Control method and system of ethylene cracking furnace

Abstract

The invention relates to a control method and system for an ethylene cracking furnace, and belongs to the technical field of intelligent adjusting.The control method for the ethylene cracking furnace comprises the steps that an infrared thermodynamic diagram of the ethylene cracking furnace is obtained; performing channel mapping processing on the infrared thermodynamic diagram to obtain a three-channel image; inputting the three-channel image into a classification model, and performing category prediction on the three-channel image through the classification model to generate current category information; inputting the infrared thermodynamic diagram into an adjustment strategy output model in response to the fact that the current category information is abnormal, and outputting an adjustment action through the adjustment strategy output model; the infrared thermodynamic diagram is input into the adjustment strategy output model, and the corresponding adjustment action is automatically generated to adjust the operation parameters of the ethylene cracking furnace, so that the limitation that a worker depends on experience to carry out manual adjustment is avoided, the accuracy and timeliness of adjustment are ensured, automatic generation of the adjustment action of the ethylene cracking furnace is realized, and the production efficiency is improved. Therefore, stable control of the ethylene cracking furnace is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent regulation, and in particular to a control method and system for an ethylene cracking furnace. Background Art

[0002] In the production process of ethylene, the thermal cracking method is one of the main production means. The carburization and coking problems of the furnace tubes of the ethylene cracking furnace in the thermal cracking method have always been the key factors restricting production efficiency and continuity. In the process of solving the coking problem of the furnace tubes, the external temperature of the furnace tubes of the cracking furnace is an important factor leading to coking. Therefore, accurately monitoring and controlling the temperature of the furnace tube surface is crucial for preventing coking and optimizing the production process.

[0003] At present, infrared thermal imaging technology is gradually applied to the temperature monitoring of ethylene cracking furnaces. Through infrared imaging, real-time temperature measurement can be provided, enabling operators to intuitively understand the temperature distribution on the surface of the furnace tubes. However, the infrared thermal imaging technology can only achieve the acquisition and display of temperature data. However, how to adjust the operating state of the ethylene cracking furnace according to the furnace tube temperature depends on the experience of workers, making the control of the ethylene cracking furnace difficult. Improper operation is likely to affect the production quality and efficiency of ethylene and may also pose potential safety hazards. Therefore, how to achieve stable control of the ethylene cracking furnace is an urgent problem to be solved at present. Summary of the Invention

[0004] In order to facilitate the stable control of the ethylene cracking furnace, the present application provides a control method and system for an ethylene cracking furnace.

[0005] In a first aspect, a control method for an ethylene cracking furnace provided by the present application adopts the following technical solution:

[0006] A control method for an ethylene cracking furnace includes:

[0007] Obtain the infrared thermal map of the ethylene cracking furnace;

[0008] Perform channel mapping processing on the infrared thermal map to obtain a three-channel image;

[0009] Input the three-channel image into a classification model, and perform class prediction on the three-channel image through the classification model to generate current class information; wherein, the current class information is used to characterize the current operating state of the ethylene cracking furnace;

[0010] In response to the current class information being abnormal, input the infrared thermal map into an adjustment strategy output model, and output an adjustment action through the adjustment strategy output model; wherein, the adjustment action is used to adjust the operating parameters of the ethylene cracking furnace.

[0011] Optionally, performing channel mapping processing on the infrared thermal image to obtain a three-channel image specifically includes:

[0012] Performing denoising processing on the infrared thermal image to obtain a denoised thermal image;

[0013] Mapping the temperature value of each pixel point in the denoised thermal image according to a preset range, and normalizing the mapping result to obtain the normalized temperature value of each pixel;

[0014] Based on a preset RGB color mapping rule, calculating the color values of the R channel, G channel, and B channel respectively according to the normalized temperature value of each pixel;

[0015] Merging the color values of the R channel, G channel, and B channel of all pixel points in the channel dimension to obtain a three-channel image.

[0016] Optionally, performing denoising processing on the infrared thermal image to obtain a denoised thermal image specifically includes:

[0017] Obtaining a historical thermal image at the previous time point adjacent to the current infrared thermal image in the time dimension;

[0018] Presetting a weighting factor for the current infrared thermal image and the historical thermal image;

[0019] Based on the weighting factor, performing corresponding fusion on each pixel in the current infrared thermal image and the historical infrared thermal image to obtain a denoised thermal image.

