Method, device and equipment for monitoring unstable state of flame combustion and medium

Abstract

The invention provides a flame combustion unstable state monitoring method, device and equipment and a medium, and belongs to the technical field of combustion monitoring. The method comprises the steps that flame characteristics of the petrochemical heating furnace in the actual combustion process are obtained, and the characteristics with the difference value between the flame characteristics and the corresponding preset flame characteristics larger than a preset threshold value are determined as target flame characteristics; classifying the target flame features by using a plurality of pre-trained different types of neural network models to obtain a plurality of flame combustion unstable states; voting is conducted on the multiple flame combustion unstable states, and one flame combustion unstable state is selected from the multiple flame combustion unstable states to serve as the target flame combustion unstable state. According to the invention, the accuracy of identifying the unstable state of flame combustion can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of combustion monitoring, and in particular to a method for monitoring unstable flame combustion states, a device for monitoring unstable flame combustion states, an electronic device, a readable storage medium and a petrochemical heating furnace. Background Art

[0002] Petrochemical heating furnace is an important equipment used in industrial production process, mainly used to heat chemical raw materials or fuel. In the use of heating furnace, flame combustion instability is a common problem.

[0003] At present, there are various methods for monitoring flame combustion status on the market, such as acoustic signal monitoring, thermal radiation monitoring, optical flame monitoring and flame ion monitoring, but the above methods all have certain limitations (interference by ambient light, insufficient sensor sensitivity, and failure can only be replaced when the equipment is shut down). Therefore, existing scholars have proposed an algorithm-based flame state recognition method.

[0004] However, flame image data is difficult to interpret, and the unexplainable nature of its deep features makes feature analysis even more difficult. In addition, the potential inducing factors of combustion instability are numerous and difficult to characterize, making it difficult to accurately diagnose the combustion state, which leads to the inability to accurately monitor the unstable state of flame combustion. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method, device, equipment and medium for monitoring the unstable state of flame combustion, so as to solve the problem that the existing flame combustion state monitoring technology cannot accurately monitor the flame combustion state.

[0006] In order to achieve the above object, an embodiment of the present invention provides a method for monitoring an unstable state of flame combustion, comprising:

[0007] Obtaining the flame characteristics of the petrochemical heating furnace during the actual combustion process, and determining the characteristics of the flame characteristics whose difference with the corresponding preset flame characteristics is greater than a preset threshold as the target flame characteristics;

[0008] Using multiple different types of pre-trained neural network models to classify target flame features, a variety of flame combustion instability states are obtained;

[0009] Voting is performed on the multiple flame combustion instability states to select one flame combustion instability state from the multiple flame combustion instability states as the target flame combustion instability state.

[0010] Optionally, the flame characteristics include: characteristic parameters of flame morphology and characteristic parameters of flame structure;

[0011] The characteristic parameters of the flame morphology include: flame area, flame brightness, flame height, flame pulsation frequency and flame inclination angle;

[0012] The characteristic parameters of the flame structure include: the inner area of ​​the flame, the middle area of ​​the flame and the outer area of ​​the flame;

[0013] The preset flame characteristics include: preset flame area, preset flame brightness, preset flame height, preset flame pulsation frequency, preset flame inclination angle, preset flame inner layer area, preset flame middle layer area and preset flame outer layer area;

[0014] The preset thresholds include: a preset flame brightness difference threshold, a preset flame height difference threshold, a preset flame pulsation frequency difference threshold, a preset flame inclination difference threshold, a preset flame inner layer area difference threshold, a preset flame middle layer area difference threshold and a preset flame outer layer area difference threshold;

[0015] The difference between the flame characteristic and the corresponding preset flame characteristic is greater than a preset threshold, including:

[0016] The difference between the flame brightness and the preset flame brightness is greater than a preset flame brightness difference threshold;

[0017] The difference between the flame height and the preset flame height is greater than a preset flame height difference threshold;

[0018] The difference between the flame pulsation frequency and the preset flame pulsation frequency is greater than the preset flame pulsation frequency difference threshold;

[0019] The difference between the flame inclination angle and the preset flame inclination angle is greater than the preset flame inclination angle difference threshold;

[0020] The difference between the inner layer area of ​​the flame and the preset inner layer area of ​​the flame is greater than the preset inner layer area difference threshold of the flame;

[0021] The difference between the flame middle layer area and the preset flame middle layer area is greater than the preset flame middle layer area difference threshold;

[0022] The difference between the flame outer layer area and a preset flame outer layer area is greater than a preset flame outer layer area difference threshold.

[0023] Optionally, the characteristic parameters of the flame morphology are obtained by:

[0024] Acquire multiple frames of images of the actual combustion process in a petrochemical heating furnace;

[0025] For each frame of image, different image recognition algorithms are used to respectively recognize the frame of image, and the flame morphology of each frame of image under each image recognition algorithm is obtained, and the flame morphology under each image recognition algorithm is determined based on the flame morphology of each frame of image under each image recognition algorithm;

[0026] Based on the characteristic parameters of the flame morphology under each image recognition algorithm, the flame morphology is corrected to obtain the corrected flame morphology;

[0027] Based on the corrected flame morphology, characteristic parameters of the flame morphology are obtained.

[0028] Optionally, the different image recognition algorithms include: any two or three of the maximum inter-class method difference method, Boolean variable and Yolov5 algorithm.

[0029] Optionally, the characteristic parameters of the flame structure are obtained by:

[0030] Acquire multiple frames of images of the actual combustion process in a petrochemical heating furnace;

[0031] Extract the RGB value of each pixel on each frame of the image;

[0032] Determine the pixel points on each frame of image that are greater than a first preset RGB value as the inner flame region in each frame of image, determine the pixel points on each frame of image that are greater than a second preset RGB value as the middle flame region in each frame of image, and determine the pixel points on each frame of image that are greater than a third preset RGB value as the outer flame region in each frame of image;

[0033] The inner flame region, the middle flame region and the outer flame region in each frame of the image are superimposed, and the probability of each pixel point appearing in the inner flame region, the middle flame region and the outer flame region formed after superposition is determined;

[0034] The pixel points whose occurrence probability is greater than a first preset probability threshold are determined as the inner region of the flame, the pixel points whose occurrence probability is greater than a second preset probability threshold are determined as the middle region of the flame, and the pixel points whose occurrence probability is greater than the first preset probability threshold are determined as the outer region of the flame;

[0035] Based on the flame inner region, flame middle region and flame outer region, the flame inner area, flame middle area and flame outer area are obtained.

