detailed features, multi-scale fusion features and process parameter features; and dynamically balancing the detailed features and multi-scale fusion features based on the dynamic gating weights to generate the dynamic fusion features.

[0014] In the above implementation process, the reactor operation status monitoring method provided in this application generates dynamic gating weights by jointly modeling detailed features, multi-scale fusion features, and process parameter features. This allows the system to autonomously adjust the feature fusion ratio according to different monitoring scenarios (such as edge detection, bubble recognition, or liquid surface morphology analysis), effectively enhancing the model's adaptability to complex reactor operating conditions. On the other hand, this method achieves dynamic enhancement of key features while maintaining the integrity of feature information. For example, detailed features such as microbubbles can be assigned a weight close to 1 to improve detection sensitivity, while the weight of global context features is adaptively reduced to maintain the stability of overall morphology analysis. This intelligent feature fusion mechanism based on process parameters not only overcomes the limitations of traditional skip-connection fixed-weight fusion but also significantly improves the state monitoring model's ability to represent multi-scale features, providing reliable technical support for refined reactor monitoring.

[0015] Optionally, in this embodiment of the application, optimizing the initial semantic segmentation result to monitor the operating status of the reactor includes: inputting the initial semantic segmentation result into a conditional random field model and adjusting the parameters of the conditional random field model; obtaining the output of the conditional random field model in the case of minimizing the energy function, and using the output as the updated semantic segmentation result; and analyzing the operating status of the reactor based on the updated semantic segmentation result to monitor the operating status of the reactor.

[0016] In the above implementation process, the reactor operation status monitoring method provided in this application embodiment uses a CRF model with dynamically adjusted parameters to refine the segmentation results, effectively eliminating edge jaggedness and noise interference, and significantly improving the boundary accuracy of key areas such as liquid surface and bubbles; and performs multi-dimensional state analysis based on the segmentation results. The reactor operation status monitoring method provided in this application embodiment can dynamically adjust the detection sensitivity according to the process conditions, and achieve comprehensive and reliable monitoring of the reactor operation status while ensuring real-time performance.

[0017] Optionally, in an embodiment of this application, the process parameter features include sensor features for monitoring the reaction process in the reactor.

[0018] Secondly, embodiments of this application provide a reactor operation status monitoring system, which includes: a feature acquisition module, a feature enhancement module, a semantic segmentation module, and a safety monitoring module; the feature acquisition module is used to acquire visual features and process parameter features of the reactor operation; the feature enhancement module is used to spatialize the process parameter features based on the visual features to obtain spatialized process parameter features; the semantic segmentation module is used to obtain enhanced features based on the visual features and the spatialized process parameter features; and the safety monitoring module is used to perform dynamic semantic segmentation based on the visual features and the enhanced features to monitor the operation status of the reactor.

[0019] Thirdly, embodiments of this application provide an electronic device, which includes a memory and a processor. The memory stores program instructions, and when the processor reads and runs the program instructions, it executes the steps in any of the above implementation methods.

[0020] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the steps in any of the above implementations.

[0021] Fifthly, embodiments of this application also provide a computer program product, the computer program product including a computer program / instruction, which, when executed by a processor, performs the steps in any of the above implementation methods. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating the reactor operation status monitoring method provided in this application embodiment;

[0024] Figure 2 A flowchart illustrating the acquisition of enhanced features provided in embodiments of this application;

[0025] Figure 3 A flowchart of dynamic semantic segmentation provided for embodiments of this application;

[0026] Figure 4 A flowchart illustrating the generation process of multi-scale fusion features provided in embodiments of this application;

[0027] Figure 5 A flowchart illustrating the generation process of dynamic fusion features provided in this application embodiment;

[0028] Figure 6 A flowchart for optimizing the initial semantic segmentation results provided in the embodiments of this application;

[0029] Figure 7 This is a schematic diagram of the modules of the reactor operation status monitoring system provided in the embodiments of this application;

[0030] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. For example, the flowcharts and block diagrams in the drawings illustrate the architecture, functions, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, program segment, or part of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions. In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0032] Reactors are core equipment in modern chemical production, integrating high temperature, high pressure, stirring, and airtight control. They are widely used in chemical, pharmaceutical, and energy fields. By precisely controlling parameters such as temperature and pressure, they achieve efficient material conversion. Typical applications include organic synthesis, catalytic reactions, and fine chemical production in the chemical industry; drug synthesis and bio-fermentation in the pharmaceutical field; heavy oil cracking and polymer material preparation in the petrochemical industry; and key processes in the new energy, environmental protection, food, and daily chemical industries.