[0020] Optionally, the classification model includes an image segmentation layer, a linear mapping layer, a position encoding layer, an encoder layer, and a classification layer;

[0021] The image segmentation layer is used to segment the three-channel image to obtain a plurality of image patches;

[0022] The linear mapping layer receives all the image patches and vectorizes each of the image patches to obtain corresponding mapping vectors;

[0023] The position encoding layer receives the mapping vectors and adds position identifiers to the mapping vectors to obtain identifier vectors corresponding to each image patch; wherein, the position identifiers are used to represent the positions of the image patches;

[0024] The encoder layer receives the identifier vectors, analyzes the identifier vectors, and outputs a class vector and feature vectors corresponding to each image patch;

[0025] The classification layer analyzes the probability distribution of each class to which the three-channel image belongs according to the class vector and the feature vectors, and generates current class information based on the probability distribution.

[0026] Optionally, adding a position identifier to the mapping vector to obtain an identifier vector corresponding to each tile specifically includes:

[0027] Adding the mapping vector corresponding to each tile and the position identifier correspondingly to obtain the identifier vector of each tile.

[0028] Optionally, it further includes a training step of training a preset reinforcement learning network to obtain an adjusted policy output model;

[0029] The training step includes:

[0030] Obtain a training data set, where the training data set includes state information and adjustment actions; the state information includes a temperature observation value and category information; the adjustment action includes a temperature adjustment value for the ethylene cracking furnace;

[0031] Input the current state information into a preset reinforcement learning network, and the reinforcement learning network selects a corresponding adjustment action based on the current state information through a Q function and a preset policy;

[0032] Obtain an infrared thermal image after executing the adjustment action, and calculate the reward value of the adjustment action based on the infrared thermal image after executing the adjustment action;

[0033] Iteratively update the parameters of the Q function and the preset policy according to the current state information and the reward value to obtain an adjusted policy output model.

[0034] Optionally, the preset policy is a greedy policy.

[0035] In a second aspect, the present application provides a control system for an ethylene cracking furnace, adopting the following technical solutions:

[0036] A control system for an ethylene cracking furnace includes:

[0037] A thermal image acquisition module, configured to acquire an infrared thermal image of the ethylene cracking furnace;

[0038] An image preprocessing module, configured to perform channel mapping processing on the infrared thermal image to obtain a three-channel image;

[0039] A state discrimination module, configured to input the three-channel image into a classification model, and perform category prediction on the three-channel image through the classification model to generate current category information; wherein, the current category information is used to represent the current operating state of the ethylene cracking furnace;

[0040] An adjustment policy output module, configured to, in response to the current category information being abnormal, input the infrared thermal image into the adjustment policy output model, and output an adjustment action through the adjustment policy output model; wherein, the adjustment action is used to adjust the operating parameters of the ethylene cracking furnace.

[0041] In a third aspect, the present application provides a computer device, adopting the following technical solution:

[0042] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program of any of the above methods.

[0043] In a fourth aspect, the present application provides a computer-readable storage medium, adopting the following technical solution:

[0044] A computer-readable storage medium includes a computer program stored therein that can be loaded and executed by a processor and is in any of the above methods.

[0045] In summary, the present application includes the following beneficial technical effects:

[0046] 1. By obtaining the infrared thermal image of the ethylene cracking furnace, converting it into a three-channel image through channel mapping processing, and using a classification model for automatic state recognition. When it is found that the ethylene cracking furnace is in an abnormal state, the infrared thermal image is input into the adjustment strategy output model to automatically generate corresponding adjustment actions to adjust the operating parameters of the ethylene cracking furnace, avoiding the limitations of manual adjustment by workers relying on experience, ensuring the accuracy and timeliness of adjustment, realizing the automatic generation of adjustment actions for the ethylene cracking furnace, and thus improving the stable control of the ethylene cracking furnace.