[0036] Optionally, the multiple flame combustion unstable states include: a first flame combustion unstable state, a second flame combustion unstable state and a third flame combustion unstable state;

[0037] The target flame characteristics are classified by using a plurality of different types of pre-trained neural network models to obtain a plurality of flame combustion instability states, including:

[0038] Using a pre-trained convolutional neural network model to classify characteristic parameters of the target flame morphology, and obtain the first flame combustion unstable state;

[0039] Using a pre-trained recurrent neural network model to classify characteristic parameters of the target flame morphology, and obtain the second flame combustion unstable state;

[0040] The characteristic parameters of the target flame morphology are classified using a pre-trained self-organizing neural network model to obtain the third flame combustion unstable state.

[0041] Optionally, determining one of the multiple flame combustion unstable states as a target flame combustion unstable state includes:

[0042] Voting on the multiple flame combustion unstable states to obtain the number of votes corresponding to each of the multiple flame combustion unstable states;

[0043] The flame combustion unstable state having the largest number of votes among the multiple flame combustion unstable states, and having more than half of the votes, is determined as the flame combustion unstable state.

[0044] Optionally, determining one of the multiple flame combustion unstable states as a target flame combustion unstable state includes:

[0045] Voting on the multiple flame combustion instability states using a relative majority voting method to obtain a voting value corresponding to each of the multiple flame combustion instability states;

[0046] The flame combustion unstable state having the largest number of votes among the plurality of flame combustion unstable states is determined as the target plurality of flame combustion unstable states.

[0047] Optionally, determining one of the multiple flame combustion unstable states as a target flame combustion unstable state includes:

[0048] Voting for each of the plurality of flame combustion instability states using a weighted voting method to obtain a plurality of voting values;

[0049] Performing weighted summation on the multiple voting values ​​to obtain multiple voting scores;

[0050] The flame combustion instability state having the highest voting score among the multiple voting scores is determined as the target flame combustion instability state.

[0051] Optionally, the method further comprises:

[0052] Based on the target flame combustion unstable state, alarm information corresponding to the target flame combustion unstable state is generated.

[0053] In a second aspect of the embodiments of the present invention, a monitoring device for an unstable state of flame combustion is provided, comprising:

[0054] A feature acquisition module is used to acquire the flame features of the petrochemical heating furnace during the actual combustion process, and determine the features whose difference between the flame features and the corresponding preset flame features is greater than a preset threshold as target flame features;

[0055] A state recognition module is used to analyze the target flame characteristics using multiple different types of pre-trained neural network models to obtain multiple flame combustion instability states;

[0056] The target determination module is used to vote on multiple flame combustion instability states to select one flame combustion instability state from the multiple flame combustion instability states as a target flame combustion instability state.

[0057] In a third aspect of the embodiments of the present invention, an electronic device is provided, comprising: a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the machine-readable instructions are executed by the processor, the above-mentioned method for monitoring unstable flame combustion state is executed.

[0058] In a fourth aspect of the embodiments of the present invention, a computer-readable storage medium is provided, storing computer instructions. When the computer instructions are executed on a computer, the computer executes the above-mentioned method for monitoring unstable flame combustion state.

[0059] In a fifth aspect of the embodiment of the present invention, a petrochemical heating furnace is provided, comprising a boiler furnace and the above-mentioned monitoring device for the unstable state of flame combustion.

[0060] In an embodiment of the present invention, the flame characteristics of a petrochemical heating furnace in an actual combustion process are obtained, and the characteristics whose difference between the flame characteristics and the corresponding preset flame characteristics is greater than a preset threshold are determined as target flame characteristics; the target flame characteristics are classified using a plurality of different types of pre-trained neural network models to obtain a plurality of unstable flame combustion states; the plurality of unstable flame combustion states are voted on to select one unstable flame combustion state from the plurality of unstable flame combustion states as the target unstable flame combustion state. The present invention provides effective data input for subsequent models by screening out target flame characteristics that can accurately characterize the changes in flame combustion states from the flame characteristics. In addition, by using different types of models for ensemble learning, the advantages and disadvantages of different types of models can be combined, thereby improving the accuracy of identifying unstable flame combustion states.

[0061] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0063] Figure 1 It is a flow chart of a method for monitoring unstable state of flame combustion provided by an embodiment of the present invention;

[0064] Figure 2 is a schematic diagram of a flame morphology provided by an embodiment of the present invention;

[0065] Figure 3 is a schematic diagram of a flame structure provided by an embodiment of the present invention;

[0066] Figure 4 is a schematic diagram of the structure of a convolutional neural network provided by an embodiment of the present invention;

[0067] Figure 5 is a schematic diagram of the structure of a bidirectional long short-term memory network provided by an embodiment of the present invention;

[0068] Figure 6 is a schematic diagram of the structure of a self-organizing neural network provided by an embodiment of the present invention;

[0069] Figure 7 is a schematic diagram of voting on multiple different types of neural network models provided by an embodiment of the present invention;

[0070] Figure 8 It is a schematic diagram of the architecture of a monitoring device for unstable flame combustion provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0072] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0073] In the description of the embodiments of the present invention, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present invention, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0074] Embodiment 1

[0075] Figure 1 1 is a flow chart of a method for monitoring an unstable state of flame combustion provided by an embodiment of the present invention, the method comprising the following steps:

[0076] S100, obtaining flame characteristics of a petrochemical heating furnace in an actual combustion process, and determining a characteristic whose difference between a flame characteristic and a corresponding preset flame characteristic is greater than a preset threshold as a target flame characteristic;

[0077] Petrochemical heating furnace is a kind of equipment specially used for heating product materials and workpieces. Different from drying equipment, the main purpose of heating furnace is not to heat general materials, but to heat metal products at high temperature without or with little oxidation so that they can be further processed.

[0078] Flame characteristics refer to the flame characteristics of the nozzle in the petrochemical heating furnace during combustion, including: characteristic parameters of flame morphology and characteristic parameters of flame structure.