[0033] Currently, reactors face many challenges in safe production, such as the risk of leakage or explosion caused by extreme working conditions (high temperature and high pressure), the potential for structural failure of equipment due to corrosive media, the risk of runaway reaction (such as excessively rapid temperature rise or stirring failure) which may trigger chain side reactions, and safety hazards caused by human error (such as monitoring delay).

[0034] The inventors discovered that traditional monitoring technologies have significant limitations in the safe application of reactors. On the one hand, existing systems often rely on single sensors (such as pressure gauges or thermocouples) or manual inspections, resulting in limited data dimensions and making it difficult to comprehensively monitor multimodal risks (such as localized overheating, minor leaks, or material adhering to the vessel walls). On the other hand, existing reactor monitoring lacks real-time performance, with low frequency of manual inspections and sensor signal transmission delays, potentially leading to delayed warnings. Furthermore, they have poor adaptability to complex operating conditions; highly corrosive and dusty environments can easily cause sensor failure, while visual monitoring is susceptible to obstruction by steam and foam. In addition, there are blind spots in the monitoring of critical safety hazards such as abnormal sparks and smoke, increasing the risk of major accidents.

[0035] Based on this, this application proposes a method, system, and electronic device for monitoring the operating status of a reactor. This method integrates visual information and process information (such as acoustic fingerprint information and sensor information) to achieve multi-dimensional real-time monitoring of the reactor's operating status, significantly improving the early warning capability of safety hazards (leakage, overheating, and mechanical failure) during the safe operation of the reactor, and providing intelligent protection for safe production in industries such as chemical, pharmaceutical, and new energy.

[0036] Please refer to Figure 1 , Figure 1 This is a flowchart of a reactor operating status monitoring method provided in an embodiment of this application; this application provides a reactor operating status monitoring method that can... Figure 8 The electronic equipment is used for execution. This method for monitoring the operating status of the reactor includes the following steps:

[0037] Step S100: Obtain the visual characteristics and process parameter characteristics of the reactor operation.

[0038] In step S100 above, visual characteristics and process parameter characteristics of the reactor operation are obtained.

[0039] Among them, visual characteristics include the color and / or morphological characteristics of the materials in the reactor. Morphological characteristics include normal and abnormal morphological characteristics. Normal morphological characteristics include material viscosity, wall adhesion and splashing. Abnormal morphological characteristics include the generation of sparks, excessive accumulation of materials in some areas, etc.

[0040] Process parameter characteristics include temperature characteristics, acoustic characteristics, and pressure characteristics in the reactor, among which temperature and pressure characteristics can be obtained based on temperature and pressure sensors.

[0041] A multispectral imaging unit can be set up around the reactor. For example, the multispectral imaging unit includes a high-pressure explosion-proof camera group (including visible light / infrared / hyperspectral lenses) and a multi-angle stereo imaging device (including a rotating pan-tilt head and a wide-angle lens) to obtain visual information. The visual information is then processed to obtain visual features.

[0042] In the above implementation process, visual information and process parameter information are processed to obtain visual features and process parameter features, including step S110 to extract features from visual information to obtain visual features, and step S120 to extract features from process parameter information to obtain process parameter features.

[0043] The process of extracting visual features from visual information in step S110 above is as follows:

[0044] The input visual information, the RGB image (H×W×3), is preprocessed, for example, by normalization to scale pixel values ​​to [0,1] and by histogram equalization to enhance contrast (e.g., material edges in low-light environments).