[0047] 2. By obtaining the infrared thermal image after performing the adjustment action and calculating the reward value of the adjustment action for quantitative evaluation of the adjustment effect, the infrared thermal image can intuitively reflect the temperature distribution of the ethylene cracking furnace. By comparing the infrared thermal images before and after performing the adjustment action, the influence of the adjustment action on the temperature distribution can be accurately evaluated. The calculation of the reward value further quantifies the quality of the adjustment effect, providing a basis for optimizing the adjustment strategy. According to the current state information and the reward value, the parameters of the Q function and the preset strategy are iteratively updated. Through continuous iterative learning, the reinforcement learning network can gradually approach the optimal adjustment strategy, making the selection of adjustment actions more accurate and efficient, avoiding the subjectivity and inaccuracy of manual intervention and empirical judgment, and improving the accuracy and efficiency of adjustment. Description of the Drawings

[0048] Figure 1 is a flowchart of the control method for the ethylene cracking furnace in one embodiment of the present application.

[0049] Figure 2 is a flowchart of the method for generating a three-channel image in one embodiment of the present application.

[0050] Figure 3 is a schematic diagram of a three-channel image in one embodiment of the present application.

[0051] Figure 4 It is a schematic structural diagram of a classification model according to one embodiment of the present application.

[0052] Figure 5 It is a flowchart of a training method for an adjustment strategy output model according to one embodiment of the present application.

[0053] Figure 6 It is a block diagram of a control system for an ethylene cracking furnace according to one embodiment of the present application. Detailed implementation manners

[0054] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] An embodiment of the present application discloses a control method for an ethylene cracking furnace. Referring to Figure 1 , a control method for an ethylene cracking furnace includes:

[0056] Step S101: Obtain an infrared thermal image of the ethylene cracking furnace;

[0057] Among them, the infrared thermal image is an image that shows the temperature distribution on the surface of an object through infrared thermal imaging technology. In this embodiment, the infrared thermal image is the original image obtained by an infrared imager, which is a single-channel image composed only of temperature information, that is, the infrared thermal image represents different temperature ranges through different levels of gray scale (from black to white).

[0058] Step S102: Perform channel mapping processing on the infrared thermal image to obtain a three-channel image;

[0059] Step S103: Input the three-channel image into the classification model, and perform category prediction on the three-channel image through the classification model to generate current category information; wherein, the current category information is used to characterize the current operating state of the ethylene cracking furnace;

[0060] Among them, the classification model is a Vision Transformer (ViT) model. Vision Transformer is an emerging image classification model that uses a structure similar to Transformer in natural language processing to process images.

[0061] Among them, the category information includes normal, coking, temperature anomaly, local overheating, carburization, etc.; when coking, temperature anomaly, local overheating and carburization occur, the category information is all abnormal.

[0062] Step S104: In response to the current category information being abnormal, input the infrared thermal map into the adjustment strategy output model, and output an adjustment action through the adjustment strategy output model; wherein, the adjustment action is used to adjust the operating parameters of the ethylene cracking furnace.

[0063] It should be understood that the ethylene cracking furnace adjusts the operating state of the furnace chamber through burners inside the furnace. The burners are used to transfer the heat generated by fuel combustion to the furnace chamber to maintain the required high-temperature environment inside the furnace. The burners precisely control the supply of fuel and the mixing ratio of air to achieve an efficient combustion process, and precisely control the furnace chamber temperature by adjusting the fuel flow rate and air flow rate of the burners. That is, when it is necessary to increase the furnace chamber temperature of the ethylene cracking furnace, the fuel flow rate can be increased or the mixing ratio of air and fuel can be optimized to promote more complete combustion and generate more heat. On the contrary, when it is necessary to lower the temperature, the fuel flow rate can be reduced or the operation mode of the burners can be adjusted to reduce heat generation.

[0064] It should also be understood that if the current category information is normal, steps S101 - S104 are cyclically executed to achieve real-time monitoring of the ethylene cracking furnace.