[0079] See also Figure 2 , Figure 2 FIG. 1 is a schematic diagram of flame morphology. The characteristic parameters of flame morphology include flame area, flame brightness, flame height, flame pulsation frequency and flame inclination. In one embodiment, the characteristic parameters of flame morphology can be obtained by the following method:

[0080] Step 1: Acquire multiple frames of images during the actual combustion process in a petrochemical heating furnace;

[0081] It should be noted that a plurality of furnace nozzles are provided in the heating furnace, and the fuel sprayed from the furnace nozzles burns to form flames, that is, the video information of the actual combustion process in the heating furnace includes the video information of the plurality of furnace nozzles in the actual combustion process.

[0082] It is understandable that the number of multi-frame images of the actual combustion process in the heating furnace is determined by the frame rate and duration of the video information of the actual combustion process in the heating furnace, and the frame rate and duration of the video information of the actual combustion process in the heating furnace are determined by the camera, and the embodiment of the present invention does not specifically limit this. For example: the video frame rate of the camera is 50fps, and it takes a total of 10S to shoot, then the number of multi-frame images of the actual combustion process in the heating furnace is 50fps*10S=500.

[0083] The video information of the actual combustion process in the heating furnace is obtained by shooting with a camera.

[0084] Step 2: for each frame of image, use different image recognition algorithms to recognize the frame of image respectively, obtain the flame morphology of each frame of image under each image recognition algorithm, and determine the flame morphology under each image recognition algorithm based on the flame morphology of each frame of image under each image recognition algorithm;

[0085] The image recognition algorithm includes: any two or three of the maximum inter-class method difference method, Boolean variable and Yolo v5 algorithm. Of course, it can also be other image recognition algorithms, which are not specifically limited in the embodiment of the present invention.

[0086] Step 3: Based on the characteristic parameters of the flame morphology under each image recognition algorithm, the flame morphology is corrected to obtain the corrected flame morphology;

[0087] In order to avoid errors in the image recognition algorithm itself, or errors in the image recognition algorithm caused by the complex combustion environment in the heating furnace, the differences between the characteristic parameters of the flame morphology under different image recognition algorithms are compared. If the differences between different image recognition algorithms are very large, it means that there is an error in one of the algorithms. At this time, the flame morphology needs to be corrected.

[0088] Step 4: Based on the corrected flame morphology, characteristic parameters of the flame morphology are obtained.

[0089] Flame height refers to the vertical distance between the lowest and highest points of the flame shape.

[0090] The flame width refers to the lateral distance between the leftmost and rightmost sides of the flame shape.

[0091] Flame area refers to the size of the flame shape, which can be calculated in the following way: flame area = 5 / 6*flame width.

[0092] Flame pulsation frequency refers to the phenomenon of periodic changes in flame length. Flame pulsation frequency is the reciprocal of the number of times the maximum / minimum height of the same flame occurs within one second.

[0093] Flame inclination refers to the angle at which the line connecting the highest point and the lowest point of the flame deviates from the center line.

[0094] In one embodiment, the flame brightness can be calculated by the following formula: Light = 0.2126*R+0.7152*G+0.0722*B; Light represents the flame brightness; R represents the red channel value of the pixel; G represents the green channel value of the pixel; B represents the red channel value of the pixel.

[0095] See also Figure 3 , Figure 3 FIG. 1 is a schematic diagram of a flame structure. The characteristic parameters of the flame structure include: the inner area of ​​the flame, the middle area of ​​the flame, and the outer area of ​​the flame. In one embodiment, the characteristic parameters of the flame structure can be obtained by the following method:

[0096] Step 1: Acquire multiple frames of images during the actual combustion process in a petrochemical heating furnace;

[0097] Step 2: Extract the RGB value of each pixel on each frame of the image;

[0098] Step 3: Determine the pixel points on each frame of image that are greater than the first preset RGB value as the inner flame region in each frame of image, determine the pixel points on each frame of image that are greater than the second preset RGB value as the middle flame region in each frame of image, and determine the pixel points on each frame of image that are greater than the third preset RGB value as the outer flame region in each frame of image;

[0099] The first preset RGB value is the RGB value of the pixel point representing the inner layer of the flame, the second preset RGB value is the RGB value of the pixel point representing the middle layer of the flame, and the third preset RGB value is the RGB value of the pixel point representing the outer layer of the flame.

[0100] Step 4: superimpose the flame inner region, flame middle region and flame outer region in each frame image, and determine the probability of each pixel appearing in the flame inner region, flame middle region and flame outer region formed after superposition;

[0101] Step 5: Determine the inner flame region for pixels whose probability of occurrence is greater than the first preset probability threshold, determine the middle flame region for pixels whose probability of occurrence is greater than the second preset probability threshold, and determine the outer flame region for pixels whose probability of occurrence is greater than the first preset probability threshold.

[0102] The first preset probability refers to the probability value of determining the inner area of ​​the flame;

[0103] The second preset probability refers to the probability value of determining the middle layer area of ​​the flame;

[0104] The third preset probability refers to the probability value of determining the outer area of ​​the flame;

[0105] Specifically, after each frame of the image is superimposed, the number of pixels in each frame of the image that are in the inner region of the flame is calculated, and then divided by the number of frames of the image to obtain the probability that each pixel is in the inner region of the flame, and then the probability that each pixel is in the inner region of the flame is compared with a preset probability value, and the pixels greater than the preset probability value are taken as pixels in the inner region of the flame, and finally all the pixels that meet the above conditions constitute the inner region of the flame. The method for obtaining the middle region of the flame and the outer region of the flame is the same as the method for obtaining the inner region of the flame, and will not be repeated here.

[0106] Step 6: Based on the flame inner area, flame middle area and flame outer area, the flame inner area, flame middle area and flame outer area are obtained.

[0107] In the above-mentioned step of obtaining characteristic parameters of flame morphology and characteristic parameters of flame structure, since the nozzles in different types of petrochemical heating furnaces are arranged in different ways, such as side-by-side, stacked and ring forms, the multiple frames of images can be images of the petrochemical heating furnace in the actual combustion process at different angles. In this embodiment, since the images of the petrochemical heating furnace in the actual combustion process at different angles can accurately show the flame morphology and flame structure, the subsequent image recognition algorithm can accurately identify the flame morphology and flame structure.

[0108] The preset thresholds include: a preset flame brightness difference threshold, a preset flame height difference threshold, a preset flame pulsation frequency difference threshold, a preset flame inclination difference threshold, a preset flame inner layer area difference threshold, a preset flame middle layer area difference threshold and a preset flame outer layer area difference threshold.