[0045] Furthermore, multi-level convolutions are performed on the preprocessed visual information to obtain multi-scale visual features. For example, a lightweight CNN (such as ResNet-18) is used to extract spatial features, where some layers are used to extract local textures / edges (suitable for segmentation tasks to preserve details), and some layers are used to capture global semantics (such as material category semantics) to obtain visual features Fimg{P3,P4,P5}.

[0046] The above step S120, which extracts features from process parameter information to obtain process parameter features, will be explained using sensor data as an example:

[0047] Obtain sensor time-series data (T×S, where T is the time step and S is the number of sensors), standardize the sensor time-series data to eliminate dimensional differences, and perform linear interpolation to ensure time step alignment.

[0048] Furthermore, a bidirectional LSTM or Transformer encoder can be used to capture temporal dependencies, and the output of the last time step can be taken as the feature vector to obtain the sensor feature vector Fsen.

[0049] Step S200: Based on visual features, spatialize the process parameter features to obtain spatialized process parameter features.

[0050] In step S200 above, the process parameter features are spatialized based on visual features to obtain spatialized process parameter features. In this embodiment of the application, if it is desired to combine multimodal information, it is necessary to combine visual features and process parameter features. However, process parameter features lack spatial dimension and cannot directly interact with visual features. Therefore, it is necessary to first convert the process parameter features into process parameter spatial features with spatial dimension.

[0051] Step S300: Spatialize the features based on visual features and process parameters to obtain enhanced features.

[0052] In step S300 above, the visual features and process parameter spatial features are stitched together to obtain the enhanced features. In this embodiment, the enhanced features are the features obtained after aligning and interacting the visual features and process parameter features. They can map sensor signals (such as temperature) to local areas of the image and establish an explicit connection between physical state and visual features. At the same time, irrelevant backgrounds are suppressed and features related to the process state are enhanced, such as bubbles, liquid surface fluctuations, etc.

[0053] Step S400: Perform dynamic semantic segmentation based on visual features and enhanced features to monitor the operating status of the reactor.

[0054] In step S400 above, dynamic semantic segmentation is performed based on visual features and enhancement features to obtain semantic segmentation results, and the operating status of the reactor is monitored based on the semantic segmentation results.

[0055] pass Figure 1 As can be seen, the embodiments of this application achieve high-precision real-time monitoring of the reactor's operating status through multimodal feature fusion and dynamic attention mechanism. By spatially aligning and adaptively weighting visual features with process parameters (such as temperature and pressure), the segmentation accuracy of key areas (such as liquid surface and bubbles) is significantly improved, thereby significantly increasing the accuracy of anomaly detection and reducing the false alarm rate. This provides a reliable intelligent monitoring method for chemical production processes using reactors.

[0056] Please refer to Figure 2 , Figure 2The flowchart for obtaining enhanced features provided in this application embodiment is shown below. In an optional embodiment of this application embodiment, step S200 above, based on visual features, spatializes the process parameter features to obtain spatialized process parameter features, which can be achieved through the following steps:

[0057] Step S210: Concatenate the visual features and the spatialized features of the process parameters to obtain the concatenated features.

[0058] In step S210 above, the visual features and the spatialized features of the process parameters are spliced ​​together to obtain the spliced ​​features. Specifically, the visual Fimg and the spatialized sensor features Fsen_spatia are spliced ​​along the channel dimension to obtain the spliced ​​features Fca; thereby establishing the association between the local region of the image and the global state of the sensor. For example, the texture features of a certain region may be related to high temperature.

[0059] Step S220: Generate attention weights based on the spliced ​​features.

[0060] In step S220 above, attention weights are generated based on the spliced ​​features. The channels of the spliced ​​feature Fcat are compressed to 1 using a 1x1 convolution, allowing the use of a single-channel convolutional kernel to learn the weight distribution of cross-modal interactions. Furthermore, sigmoid activation is used to generate attention weights; the closer the value is to 1, the greater the influence of process parameters on that spatial location (e.g., enhancing the feature response in high-temperature regions).

[0061] Step S230: Spatially weight the visual features using attention weights to obtain enhanced features.

[0062] In step S230 above, visual features are enhanced according to attention weights to obtain enhanced features. Among them, high-weight regions enhance features related to process parameters, such as texture anomalies caused by temperature; low-weight regions suppress irrelevant background interference.