[0065] In the above embodiment, by obtaining the infrared thermal map of the ethylene cracking furnace, converting it into a three-channel image through channel mapping processing, and using a classification model for automatic state recognition. When it is found that the ethylene cracking furnace is in an abnormal state, the infrared thermal map is input into the adjustment strategy output model to automatically generate corresponding adjustment actions to adjust the operating parameters of the ethylene cracking furnace, avoiding the limitations of manual adjustment by workers relying on experience, ensuring the accuracy and timeliness of adjustment, realizing the automatic generation of adjustment actions for the ethylene cracking furnace, and thus improving the stable control of the ethylene cracking furnace.

[0066] Refer to Figure 2 、 3 , as an embodiment of step S102, step S102 specifically includes:

[0067] Step S1021: Denoise the infrared thermal map to obtain a denoised thermal map;

[0068] Step S1022: Map the temperature value of each pixel point in the denoised thermal map according to a preset range and normalize the mapping result to obtain the normalized temperature value of each pixel;

[0069] Among them, since the normal temperature range of the ethylene cracking furnace is 800°C - 860°C, and exceeding 1100°C is severe coking. Based on this, in this embodiment, the preset range is set to 600°C - 1400.

[0070] Specifically, the temperature obtained by mapping each pixel point according to the preset range

[0071]

[0072] Then, the mapped temperature T mapped is normalized, and the normalized temperature value After normalization, the range of the normalized temperature value always remains between 0 and 1.

[0073] Step S1023: Based on the preset RGB color mapping rule, calculate the color values of the R channel, G channel, and B channel respectively according to the normalized temperature value of each pixel;

[0074] Specifically, the preset RGB color mapping rule is that low temperature (600 °C) is mapped to blue: RGB is (0, 0, 255). High temperature (1400 °C) is mapped to red, RGB is (255, 0, 0). Therefore, the RGB color value of each pixel point can be expressed as C p (R p , G p , B p ), where R p = 255 × T norm ; G p = 0; B p = 255 × (1 - T morm ). R p is the color value of the R channel, representing the mapping of the normalized temperature to the red channel; G p is the color value of the G channel, representing the mapping of the normalized temperature to the green channel, and B p is the color value of the B channel, representing the mapping of the normalized temperature to the blue channel.

[0075] Step S1024: Combine the color values of the R channel, G channel, and B channel of all pixel points in the channel dimension to obtain a three-channel image.

[0076] Combined with Figure 3 , Figure 3 (a) is a single-channel infrared thermal image, Figure 3 (b) is a three-channel image (grayscale processed).

[0077] In the above embodiments, by denoising the infrared thermal image, a denoised thermal image is obtained to improve the quality of the image, making the temperature distribution information clearer and more accurate. The temperature value of each pixel in the denoised thermal image is mapped according to a preset range, and the mapping result is normalized to map different temperature values to a unified numerical range, so that the temperature information can be reasonably represented in the subsequent RGB color mapping process. Based on the preset RGB color mapping rule, according to the normalized temperature value of each pixel, the color values of the R channel, G channel, and B channel are calculated respectively, and the temperature information is converted into color information, so that the original infrared thermal image with only single temperature information is transformed into a three-channel image containing rich color information. And the RGB color mapping rule can be set according to actual needs to perform conversion according to the required temperature range, obtaining a three-channel image with a specific color representation, which is convenient for intuitively observing and identifying the temperature distribution situation.

[0078] As an implementation manner of step S1021, step S1021 specifically includes:

[0079] Step S10211: Obtain a historical thermal image at the previous time point adjacent to the current infrared thermal image in the time dimension;

[0080] It should be understood that the infrared thermal image is obtained from the continuous video captured by the infrared camera. Each frame of the continuous video stream is a static picture in the video or image sequence. In the video captured by the infrared camera, the historical thermal image at the previous time point adjacent to the current infrared thermal image in the time dimension is the previous frame image or the image at the preset previous time point.

[0081] Step S10212: Preset the weighting factors of the current infrared thermal image and the historical thermal image;

[0082] Among them, the weighting factor of the infrared thermal image and the weighting factor of the historical thermal image are complementary relationships, that is, the sum of the weighting factor of the infrared thermal image and the weighting factor of the historical thermal image is equal to one.

[0083] Step S10213: Based on the weighting factors, perform corresponding fusion on each pixel in the current infrared thermal image and the historical infrared thermal image to obtain a denoised thermal image.