[0109] Among them, the preset threshold can be set according to application requirements. For example, when particularly high accuracy is not required to identify the unstable state of flame combustion, the preset threshold can be set to a smaller value. When particularly high accuracy is required to identify the unstable state of flame combustion, the preset threshold can be set to a larger value.

[0110] For ease of understanding, the following is an example of a process of determining a feature whose difference between a flame feature and a corresponding preset flame feature is greater than a preset threshold as a target flame feature:

[0111] Assume the flame area is 2m 2 、Flame brightness is 3cd / m 2 , flame height is 4m, flame pulsation frequency is 50Hz, flame inclination angle is 3°, flame inner area is 12m 2 、The area of ​​the middle layer of the flame is 9m 2 The outer area of ​​the flame is 8m 2 , the preset flame area is 4m 2 , the preset flame brightness is 2cd / m2 , preset flame height is 1m, preset flame pulsation frequency is 40Hz, preset flame inclination angle is 13°, preset flame inner layer area is 20m 2 、The preset flame middle layer area is 11m 2 The outer area of ​​the flame is set to 7m 2 , and the preset flame area difference threshold is 2m 2 , the preset flame brightness difference threshold is 1cd / m 2 , the preset flame height difference threshold is 3m, the preset flame pulsation frequency difference threshold is 8Hz, the preset flame inclination difference threshold is 8°, and the preset flame inner layer area difference threshold is 6m 2 , the preset flame middle layer area difference threshold is 2m 2 The difference threshold between the preset flame outer area and the flame outer area is 1m 2 .

[0112] Then, the differences between the flame area, flame brightness, flame height, flame pulsation frequency, flame inclination, flame inner area, flame middle area and flame outer area and the preset flame area, flame brightness, flame height, flame pulsation frequency, flame inclination, flame inner area, flame middle area and flame outer area are calculated respectively, and the error value of the flame area is 2m 2 , the error value of flame brightness is 1cd / m 2 The error value of flame height is 3m, the error value of flame pulsation frequency is 10Hz, the error value of flame inclination is 10°, and the error value of flame inner area is 8m 2 , the error value of the flame middle layer area is 2m 2 The error value of the outer area of ​​the flame is 1m 2 .

[0113] Finally, the error values ​​of flame area, flame brightness, flame height, flame pulsation frequency, flame inclination, flame inner area, flame middle area and flame outer area are compared with the preset flame brightness difference threshold, flame height difference threshold, flame pulsation frequency difference threshold, flame inclination difference threshold, flame inner area difference threshold, flame middle area difference threshold and flame outer area difference threshold, respectively, so that it is known that the target flame characteristics are that the flame pulsation frequency, flame inclination and flame inner area have the largest differences, which are 2Hz, 2° and 2m respectively. 2 , and the others are all 0, that is, these flame characteristics change most obviously when the flame combustion state changes, and can obviously characterize the changes in the flame combustion state, thereby providing effective input data for subsequent model identification.

[0114] S200, using a plurality of different types of pre-trained neural network models to classify target flame characteristics and obtain a plurality of flame combustion instability states;

[0115] A neural network model refers to a neural network model obtained by training the above neural network using preset training data. The preset training data here can be set according to the specific actual situation. For example, in the task of identifying the unstable state of flame combustion, the preset training data refers to the flame characteristics. In the supervised learning training process, the training data needs to be correctly labeled. Different types of neural network models refer to different types of neural network models, such as convolutional neural network models and self-organizing neural network models.

[0116] The multiple flame combustion unstable states here may include: a first flame combustion unstable state, a second flame combustion unstable state and a third flame combustion unstable state, and a method for obtaining the multiple flame combustion unstable states may include the following steps:

[0117] S210, using a pre-trained convolutional neural network model to classify characteristic parameters of the target flame morphology to obtain a first flame combustion unstable state;

[0118] Please refer to Figure 4 , Figure 4 The convolutional neural network in includes: a first convolution (conv1) layer, a first pooling (pool1) layer, a second convolution (conv2) layer, a second pooling (pool2) layer, a full connection layer, a dropout layer and a normalized exponential function (Softmax) layer; wherein, the first convolution layer, the first pooling layer, the second convolution layer, the second pooling layer, the full connection layer, the dropout layer and the normalized exponential function layer are connected in sequence.

[0119] A real-time method for classifying characteristic parameters of target flame morphology using a pre-trained convolutional neural network model is, for example: using the above-mentioned convolutional neural network to train a convolutional neural network, and then using the trained convolutional neural network to classify target flame features to obtain a first flame combustion instability state; wherein the normalized exponential function (Softmax) layer can output the probabilities of multiple flame combustion instability states, and the probability value of the probability of multiple flame combustion instability states with the largest probability value can be used as the first flame combustion instability state, or the probability value of the probability of multiple flame combustion instability states with the largest probability value can be directly used as the first flame combustion instability state. In other words, the first flame combustion instability state can be either a classification label or a classification probability, and the definition here will be used in the following voting method.

[0120] S220, using a pre-trained recurrent neural network model to classify characteristic parameters of the target flame morphology to obtain a second flame combustion unstable state;

[0121] A recurrent neural network (RNN) model refers to a neural network model obtained by training the above recurrent neural network using preset training data. Common recurrent neural network models include LSTM models and Bi-LSTM models. In an embodiment of the present application, the recurrent neural network model may be a long short-term memory network model (i.e., an LSTM model) or a bidirectional long short-term memory network model (i.e., a Bi-LSTM model).

[0122] For ease of understanding and explanation, the recurrent neural network is described as a bidirectional long short-term memory network (Bi-LSTM) as follows:

[0123] See also Figure 5 , Figure 5 The bidirectional long short-term memory network in includes: input layer, forward layer, backward layer and output layer; wherein, the input layer, forward layer, backward layer and output layer are connected in sequence, and the connecting lines between the layers represent the connection weights between the layers, i.e., w and values ​​in the figure. Specifically, for example, w1 in the figure represents the connection weight between the input layer and the forward layer. The bidirectional long short-term memory network can be used to train and identify flame features. Bidirectional LSTM is better than unidirectional LSTM. Bidirectional LSTM can use information in both positive and negative directions at the same time, so that the final prediction is more accurate. The main structure of the above-mentioned bidirectional LSTM is the combination of two unidirectional LSTMs. At each time t, the input will be provided to the two neural networks in opposite directions at the same time, and the output is jointly determined by the states of the two unidirectional LSTMs.