[0063] pass Figure 2 As can be seen, the reactor operation status monitoring method provided in this application realizes the dynamic fusion of process parameters and visual features through sensor feature spatialization and cross-modal attention mechanism. It realizes the mapping of sensor signals (such as temperature) to local areas of the image, establishing an explicit connection between physical state and visual features; and automatically focuses on key areas (such as high-temperature deformation zones) through attention weights, thereby improving the model's sensitivity to abnormal reactor operation status.

[0064] Please refer to Figure 3 , Figure 3A flowchart of dynamic semantic segmentation provided for embodiments of this application; in optional embodiments of this application, step S400 above, which performs dynamic semantic segmentation based on visual features and enhanced features to monitor the operating status of the reactor, can be implemented through the following steps:

[0065] Step S410: Use the encoder to downsample the enhanced features to obtain detailed features.

[0066] In step S410 above, the encoder is used to downsample the enhanced features to obtain detailed features. For example, 3D-DyConv dynamic convolution can be used to perform 4-level downsampling to adapt to changes in the state of the material in the reactor, thereby preserving more detailed features in the reactor, such as bubbles and liquid surface fluctuations.

[0067] Step S420: Input the visual features into a multi-level bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of the visual features.

[0068] In step S420 above, the original visual features are input into a multi-level bidirectional feature pyramid. The multi-level bidirectional feature pyramid processes the original visual features in sequence according to the hierarchy. After all the bidirectional feature pyramid levels have been processed, multi-scale fused features are obtained.

[0069] Bidirectional Feature Pyramid (BiFPN) is an efficient multi-scale feature fusion architecture that enhances the network’s feature extraction capabilities through bidirectional cross-scale connections and weighted feature fusion mechanisms.

[0070] Step S430: Input the detailed features and multi-scale fused features into the skip connection layer to obtain dynamic fused features.

[0071] In step S430 above, detailed features and multi-scale fused features are input into the skip connection layer to obtain dynamic fused features. Using the skip connection layer to connect the encoder's detailed features with the multi-scale fused features output by the multi-layer BiFPN can adaptively balance local details and global context, and by reasonably setting the weights, the biased features can be effectively controlled.

[0072] Step S440: Use the decoder to upsample the dynamically fused features to obtain the initial semantic segmentation result.

[0073] Step S450: Optimize the initial semantic segmentation results to monitor the operating status of the reactor.

[0074] In steps S440 to S450 above, the decoder is used to upsample the dynamically fused features to obtain initial semantic features, and upsampling is performed step by step to restore the resolution, resulting in an initial semantic segmentation result. Further, the initial semantic segmentation result is optimized (e.g., using CRF to optimize edge accuracy, morphological post-processing, or filter guidance) to obtain the final semantic segmentation result, thereby monitoring the operating status of the reactor based on the final semantic segmentation result.

[0075] pass Figure 3 As can be seen from the embodiments of this application, the encoder retains key detailed features through 3D dynamic convolution with adaptive process parameters, constructs cross-level feature associations through multi-layer bidirectional pyramids, and dynamically balances local and global information through gated jump connections. While ensuring real-time performance, it achieves multi-scale accurate perception of the reactor status, reliably identifying risk states such as abnormal liquid levels and bubble accumulation, providing intelligent protection for the safe operation of industrial reactors.

[0076] Please refer to Figure 4 , Figure 4 The flowchart for generating multi-scale fusion features provided in this application embodiment is shown below. In an optional embodiment of this application, step S420, which inputs visual features into a multi-layer bidirectional feature pyramid to obtain multi-scale fusion features after bidirectional fusion of the visual features, can be achieved through the following steps:

[0077] Step S421: Divide the multi-level visual features into multiple groups; wherein the number of groups of visual features is the same as the number of layers of the bidirectional pyramid, and adjacent groups of visual features include at least two overlapping visual features.