[0084] Specifically, the temperature value of the pixel point in the denoised thermal image is T i，j，t = α·T i，j，t-1 +(1 - α)·T i，j，current ; where α is the weighting factor, and the range is between 0 and 1. The weighting factor is used to determine the relative importance of the current measurement value and the historical value in the temperature estimation. T i,j,t represents the temperature estimation value of the pixel point located in the i-th row and j-th column of the denoised thermal image at time point t. Ti,j,current is the temperature value measured at the pixel point in the i-th row and j-th column of the current infrared thermal image. T i,j,t-1 is the temperature value measured at the pixel point in the i-th row and j-th column at the previous time point t - 1 of the pixel point in the historical infrared thermal image.

[0085] In the above embodiment, exponential weighted averaging is performed on each pixel point in the time dimension, making full use of the information in the time series, reducing the influence of various factors such as environmental noise and equipment errors on the temperature data captured by the infrared camera. By weighted averaging, the noise is smoothed and its impact on the temperature reading is reduced. At the same time, since the weighting factor is preset according to actual needs, it can be flexibly adjusted to balance the weights of the current frame and the historical frame, thereby effectively retaining the key information of the temperature change.

[0086] Referring to Figure 4 , as an embodiment of the classification model, the classification model includes an image segmentation layer, a linear mapping layer, a position encoding layer, an encoder layer, and a classification layer;

[0087] The image segmentation layer is used to segment the three-channel image to obtain a number of tiles;

[0088] Specifically, the three-dimensional channel image is segmented into N tiles P i , where i = 1, 2,...., N. The size of the tiles can be set according to the actual situation, for example, set to 32×32, 16×16, or 14×14.

[0089] The linear mapping layer receives all the tiles and vectorizes each of the tiles to obtain corresponding mapping vectors;

[0090] Specifically, the tile P i is converted into a D-dimensional vector X i = LinearProjection(P i ), that is, each tile uniquely corresponds to a generated mapping vector.

[0091] The position encoding layer receives the mapping vectors and adds position identifiers to the mapping vectors to obtain the identifier vectors corresponding to each tile; wherein, the position identifiers are used to represent the positions of the tiles;

[0092] Among them, adding position identifiers to the mapping vectors to obtain the identifier vectors corresponding to each tile specifically includes: adding the mapping vectors corresponding to each tile and the position identifiers correspondingly to obtain the identifier vectors of each tile.

[0093] Specifically, each vector X iAdd position identifier E pos,i The obtained identification vector Z0 = [X1 + E pos,1 ; X2 + E pos,2 ;...; X N + E pos,N .

[0094] The encoder layer receives the identification vector and analyzes the identification vector to output a category vector and a feature vector corresponding to each patch;

[0095] The classification layer analyzes the probability distribution of each category to which the three-channel image belongs based on the category vector and the feature vector, and generates current category information based on the probability distribution.

[0096] Specifically, the category vector contains the information most relevant to the overall classification of the three-channel image and synthesizes the information of all important features in the three-channel image. Then, through the linear classification layer (Linear): map the category vector to a score vector corresponding to the number of categories. Then, use the softmax function to convert the score vector output by the linear classification layer into a probability distribution, that is, convert each score into a positive number and ensure that the sum of the probabilities of all categories is 1. In this way, the output of each category can be interpreted as the probability that the input image belongs to that category, so as to obtain the probability distribution of each category, and then find the index of the maximum value in the probability distribution (the category predicted by the model) as the current category information.

[0097] Refer to Figure 5 As a further implementation of the control method of the ethylene cracking furnace, the control method of the ethylene cracking furnace further includes a training step of training a preset reinforcement learning network to obtain an adjustment strategy output model;

[0098] The training steps include:

[0099] Step S201: Obtain a training data set, where the training data set includes state information and adjustment actions; the state information includes temperature observation values and category information; the adjustment actions include temperature adjustment values for the ethylene cracking furnace;

[0100] It should be understood that after generating the temperature adjustment value, the temperature is adjusted by adjusting the fuel flow rate and air flow rate of the combustion furnace in the ethylene cracking furnace.