[0124] A specific implementation method for classifying the characteristic parameters of the target flame morphology using a pre-trained recurrent neural network model is, for example: using the above-mentioned recurrent neural network to train to obtain a recurrent neural network model, and then using the trained recurrent neural network model to classify the target flame characteristics to obtain a second flame combustion instability state; in the specific implementation process, a normalized exponential function (Softmax) layer can be connected to the final output of the recurrent neural network model, where the normalized exponential function layer can output the probabilities of multiple flame combustion instability states, and the probability value with the largest probability value among the multiple flame combustion instability states can be used as the second flame combustion instability state, or the probability value with the largest probability value among the multiple flame combustion instability states can be directly used as the second flame combustion instability state.

[0125] S230, using a pre-trained self-organizing neural network model to classify characteristic parameters of the target flame morphology to obtain a third flame combustion unstable state.

[0126] See 6, Figure 6 The self-organizing neural network model in is also called self-organizing competitive neural network, which includes: input layer and competitive layer. It automatically finds the inherent rules and essential attributes in the sample, and self-organizes and adaptively changes the network parameters and structure. It is usually implemented through competitive learning, which is suitable for solving problems in pattern classification and recognition.

[0127] The self-organizing neural network belongs to the feedforward neural network and adopts the unsupervised learning algorithm. Its principle is: the neurons in the competition layer compete (matching with the input pattern) to select a winner, and its output represents the classification of the input pattern.

[0128] S300: determining one of the multiple flame combustion unstable states as a target flame combustion unstable state.

[0129] See also Figure 7 , Figure 7 The convolutional neural network model, the recurrent neural network model and the self-organizing neural network model are shown in the figure. The convolutional neural network model, the recurrent neural network model and the self-organizing neural network model respectively classify the target flame characteristics, and respectively obtain the first flame combustion instability state, the second flame combustion instability state and the third flame combustion instability state. The multiple flame combustion instability states here may include the above-mentioned first flame combustion instability state and the second flame combustion instability state, and may also include the above-mentioned first flame combustion instability state, the second flame combustion instability state and the third flame combustion instability state. Of course, in the specific implementation process, if there are more flame combustion instability states obtained by different types of neural network models, the flame combustion instability state may also be included in the multiple flame combustion instability states.

[0130] Specifically, there are many ways to determine one of the multiple flame combustion unstable states as the target flame combustion unstable state. Here are only a few ways to determine the target flame combustion unstable state: the first is to determine the target classification based on the absolute majority voting method; the second is to determine the target classification based on the relative majority voting method; the third is to determine the target classification based on the weighted voting method; the following will describe these three methods in detail respectively, and here first introduces the first method of determining the target flame combustion unstable state based on the absolute majority voting method, which may include the following steps:

[0131] Step 1: Use the relative majority voting method to vote on multiple flame combustion instability states to obtain the voting value corresponding to each flame combustion instability state in the multiple flame combustion instability states;

[0132] Step 2: Determine the flame combustion unstable state with the most votes among the multiple flame combustion unstable states as the target multiple flame combustion unstable state.

[0133] First, a classifier is pre-built. The specific classifier is:

[0134] Assume that for an N-classification problem, the category labels are {C1,C2,...,C N}, if the i-th classifier is represented as h i , the predicted flame combustion instability state of the i-th classifier is represented as h i (x), then the prediction output of the classifier on sample x is an N-dimensional vector (h1(x); h2(x); ...; h i (x); ...; h N (x)).

[0135] Then, multiple flame combustion instability states are input into a pre-built classifier to obtain the votes among the multiple flame combustion instability states, and the flame combustion instability state with the most votes and more than half of the votes among the multiple flame combustion instability states is determined as the target flame combustion instability state, that is, if the votes of a certain mark exceed half, it is predicted as the mark, otherwise the prediction is rejected.

[0136] As described above, the first method for determining the target flame combustion unstable state based on the absolute majority voting method is introduced above. The second method for determining the target flame combustion unstable state based on the relative majority voting method is introduced below. The method may include the following steps:

[0137] Step 1: Use the relative majority voting method to vote on multiple flame combustion instability states to obtain the voting value corresponding to each flame combustion instability state in the multiple flame combustion instability states;

[0138] Step 2: Determine the flame combustion unstable state with the most votes among the multiple flame combustion unstable states as the target multiple flame combustion unstable state.

[0139] First, a classifier is pre-built. The specific classifier is:

[0140] Assume that for an N-classification problem, that is, to determine one unstable flame state from N unstable flame states according to the target flame feature, the classification category label of the target flame feature is {C1, C2, ..., C N}, the learner (e.g., represented by h i ) on the sample data (such as the target flame feature above, represented by x) is an N-dimensional vector represented by (h1(x); h2(x); ...; h N (x)).

[0141] Then, multiple flame combustion instability states are input into the pre-built classifier, and the following formula is used for calculation to obtain the voting value corresponding to each flame combustion instability state in the multiple flame combustion instability states:

[0142]

[0143] That is, the prediction is the tag with the most votes. If there are multiple tags with the highest votes, one of them is randomly selected as the prediction result. In formula (1), h i is the above-mentioned learner or classifier, x is the above-mentioned sample data (such as the above-mentioned target flame characteristics), and T is the number of learners or classifiers. From the above analysis, it can be seen that when T=3, the results of the absolute majority voting method and the relative majority voting method are completely consistent.

[0144] As described above, there are three ways to determine the unstable state of target flame combustion. The second way to determine the unstable state of target flame combustion based on the relative majority voting method is introduced above. The third way to determine the unstable state of target flame combustion based on the weighted voting method is introduced below:

[0145] Step 1: Use a weighted voting method to vote on each of the multiple flame combustion instability states to obtain multiple voting values;

[0146] First, a classifier is pre-built. The specific classifier is:

[0147] Assume that for an N-classification problem, that is, to determine a flame combustion instability state from N categories according to the target flame characteristics, the classification category labels of the target flame characteristics are {C1, C2, ..., C N}, the learner (for example, represented by hi) predicts the output of the sample data (for example, the target flame feature above, represented by x) as an N-dimensional vector represented by (h1(x); h2(x); ...; h N (x)).