[0078] In step S421 above, for example, if the input visual features are Fimg{P3,P4,P5,P6,P7} and the multi-layer bidirectional feature pyramid is set to 3 layers, then the first layer processes {P3,P4,P5}, the second layer processes {P4,P5,P6}, and the third layer processes {P5,P6,P7}.

[0079] Step S422: Perform bidirectional fusion according to the hierarchy to update the intermediate visual features of each level until the bidirectional fusion of the last layer of the bidirectional feature pyramid is completed, and multi-scale fused features are obtained.

[0080] In step S422 above, bidirectional fusion is performed according to the hierarchy. First, taking the visual feature Fimg{P3,P4,P5} as an example, the feature fusion process of a single-layer BiFPN is as follows: the high-level semantics (P5) is propagated to the low-level (P4), that is, from top to bottom, to improve the classification accuracy of small objects. The high-level features are upsampled and fused with the low-level features (upsampling P5 to get P5up; weighted fusion of P5up and P4 to get P4td); and the low-level details (P3) are passed to the high-level (P4), that is, from bottom to top, to repair blurred edges. The low-level features are downsampled and fused with the high-level features (downsampling P3 to get p3down; weighted fusion of P4td and p3down, followed by lightweight convolution to get p4new).

[0081] In this embodiment, the input is Fimg{P3,P4,P5,P6,P7}, and the multi-layer bidirectional feature pyramid is set to 3 layers. The first layer yields p3new, the second layer yields p4new, and the third layer yields p5new, thereby establishing cross-scale feature association and enhancing the detection capability for targets of different sizes. The fusion weights of each layer are learnable weights.

[0082] pass Figure 4 As can be seen, the reactor operation status monitoring method provided in this application establishes cross-scale feature association, which not only retains the detailed information of low-level features (such as bubble edges) but also integrates the semantic context of high-level features (such as the overall shape of the liquid surface); the multi-level bidirectional feature pyramid structure supports repeated stacking, enhances the feature fusion capability, and significantly improves the detection sensitivity of the segmentation network for minor anomalies.

[0083] Please refer to Figure 5 , Figure 5 The flowchart for generating dynamic fusion features provided in this application embodiment is shown below. In an optional embodiment of this application, step S430, which inputs detailed features and multi-scale fusion features into the skip connection layer to obtain dynamic fusion features, can be implemented through the following steps:

[0084] Step S431: Generate dynamic gating weights based on detailed features, multi-scale fusion features, and process parameter features.

[0085] In step S431 above, dynamic gating weights are generated based on detailed features, multi-scale fusion features, and process parameter features. First, the detailed features, multi-scale fusion features, and process parameter features are concatenated; for the features generated after concatenation, a lightweight gating network is used to generate dynamic gating weights.

[0086] Step S432: Based on dynamic gating weights, dynamically balance detailed features and multi-scale fusion features to generate dynamic fusion features.

[0087] In step S432 above, the detailed features and multi-scale fusion features are dynamically balanced based on dynamic gating weights. It should be noted that the reactor operation status monitoring method provided in this application embodiment considers process parameter features when generating dynamic gating weights; for example, the weights can be adjusted to be closer to 1 for the detection of details such as edges and microbubbles, and closer to 0 for contextual analysis such as overall liquid surface morphology analysis. This application embodiment uses a skip connection layer to achieve adaptive fusion of cross-layer features between the encoder and decoder.

[0088] pass Figure 5 As can be seen, the reactor operation status monitoring method provided in this application generates dynamic gating weights by jointly modeling detailed features, multi-scale fusion features, and process parameter features. This allows the system to autonomously adjust the feature fusion ratio according to different monitoring scenarios (such as edge detection, bubble recognition, or liquid surface morphology analysis), effectively enhancing the model's adaptability to complex reactor operating conditions. On the other hand, this method achieves dynamic enhancement of key features while maintaining the integrity of feature information. For example, detailed features such as microbubbles can be assigned a weight close to 1 to improve detection sensitivity, while the weight of global context features is adaptively reduced to maintain the stability of overall morphology analysis. This intelligent feature fusion mechanism based on process parameters not only overcomes the limitations of traditional skip-connection fixed-weight fusion but also significantly improves the state monitoring model's ability to represent multi-scale features, providing reliable technical support for refined reactor monitoring.