[0101] Step S202: Input the current state information into a preset reinforcement learning network, and the reinforcement learning network selects corresponding adjustment actions based on the current state information through the Q function and a preset policy;

[0102] Among them, the preset policy is the greedy policy. In Q-learning, the greedy policy is used to select adjustment actions according to the current Q-values. Specifically, the greedy policy finds a balance between exploitation and exploration. Exploitation means, based on the current Q-value table or Q-network, selecting the action with the highest Q-value. Exploration means selecting non-greedy actions with a certain probability (controlled by the probability ε), that is, actions that are not the ones with the highest current Q-values, which helps the agent discover potentially better strategies, avoid getting stuck in local optima, and handle possible Q-value estimation errors.

[0103] Step S203: Obtain the infrared thermal image after performing the adjustment action, and calculate the reward value of the adjustment action based on the infrared thermal image after performing the adjustment action;

[0104] Step S204: Iteratively update the parameters of the Q-function and the preset policy according to the current state information and the reward value to obtain an adjustment policy output model.

[0105] Among them, the parameter of the preset policy is the probability ε. For example, a relatively high ε value can be set at the beginning of training to promote extensive exploration, and then the ε value is gradually decreased as training progresses, so as to finally converge to a stable policy.

[0106] Among them, the iterative update of the Q-function follows where α is the learning rate, γ is the discount factor, s t and a t are the current state information and the selected adjustment action respectively, s t+1 and a t+1 are the next state information and the possible selected adjustment actions respectively.

[0107] In the above embodiment, by obtaining the infrared thermal image after performing the adjustment action and calculating the reward value of the adjustment action for quantitative evaluation of the adjustment effect, the infrared thermal image can intuitively reflect the temperature distribution of the ethylene cracking furnace. By comparing the infrared thermal images before and after performing the adjustment action, the impact of the adjustment action on the temperature distribution can be accurately evaluated. The calculation of the reward value further quantifies the quality of the adjustment effect, providing a basis for optimizing the adjustment policy. Iteratively updating the parameters of the Q-function and the preset policy according to the current state information and the reward value, through continuous iterative learning, the reinforcement learning network can gradually approach the optimal adjustment policy, making the selection of adjustment actions more accurate and efficient, avoiding the subjectivity and inaccuracy of manual intervention and empirical judgment, and improving the accuracy and efficiency of the adjustment.

[0108] As a further implementation of the control method for an ethylene cracking furnace, the control method for the ethylene cracking furnace further includes a step of periodically training the classification model and the adjustment strategy output model. Specifically, by recording the temperature changes before and after burner adjustment, the temperature data of each measuring point in the furnace before and after each adjustment action, as well as indicators such as production efficiency and product quality, historical data is collected, and the classification model and the adjustment strategy output model are periodically retrained and optimized to adapt to the changes in the operating state of the ethylene cracking furnace and operation experience. Through continuous learning and adaptation, the classification model can more accurately identify the abnormal state of the cracking furnace. By analyzing the historical adjustment effects, the adjustment strategy output model continuously optimizes the adjustment strategy to find a more effective method to adjust the burner, so as to restore or maintain the normal temperature of the cracking furnace, improve production efficiency and product quality, and reduce energy consumption at the same time.

[0109] In addition, the embodiments of the present application disclose a control system for an ethylene cracking furnace. The control system for the ethylene cracking furnace can be applied to a computer device, and is a schematic diagram of the architecture of the computer device provided by the embodiments of the present invention for implementing the above method. In this embodiment, the computer device may include a control system for the ethylene cracking furnace, a machine-readable storage medium, and a processor.

[0110] In this embodiment, the machine-readable storage medium and the processor may be located in the computer device and are separately arranged. The machine-readable storage medium may also be independent of the computer device and accessible by the processor. The control system for the ethylene cracking furnace may include a plurality of functional modules stored in the machine-readable storage medium, such as each software functional module included in the control system for the ethylene cracking furnace. When the processor executes the computer program corresponding to the software functional module in the control system for the ethylene cracking furnace, the control system for the ethylene cracking furnace provided by the foregoing method embodiment is implemented.