[0148] Then, multiple flame combustion instability states are input into the pre-built classifier, and the following formula is used for calculation to obtain the voting value corresponding to each flame combustion instability state in the multiple flame combustion instability states:

[0149]

[0150] Among them, w i is the classifier h i The weight of w is , and T is the number of classifiers. i The general conditions that must be met are x is the x-th unstable flame combustion state among the multiple unstable flame combustion states, and H(x) is the voting value corresponding to each unstable flame combustion state among the multiple unstable flame combustion states.

[0151] It should be noted that the above formulas (1) and (2) do not restrict the output type of each classifier h(x). Different classifiers may have different output types. Common h(x) output types include class labels and class probabilities. For class labels: belongs to {0,1}, that is, 0 or 1, that is, if the classifier h i Predict sample x as c j , its value is 1, otherwise it is 0; and for class probability: belongs to [0,1], that is, any number between 0 and 1, which includes 0 and 1. This is equivalent to the posterior probability P(c j |x) is an estimate of the value of

[0152] Step 2: Perform weighted summation on multiple voting values ​​to obtain multiple voting scores;

[0153] Specifically, the above-mentioned multiple flame combustion unstable states can be three flame combustion unstable states, and the corresponding multiple voting scores are three voting scores, and the weights of these three voting scores are 0.3, 0.3 and 0.4 respectively; the first voting score is for the first flame combustion unstable state, the second voting score is for the second flame combustion unstable state, and the third voting score is for the third flame combustion unstable state, then the voting scores of these three flame combustion unstable states are: 0.3, 0.3 and 0.4 respectively.

[0154] Step 3: Determine the flame combustion instability state with the highest voting score among the multiple voting scores as the target flame combustion instability state.

[0155] Specifically, the unstable flame combustion state of the three models, namely the recurrent neural network model, the self-organizing neural network model and the convolutional neural network model, each has one vote, and the weights of the three unstable flame combustion states are 0.220, 0.301 and 0.479 respectively. Then the voting scores of the three unstable flame combustion states are 0.220, 0.301 and 0.479 respectively. The unstable flame combustion state with the highest voting score among multiple voting scores should be the unstable flame combustion state corresponding to 0.479, that is, the unstable flame combustion state obtained by the convolutional neural network model is taken as the target unstable flame combustion state.

[0156] In the specific implementation process, the embodiment of the present application adopted a variety of deep learning neural network models to conduct a comparative study on the recognition of flame characteristics. The test results of each network model on the test set are shown in the following table.

[0157]

[0158] As can be seen from the table, the comparison results show that when accuracy is selected as the evaluation indicator of the model, the results vary greatly for different networks. As for the self-organizing neural network model, since the self-organizing neural network can adapt to the weights and easily find the optimal solution, it achieves a recognition rate of 94.58%.

[0159] The method of ensemble learning was used to identify all test sets using the relative majority voting method and weighted voting method for the three trained models, namely the convolutional neural network model, the recurrent neural network model and the self-organizing neural network model. The best result was achieved by the weighted voting method based on class probability, which reached a recognition rate of 96.33%, which was greatly improved compared with the previous single deep learning networks. The reason is that the method based on ensemble learning and weighted voting was used to combine the three deep learning neural networks, namely the convolutional neural network model, the recurrent neural network model and the self-organizing neural network model, into a meta-classifier. Compared with a single classifier, the meta-classifier has better generalization performance. On this basis, the weighted voting method was used to make the optimal deep learning neural network have the largest voting weight, and the suboptimal deep learning neural network have a larger voting weight. Therefore, it is guaranteed that the final recognition result is better than that of a single deep learning neural network.

[0160] In this embodiment, the flame characteristics of the petrochemical heating furnace in the actual combustion process are obtained, and the characteristics whose difference between the flame characteristics and the corresponding preset flame characteristics is greater than the preset threshold are determined as the target flame characteristics; the target flame characteristics are classified using multiple different types of pre-trained neural network models to obtain multiple flame combustion instability states; the multiple flame combustion instability states are voted to select one flame combustion instability state from the multiple flame combustion instability states as the target flame combustion instability state. The present invention provides effective data input for subsequent models by screening out target flame characteristics that can accurately characterize the changes in flame combustion states from the flame characteristics. In addition, by using different types of models for ensemble learning, the advantages and disadvantages of different types of models can be combined, thereby improving the accuracy of flame combustion instability state recognition.

[0161] Embodiment 2

[0162] Optionally, the above step S300 may further include:

[0163] S400, based on the unstable combustion state of the target flame, generating alarm information corresponding to the unstable combustion state of the target flame.

[0164] Unstable flame combustion states include, but are not limited to, flame extinction, flame oscillation, flame instability, and so on.

[0165] Alarm information includes but is not limited to: light alarm reminder, voice prompt, etc. Among them, the light alarm reminder can be a flashing indicator light of the equipment, a red light warning; the content of the voice prompt can be "pay attention to the burning status of the flame in time", etc.

[0166] A specific embodiment is given below for illustration:

[0167] When the unstable state of flame combustion is flame extinction, the indicator light turns red, and the message "Flame is extinct, please check the cause in time" is broadcasted; when the unstable state of flame combustion is flame oscillation, the indicator light turns orange, and the message "Flame is in an oscillating state, please close the nozzle in time and check the cause" is broadcasted; when the unstable state of flame combustion is flame instability, the indicator light turns yellow, and the message "Flame is in an unstable state, please shut down the device in time" is broadcasted.

[0168] In this embodiment, different alarm information is generated based on different unstable flame combustion states, that is, differentiated alarms are implemented, so that the user can clearly distinguish the type of the current unstable flame combustion state and then provide targeted treatment measures.

[0169] Embodiment 3

[0170] Reference Figure 8 , Figure 81 is a schematic diagram of the structure of a monitoring device for an unstable flame combustion state provided by an embodiment of the present invention. The monitoring device 200 for an unstable flame combustion state includes:

[0171] The feature acquisition module 210 is used to acquire the flame feature of the petrochemical heating furnace in the actual combustion process, and determine the feature whose difference between the flame feature and the corresponding preset flame feature is greater than a preset threshold as the target flame feature;

[0172] The state recognition module 220 is used to analyze the target flame characteristics using a plurality of different types of pre-trained neural network models to obtain a plurality of flame combustion instability states;

[0173] The target determination module 230 is used to vote on multiple flame combustion instability states to select one flame combustion instability state from the multiple flame combustion instability states as a target flame combustion instability state.