[0089] Please refer to Figure 6 , Figure 6 The flowchart for optimizing the initial semantic segmentation result provided in this application embodiment is shown below. In an optional embodiment of this application, the above step S450, optimizing the initial semantic segmentation result to monitor the operating status of the reactor, can be achieved in the following ways:

[0090] Step S451: Input the initial semantic segmentation results into the conditional random field model and adjust the parameters of the conditional random field model.

[0091] In step S451 above, the initial semantic segmentation result is input into the Conditional Random Field (CRF) model, and the parameters of the CRF model are continuously adjusted. For example, the CRF model receives the initial segmentation result and the original RGB image, performs Softmax normalization on the initial segmentation result, and obtains a class probability map. The hyperparameters of the CRF can be adjusted according to the current process parameters of the reactor, for example, to enhance spatial constraints under high temperature and high pressure conditions.

[0092] Step S452: Obtain the output of the conditional random field model when the energy function is minimized, which is the updated semantic segmentation result.

[0093] Step S453: Based on the updated semantic segmentation results, analyze the operating status of the reactor to monitor its operation.

[0094] In steps S452 and S453 above, iterative optimization is continuously performed to minimize the energy function of the conditional random field model. The output of the conditional random field model with the minimum energy function is used as the updated semantic segmentation result. Furthermore, the operating status of the reactor is analyzed based on the updated semantic segmentation result, thereby enabling the monitoring of the reactor's operating status.

[0095] For example, bubble regions can be detected based on the segmentation results. Pixels representing bubble categories are extracted from the segmentation results, and the total number of bubble category pixels is calculated to obtain the bubble area during the current operation of the reactor. An alarm can be triggered based on the bubble area; for example, the alarm threshold can be adjusted according to the current reactor temperature: low temperature conditions (temp < 1500℃): the threshold is set to 500 pixels; high temperature conditions (temp ≥ 1500℃): the threshold is tightened to 300 pixels (higher risk of bubbles at high temperatures). If the bubble area exceeds the threshold, an alarm signal is returned, indicating a risk of bubble accumulation.

[0096] For example, the liquid surface condition can be evaluated based on the segmentation results. Liquid category pixels are extracted from the segmentation results to generate a binary image; a smoothness index is calculated, for example, using the Laplacian operator to detect edge intensity, thereby enabling turbulence anomaly detection. An appropriate threshold is set, and when the smoothness exceeds the threshold, a warning message, such as "liquid surface turbulence anomaly," is returned.

[0097] In an optional embodiment, taking the graphene reaction as an example, when abnormal conditions such as flames or blue smoke are detected, a three-level emergency response is immediately activated: a primary warning alerts the operator, a secondary intervention automatically reduces the feed rate, and in an emergency, the potassium permanganate feed is cut off within 0.5 seconds and protective measures are activated. To address the issue of lens yellowing, an image processing algorithm is used to eliminate smoke interference in real time, ensuring a clear and usable monitoring image.

[0098] In an optional embodiment, an intelligent analysis model is established based on the material's color spectral characteristics to accurately track the color change trajectory of the material during the reaction process, such as the graphene reaction, where the color changes from black to sapphire blue and then to dark green. When abnormal bubbles or violent reactions are detected, the model automatically correlates with multi-dimensional parameters such as temperature and pressure for comprehensive judgment, promptly stops feeding, and provides handling suggestions. Simultaneously, reaction parameters at each stage are continuously recorded to provide data support for process optimization.

[0099] pass Figure 6As can be seen, the reactor operation status monitoring method provided in this application embodiment uses a CRF model with dynamically adjusted parameters to refine the segmentation results, effectively eliminating edge jaggedness and noise interference, and significantly improving the boundary accuracy of key areas such as liquid surface and bubbles; and performs multi-dimensional state analysis based on the segmentation results. The reactor operation status monitoring method provided in this application embodiment can dynamically adjust the detection sensitivity according to the process conditions, and achieve comprehensive and reliable monitoring of the reactor operation status while ensuring real-time performance.