[0111] In this embodiment, the computer device may include one or more processors. The processor may process information and / or data related to service requests to execute one or more functions described in the present invention. In some embodiments, the processor may include one or more processing engines (for example, a single-core processor or a multi-core processor). Just for example, the processor may include one or more hardware processors, such as a central processing unit CPU, an application specific integrated circuit ASIC, an application specific instruction set processor ASIP, a graphics processing unit GPU, a physics processing unit PPU, a digital signal processor DSP, a field programmable gate array FPGA, a programmable logic device PLD, a controller, a microcontroller unit, a reduced instruction set computer RISC, a microprocessor, etc., or a combination thereof or any combination similar thereto.

[0112] A machine-readable storage medium can store data and / or instructions. In some embodiments, the machine-readable storage medium can store acquired data or information. In some embodiments, the machine-readable storage medium can store data and / or instructions for execution or use by the computer device, and the computer device can implement the exemplary methods described in this application by executing or using the data and / or instructions. In some embodiments, the machine-readable storage medium can include a mass storage device, a removable storage device, a volatile read-write memory, a read-only memory (ROM), etc., or any combination of the foregoing or the like. Exemplary mass storage devices can include magnetic disks, optical disks, solid-state disks, etc. Exemplary removable storage devices can include flash drives, floppy disks, optical disks, memory cards, compact disks, magnetic tapes, etc. Exemplary volatile read-write memories can include random access memories (RAMs). Exemplary random access memories can include dynamic RAM, double data rate synchronous dynamic RAM, static RAM, thyristor RAM, and zero-capacitance RAM, etc. Exemplary ROMs can include masked ROM, programmable ROM, erasable programmable ROM, electrically erasable programmable ROM, compact disk ROM, and digital versatile disk ROM, etc.

[0113] Among them, the control system of the ethylene cracking furnace included in the computer device may include one or more software function modules. The software function modules may be programs and instructions stored in the machine-readable storage medium, and when these software function modules are executed by the corresponding processor, they are used to implement the above methods. For example, when executed by the processor of the unmanned aerial vehicle, they are used to implement the method steps executed by the above unmanned aerial vehicle, or when executed by the computer device, they are used to implement the method steps executed by the above computer device.

[0114] Referring in detail to Figure 6 , an embodiment of the present application discloses a control system for an ethylene cracking furnace. A control system for an ethylene cracking furnace includes:

[0115] A thermal map acquisition module, configured to acquire an infrared thermal map of the ethylene cracking furnace;

[0116] An image preprocessing module, configured to perform channel mapping processing on the infrared thermal map to obtain a three-channel image;

[0117] A state discrimination module, configured to input the three-channel image into a classification model, and perform class prediction on the three-channel image through the classification model to generate current class information; wherein, the current class information is used to characterize the current operating state of the ethylene cracking furnace;

[0118] The adjustment strategy output module is configured to, in response to the current category information being abnormal, input the infrared thermal map into the adjustment strategy output model, and output an adjustment action through the adjustment strategy output model; wherein, the adjustment action is used to adjust the operating parameters of the ethylene cracking furnace.

[0119] The control system of an ethylene cracking furnace provided by this application can implement the above-mentioned control method of an ethylene cracking furnace, and the specific working process of the control system of an ethylene cracking furnace can refer to the corresponding process in the above method embodiment.

[0120] It should be noted that in the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0121] The present invention also discloses a computer-readable storage medium, which includes a computer program stored therein that can be loaded and executed by a processor as in any of the above methods.

[0122] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some communication interfaces. The indirect coupling or communication connection of the devices or units can be in an electrical, mechanical or other forms.

[0123] In addition, in each embodiment of this application, the various functional modules can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0124] The above are all the preferred embodiments of this application. Without restricting the protection scope of this application based on this, any feature disclosed in this specification (including the abstract and drawings), unless specifically described, can be replaced by other equivalent or similar-purpose alternative features. That is, unless specifically described, each feature is only an example in a series of equivalent or similar features.