[0174] It should be understood that the device corresponds to the above-mentioned flame combustion unstable state monitoring method embodiment, and can execute the various steps involved in the above-mentioned method embodiment. The specific functions of the device can be found in the above description. To avoid repetition, the detailed description is appropriately omitted here. The device includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the operating system (OS) of the device.

[0175] Embodiment 4

[0176] In a typical configuration, an electronic device includes: a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the machine-readable instructions are executed by the processor, the above-mentioned method for monitoring unstable flame combustion state is performed.

[0177] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0178] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0179] Embodiment 5

[0180] An embodiment of the present invention also provides a computer-readable storage medium, which stores instructions, and the instructions are used to execute a program with the following method steps when executed by a processor: obtaining the flame characteristics of a petrochemical heating furnace in an actual combustion process, and determining the characteristics whose difference between the flame characteristics and the corresponding preset flame characteristics is greater than a preset threshold as target flame characteristics; using pre-trained multiple different types of neural network models to classify the target flame characteristics to obtain multiple flame combustion instability states; voting on the multiple flame combustion instability states to select one flame combustion instability state from the multiple flame combustion instability states as the target flame combustion instability state.

[0181] In one embodiment, the above-mentioned method for monitoring unstable flame combustion state also includes: obtaining multiple frames of images in the petrochemical heating furnace during the actual combustion process; for each frame of image, using different image recognition algorithms to respectively identify the frame of image, and obtain the flame morphology of each frame of image under each image recognition algorithm, and determine the flame morphology under each image recognition algorithm based on the flame morphology of each frame of image under each image recognition algorithm; based on the characteristic parameters of the flame morphology under each image recognition algorithm, perform flame morphology correction to obtain the corrected flame morphology; based on the corrected flame morphology, obtain the characteristic parameters of the flame morphology.

[0182] In one embodiment, the above-mentioned method for monitoring the unstable state of flame combustion also includes: acquiring multiple frames of images in the actual combustion process of the petrochemical heating furnace; extracting the RGB value of each pixel on each frame of the image; determining the pixel points on each frame of the image that are greater than the first preset RGB value as the inner flame area in each frame of the image, and determining the pixel points on each frame of the image that are greater than the second preset RGB value as the middle flame area in each frame of the image, and determining the pixel points on each frame of the image that are greater than the third preset RGB value as the outer flame area in each frame of the image; superimposing the inner flame area, the middle flame area and the outer flame area in each frame of the image, and determining the probability of each pixel point appearing in the inner flame area, the middle flame area and the outer flame area formed after superposition; determining the inner flame area for the pixel points whose probability of appearance is greater than the first preset probability threshold, and determining the middle flame area for the pixel points whose probability of appearance is greater than the second preset probability threshold, and determining the outer flame area for the pixel points whose probability of appearance is greater than the first preset probability threshold; based on the inner flame area, the middle flame area and the outer flame area, the inner flame area, the middle flame area and the outer flame area are obtained.

[0183] In one embodiment, the above-mentioned flame combustion unstable state monitoring method also includes: using a pre-trained convolutional neural network model to classify the characteristic parameters of the target flame morphology to obtain a first flame combustion unstable state; using a pre-trained recurrent neural network model to classify the characteristic parameters of the target flame morphology to obtain a second flame combustion unstable state; using a pre-trained self-organizing neural network model to classify the characteristic parameters of the target flame morphology to obtain a third flame combustion unstable state.

[0184] In one embodiment, the above-mentioned flame combustion unstable state monitoring method also includes: voting on multiple flame combustion unstable states to obtain the number of votes corresponding to each of the multiple flame combustion unstable states; and determining the flame combustion unstable state with the most votes among the multiple flame combustion unstable states and more than half of the votes as the flame combustion unstable state.

[0185] In one embodiment, the above-mentioned flame combustion unstable state monitoring method also includes: using relative majority voting method to vote on multiple flame combustion unstable states to obtain voting values ​​corresponding to each of the multiple flame combustion unstable states; determining the flame combustion unstable state with the most votes among the multiple flame combustion unstable states as the target multiple flame combustion unstable states.

[0186] In one embodiment, the above-mentioned flame combustion instability state monitoring method also includes: using a weighted voting method to vote on each of the multiple flame combustion instability states to obtain multiple voting values; weighted summing the multiple voting values ​​to obtain multiple voting scores; and determining the flame combustion instability state with the highest voting score among the multiple voting scores as the target flame combustion instability state.

[0187] In one embodiment, the above-mentioned method for monitoring unstable flame combustion state further includes: generating alarm information corresponding to the target unstable flame combustion state based on the target unstable flame combustion state.

[0188] Embodiment 6

[0189] The embodiment of the present invention further provides a petrochemical heating furnace, comprising a boiler furnace and the above-mentioned monitoring device for the unstable state of flame combustion.

[0190] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0191] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0192] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0193] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0194] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0195] 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 independently, or two or more modules may be integrated to form an independent part.

[0196] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0197] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for monitoring unstable state of flame combustion, characterized in that: include: Obtaining the flame characteristics of the petrochemical heating furnace during the actual combustion process, and determining the characteristics of the flame characteristics whose difference with the corresponding preset flame characteristics is greater than a preset threshold as the target flame characteristics; Using multiple different types of pre-trained neural network models to classify target flame features, a variety of flame combustion instability states are obtained; Voting is performed on the multiple flame combustion instability states to select one flame combustion instability state from the multiple flame combustion instability states as the target flame combustion instability state.