[0100] Please refer to Figure 7 , Figure 7 This is a schematic diagram of the modules of the reactor operation status monitoring system provided in the embodiments of this application; this application provides a reactor operation status monitoring system 100, which includes a feature acquisition module 110, a feature enhancement module 120, a semantic segmentation module 130 and a safety monitoring module 140.

[0101] The feature acquisition module 110 is used to acquire the visual features and process parameter features of the reactor operation.

[0102] The Feature Enhancement 120 module is used to spatialize process parameter features based on visual features, thereby obtaining spatialized process parameter features.

[0103] The semantic segmentation module 130 is used to spatialize features based on visual features and process parameters to obtain enhanced features.

[0104] The safety monitoring module 140 is used for dynamic semantic segmentation based on visual features and enhanced features to monitor the operating status of the reactor.

[0105] In an optional embodiment, during the process of obtaining enhanced features based on visual features and spatialized features of process parameters, the feature enhancement module 120 is specifically used to: concatenate the visual features and spatialized features of process parameters to obtain concatenated features; generate attention weights based on the concatenated features; and spatially weight the visual features using the attention weights to obtain enhanced features.

[0106] In an optional embodiment, during the process of performing dynamic semantic segmentation based on visual features and enhancement features to monitor the operating status of the reactor, the semantic segmentation module 130 is used to: downsample the enhancement features using an encoder to obtain detailed features; input the visual features into a multi-layer bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of the visual features; input the detailed features and multi-scale fused features into a skip connection layer to obtain dynamic fused features; upsample the dynamic fused features using a decoder to obtain an initial semantic segmentation result; and optimize the initial semantic segmentation result to monitor the operating status of the reactor.

[0107] In an optional embodiment, during the process of inputting visual features into a multi-level bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of visual features, the semantic segmentation module 130 is used to: divide the multi-level visual features into multiple groups; wherein the number of groups of visual features is the same as the number of layers of the bidirectional pyramid, and adjacent groups of visual features include at least two overlapping visual features; perform bidirectional fusion according to the level to update the intermediate visual features of each level, until the bidirectional fusion of the last layer of the bidirectional feature pyramid is completed to obtain multi-scale fused features.

[0108] In an optional embodiment, during the process of inputting detailed features and multi-scale fusion features into the skip connection layer to obtain dynamic fusion features, the semantic segmentation module 130 is used to: generate dynamic gating weights based on detailed features, multi-scale fusion features and process parameter features; and generate dynamic fusion features by dynamically balancing detailed features and multi-scale fusion features based on dynamic gating weights.

[0109] In an optional embodiment, during the process of optimizing the initial semantic segmentation result to monitor the operating status of the reactor, the semantic segmentation module 130 is used to: input the initial semantic segmentation result into the conditional random field model and adjust the parameters of the conditional random field model; obtain the output of the conditional random field model in the case of minimizing the energy function, and use the output as the updated semantic segmentation result; and analyze the operating status of the reactor based on the updated semantic segmentation result to monitor the operating status of the reactor.

[0110] Please see Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. An electronic device 200 provided in this application includes: a processor 201 and a memory 202. The memory 202 stores machine-readable instructions executable by the processor 201. When the machine-readable instructions are executed by the processor 201, the method described above is performed.

[0111] Based on the same inventive concept, embodiments of this application also provide a computer program product, which includes a computer program / instructions, and when the computer program / instructions are executed by a processor, implements the steps of any of the methods described in the reactor operating status monitoring method.

[0112] Based on the same inventive concept, embodiments of this application also provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, they perform the steps in any implementation of the above-described reactor operating status monitoring method.

[0113] The computer-readable storage medium can be any medium capable of storing program code, such as Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM).