Claims

1. A control method for an ethylene cracking furnace, characterized in that Including: Obtain the infrared thermal image of the ethylene cracking furnace; Perform channel mapping processing on the infrared thermal image to obtain a three-channel image; Input the three-channel image into the classification model, and perform class prediction on the three-channel image through the classification model to generate the current class information; wherein, the current class information is used to characterize the current operating state of the ethylene cracking furnace; In response to the current class information being abnormal, input the infrared thermal image into the adjustment strategy output model, and output an adjustment action through the adjustment strategy output model; wherein, the adjustment action is used to adjust the operating parameters of the ethylene cracking furnace.

2. The method according to claim 1, wherein The performing channel mapping processing on the infrared thermal image to obtain a three-channel image specifically includes: Perform denoising processing on the infrared thermal image to obtain a denoised thermal image; Map the temperature value of each pixel point in the denoised thermal image according to a preset range, and normalize the mapping result to obtain the normalized temperature value of each pixel; Based on the preset RGB color mapping rule, calculate the color values of the R channel, G channel, and B channel respectively according to the normalized temperature value of each pixel; Merge the color values of the R channel, G channel, and B channel of all pixel points in the channel dimension to obtain a three-channel image.

3. The method according to claim 2, wherein The performing denoising processing on the infrared thermal image to obtain a denoised thermal image specifically includes: Obtain the historical thermal image at the previous time point adjacent to the current infrared thermal image in the time dimension; Preset the weighting factor of the current infrared thermal image and the historical thermal image in advance; Based on the weighting factor, perform corresponding fusion on each pixel in the current infrared thermal image and the historical infrared thermal image to obtain a denoised thermal image.

4. The method according to claim 1, characterized in that, The classification model includes an image segmentation layer, a linear mapping layer, a position encoding layer, an encoder layer, and a classification layer; The image segmentation layer is used to segment the three-channel image to obtain a plurality of patches; The linear mapping layer receives all the patches and vectorizes each of the patches respectively to obtain corresponding mapping vectors; The position encoding layer receives the mapping vectors and adds position identifiers to the mapping vectors to obtain identifier vectors corresponding to each patch; wherein, the position identifier is used to characterize the position of the patch; The encoder layer receives the identifier vectors, analyzes the identifier vectors, and outputs a class vector and a feature vector corresponding to each patch; The classification layer analyzes the probability distribution of each class to which the three-channel image belongs according to the class vector and the feature vector, and generates the current class information based on the probability distribution.

5. The method according to claim 4, characterized in that, The adding position identifiers to the mapping vectors to obtain identifier vectors corresponding to each patch specifically includes: Add the mapping vectors corresponding to each patch and the position identifiers correspondingly to obtain the identifier vectors of each patch.

6. The method according to claim 1, characterized in that It also includes the training step of training a preset reinforcement learning network to obtain the adjustment strategy output model; The training step includes: Obtain a training data set, where the training data set includes state information and adjustment actions; the state information includes temperature observation values and class information; the adjustment actions include temperature adjustment values for the ethylene cracking furnace. Input the current state information into a preset reinforcement learning network, and the reinforcement learning network selects a corresponding adjustment action based on the current state information through a Q function and a preset policy; Obtain the infrared thermal image after performing the adjustment action, and calculate the reward value of the adjustment action based on the infrared thermal image after performing the adjustment action; Iteratively update the parameters of the Q function and the preset policy according to the current state information and the reward value to obtain an adjustment policy output model.

7. The method according to claim 6, characterized in that The preset policy is a greedy policy.

8. A control system for an ethylene cracking furnace, characterized in that, It includes: A thermal image acquisition module for acquiring the infrared thermal image of an ethylene cracking furnace; An image preprocessing module for performing channel mapping processing on the infrared thermal image to obtain a three-channel image; A state discrimination module for inputting the three-channel image into a classification model, and performing class prediction on the three-channel image through the classification model to generate current class information; wherein, the current class information is used to characterize the current operating state of the ethylene cracking furnace; An adjustment policy output module for, in response to the current class information being abnormal, inputting the infrared thermal image into the adjustment policy output model, and outputting an adjustment action through the adjustment policy output model; wherein, the adjustment action is used to adjust the operating parameters of the ethylene cracking furnace.

9. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program of any one of the methods in claims 1-7.

10. A computer-readable storage medium, characterized in that, It includes a computer program stored that can be loaded and executed by a processor for any one of the methods in claims 1-7.

Images