2. The method for monitoring the unstable state of flame combustion according to claim 1, characterized in that: The flame characteristics include: characteristic parameters of flame morphology and characteristic parameters of flame structure; The characteristic parameters of the flame morphology include: flame area, flame brightness, flame height, flame pulsation frequency and flame inclination angle; The characteristic parameters of the flame structure include: the inner area of ​​the flame, the middle area of ​​the flame and the outer area of ​​the flame; The preset flame characteristics include: preset flame area, preset flame brightness, preset flame height, preset flame pulsation frequency, preset flame inclination angle, preset flame inner layer area, preset flame middle layer area and preset flame outer layer area; The preset thresholds include: a preset flame brightness difference threshold, a preset flame height difference threshold, a preset flame pulsation frequency difference threshold, a preset flame inclination difference threshold, a preset flame inner layer area difference threshold, a preset flame middle layer area difference threshold and a preset flame outer layer area difference threshold; The difference between the flame characteristic and the corresponding preset flame characteristic is greater than a preset threshold, including: The difference between the flame brightness and the preset flame brightness is greater than a preset flame brightness difference threshold; The difference between the flame height and the preset flame height is greater than a preset flame height difference threshold; The difference between the flame pulsation frequency and the preset flame pulsation frequency is greater than the preset flame pulsation frequency difference threshold; The difference between the flame inclination angle and the preset flame inclination angle is greater than the preset flame inclination angle difference threshold; The difference between the inner layer area of ​​the flame and the preset inner layer area of ​​the flame is greater than the preset inner layer area difference threshold of the flame; The difference between the flame middle layer area and the preset flame middle layer area is greater than the preset flame middle layer area difference threshold; The difference between the flame outer layer area and a preset flame outer layer area is greater than a preset flame outer layer area difference threshold.

3. The method for monitoring unstable state of flame combustion according to claim 2, characterized in that: The characteristic parameters of the flame morphology are obtained in the following way: Acquire multiple frames of images of the actual combustion process in a petrochemical heating furnace; For each frame of image, different image recognition algorithms are used to respectively recognize the frame of image, and the flame morphology of each frame of image under each image recognition algorithm is obtained, and the flame morphology under each image recognition algorithm is determined based on the flame morphology of each frame of image under each image recognition algorithm; Based on the characteristic parameters of the flame morphology under each image recognition algorithm, the flame morphology is corrected to obtain the corrected flame morphology; Based on the corrected flame morphology, characteristic parameters of the flame morphology are obtained.

4. The method for monitoring the unstable state of flame combustion according to claim 3, characterized in that: The different image recognition algorithms include: any two or three of the maximum inter-class method difference method, Boolean variable and Yolo v5 algorithm.

5. The method for monitoring unstable state of flame combustion according to claim 2, characterized in that: The characteristic parameters of the flame structure are obtained by: Acquire multiple frames of images of the actual combustion process in a petrochemical heating furnace; Extract the RGB value of each pixel on each frame of the image; Determine the pixel points on each frame of image that are greater than a first preset RGB value as the inner flame region in each frame of image, determine the pixel points on each frame of image that are greater than a second preset RGB value as the middle flame region in each frame of image, and determine the pixel points on each frame of image that are greater than a third preset RGB value as the outer flame region in each frame of image; The inner flame region, the middle flame region and the outer flame region in each frame of the image are superimposed, and the probability of each pixel point appearing in the inner flame region, the middle flame region and the outer flame region formed after superposition is determined; The pixel points whose occurrence probability is greater than a first preset probability threshold are determined as the inner region of the flame, the pixel points whose occurrence probability is greater than a second preset probability threshold are determined as the middle region of the flame, and the pixel points whose occurrence probability is greater than the first preset probability threshold are determined as the outer region of the flame; Based on the flame inner region, flame middle region and flame outer region, the flame inner area, flame middle area and flame outer area are obtained.

6. The method for monitoring unstable state of flame combustion according to claim 1, characterized in that: The multiple flame combustion unstable states include: a first flame combustion unstable state, a second flame combustion unstable state and a third flame combustion unstable state; The target flame characteristics are classified by using a plurality of different types of pre-trained neural network models to obtain a plurality of flame combustion instability states, including: Using a pre-trained convolutional neural network model to classify characteristic parameters of the target flame morphology, and obtain the first flame combustion unstable state; Using a pre-trained recurrent neural network model to classify characteristic parameters of the target flame morphology, and obtain the second flame combustion unstable state; The characteristic parameters of the target flame morphology are classified using a pre-trained self-organizing neural network model to obtain the third flame combustion unstable state.

7. The method for monitoring unstable state of flame combustion according to claim 1, characterized in that: The step of determining one of the multiple flame combustion unstable states as a target flame combustion unstable state comprises: Voting on the multiple flame combustion unstable states to obtain the number of votes corresponding to each of the multiple flame combustion unstable states; The flame combustion unstable state having the largest number of votes among the multiple flame combustion unstable states, and having more than half of the votes, is determined as the flame combustion unstable state.

8. The method for monitoring unstable state of flame combustion according to claim 1, characterized in that: The step of determining one of the multiple flame combustion unstable states as a target flame combustion unstable state comprises: Voting on the multiple flame combustion instability states using a relative majority voting method to obtain a voting value corresponding to each of the multiple flame combustion instability states; The flame combustion unstable state having the largest number of votes among the plurality of flame combustion unstable states is determined as the target plurality of flame combustion unstable states.

9. The method for monitoring unstable state of flame combustion according to claim 1, characterized in that: The step of determining one of the multiple flame combustion unstable states as a target flame combustion unstable state comprises: Voting for each of the plurality of flame combustion instability states using a weighted voting method to obtain a plurality of voting values; Performing weighted summation on the multiple voting values ​​to obtain multiple voting scores; The flame combustion instability state having the highest voting score among the multiple voting scores is determined as the target flame combustion instability state.

10. The method for monitoring unstable state of flame combustion according to claim 1, characterized in that: The method further comprises: Based on the target flame combustion unstable state, alarm information corresponding to the target flame combustion unstable state is generated.

11. A monitoring device for unstable flame combustion, characterized in that: include: A feature acquisition module is used to acquire the flame features of the petrochemical heating furnace during the actual combustion process, and determine the features whose difference between the flame features and the corresponding preset flame features is greater than a preset threshold as target flame features; A state recognition module is used to analyze the target flame characteristics using multiple different types of pre-trained neural network models to obtain multiple flame combustion instability states; The target determination module is used to vote on multiple flame combustion instability states to select one flame combustion instability state from the multiple flame combustion instability states as a target flame combustion instability state.

12. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the machine-readable instructions are executed by the processor, the method for monitoring the unstable state of flame combustion according to any one of claims 1 to 9 is executed.

13. A computer-readable storage medium storing computer instructions, characterized in that: When the computer instructions are executed on a computer, the computer is enabled to execute the method for monitoring the unstable state of flame combustion according to any one of claims 1 to 9.

14. A petrochemical heating furnace, characterized in that: The invention comprises a boiler furnace and the monitoring device for the unstable state of flame combustion as claimed in claim 11.

Images