[0114] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0115] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring the operating status of a reaction vessel, characterized in that, The method includes: Obtain visual characteristics and process parameter characteristics of the reactor operation; Based on the visual features, the process parameter features are spatialized to obtain the process parameter spatialized features; Enhanced features are obtained by spatializing the visual features and process parameters. Dynamic semantic segmentation is performed based on the visual features and enhanced features to monitor the operating status of the reactor. The step of obtaining enhanced features based on the spatialized features of the visual features and process parameters includes: The visual features and the spatialized features of the process parameters are concatenated to obtain the concatenated features; Based on the splicing features, attention weights are generated; The visual features are spatially weighted using the attention weights to obtain the enhanced features; The visual features include the color and / or morphological features of the materials in the reactor, and the morphological features include normal morphology and abnormal morphology; the normal morphology includes material viscosity, wall adhesion and splashing; the abnormal morphology includes the generation of sparks and excessive accumulation of materials in certain areas. The process parameter characteristics include temperature characteristics, acoustic characteristics, and pressure characteristics in the reactor.

2. The method according to claim 1, characterized in that, The dynamic semantic segmentation based on the visual features and enhanced features to monitor the operating status of the reactor includes: The enhanced features are downsampled using an encoder to obtain detailed features; The visual features are input into a multi-level bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of the visual features. The detailed features and multi-scale fused features are input into the skip connection layer to obtain dynamic fused features; The dynamic fusion features are upsampled using a decoder to obtain the initial semantic segmentation result; The initial semantic segmentation results are optimized to monitor the operating status of the reactor.

3. The method according to claim 2, characterized in that, The step of inputting the visual features into a multi-layer bidirectional feature pyramid to obtain multi-scale fused features after bidirectional fusion of the visual features includes: The visual features at multiple levels are divided into multiple groups; wherein the number of groups of visual features is the same as the number of layers of the bidirectional feature pyramid, and adjacent groups of visual features include at least two overlapping visual features. Bidirectional fusion is performed hierarchically to update the intermediate visual features of each level until the bidirectional fusion of the last layer of the bidirectional feature pyramid is completed, thus obtaining the multi-scale fused features.

4. The method according to claim 2, characterized in that, The step of inputting the detailed features and multi-scale fused features into the skip connection layer to obtain dynamic fused features includes: Based on the detailed features, multi-scale fusion features, and process parameter features, dynamic gating weights are generated; Based on the dynamic gating weights, the detailed features and multi-scale fusion features are dynamically balanced to generate the dynamic fusion features.

5. The method according to claim 2, characterized in that, Optimizing the initial semantic segmentation result to monitor the operating status of the reactor includes: The initial semantic segmentation result is input into the conditional random field model, and the parameters of the conditional random field model are adjusted. Obtain the output of the conditional random field model when the energy function is minimized, and use the output as the updated semantic segmentation result; Based on the updated semantic segmentation results, the operating status of the reactor is analyzed to monitor its operation.

6. The method according to claim 1, characterized in that, in, The process parameter features include sensor features used to monitor the reaction process in the reactor.

7. A reactor operation status monitoring system, characterized in that, The reactor operation status monitoring system includes: a feature acquisition module, a feature enhancement module, a semantic segmentation module, and a safety monitoring module; The feature acquisition module is used to acquire the visual features and process parameter features of the reactor operation; The feature enhancement module is used to spatialize the process parameter features based on the visual features to obtain spatialized process parameter features; The semantic segmentation module is used to spatialize features based on the visual features and process parameters to obtain enhanced features; The safety monitoring module is used to perform dynamic semantic segmentation based on the visual features and enhanced features in order to monitor the operating status of the reactor. Specifically, the semantic segmentation module is used for: The visual features and the spatialized features of the process parameters are concatenated to obtain the concatenated features; Based on the splicing features, attention weights are generated; The visual features are spatially weighted using the attention weights to obtain the enhanced features; The visual features include the color and / or morphological features of the materials in the reactor, and the morphological features include normal morphology and abnormal morphology; the normal morphology includes material viscosity, wall adhesion and splashing; the abnormal morphology includes the generation of sparks and excessive accumulation of materials in certain areas. The process parameter characteristics include temperature characteristics, acoustic characteristics, and pressure characteristics in the reactor.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores program instructions, and when the processor executes the program instructions, it performs the steps of the method according to any one of claims 1-6.

9. A computer program product, characterized in that, The computer program product includes a computer program / instruction that, when executed by a processor, implements the steps of the method according to any one of claims 1-6.

Images