Waste incineration state identification method and incineration control method based on deep learning

Abstract

A garbage incineration state identification method and incineration control method based on deep learning, comprising: firstly calculating the incineration state parameters of garbage incineration sample images in a preset garbage incineration sample image set, classifying the garbage incineration sample images according to the incineration state parameters and the predetermined incineration state categories, then using the garbage incineration sample image set as training data and the categories of the garbage incineration sample images as data labels to train a neural network model to obtain a garbage incineration state identification model, inputting the garbage incineration images into the garbage incineration state identification model to obtain the garbage incineration state, thereby realizing real-time automatic monitoring of the garbage incineration state, which is beneficial to improving the accuracy of garbage incineration state identification and the stability of garbage incineration effect, bringing higher marginal economic benefits to garbage incineration power plants, and at the same time liberating manpower, reducing labor costs, and improving the working environment of workers.

Description

Technical Field

[0001] The present invention relates to the technical field of waste incineration, and in particular to a waste incineration state identification method and an incineration control method based on deep learning. Background Art

[0002] The annual collection volume and incineration disposal rate of urban domestic waste in my country have increased year by year, and the number of waste incineration power generation projects has increased day by day. However, the composition of my country's waste is complex and changeable. In addition, with the development of the national economy, the calorific value of waste is gradually higher than the design calorific value of the incinerator, which leads to deterioration of combustion, resulting in a sharp increase in the original concentration of pollutants, incomplete incineration of waste, high thermal loss rate of slag, and many other combustion problems.

[0003] At present, the on-site optimization and control of waste incineration is mainly based on the feedback values ​​of traditional thermocouples and other measuring elements and the combustion diagnosis conclusions obtained by manual flame observation. The control method also depends on the experience and habits of the operators. However, the monitoring system based on traditional thermocouples and other measuring elements has a narrow coverage range and the measurement of parameters has a lag, and the feedback on combustion problems that occur during the waste incineration process is not timely; manual operation relies more on subjective experience, and its incineration effect is unstable and brings a great burden of personnel training to the enterprise. At the same time, the relatively harsh working environment in the incineration plant is also likely to cause harm to the health of workers. Therefore, it is necessary to improve the intelligence and automation level of waste incineration state identification during waste incineration. Summary of the invention

[0004] The present application provides a garbage incineration state identification method and an incineration control method based on deep learning, aiming to solve the problem that the judgment of the garbage incineration state in the existing garbage incineration technology mainly relies on manual experience, resulting in inaccurate judgment and unstable incineration effect.

[0005] According to the first aspect, an embodiment provides a method for identifying a waste incineration state based on deep learning, comprising:

[0006] Obtain images of waste incineration;

[0007] Inputting the garbage incineration image into a garbage incineration state recognition model to obtain the garbage incineration state;

[0008] The waste incineration state identification model is obtained by the following steps:

[0009] Calculating the burning state parameters of the garbage incineration sample images in the preset garbage incineration sample image set;

[0010] Classifying the garbage incineration sample images according to the predetermined incineration state categories based on the incineration state parameters;

[0011] A neural network model is constructed, the garbage incineration sample image set is used as training data, the categories of the garbage incineration sample images are used as data labels, and the neural network model is trained to obtain the garbage incineration state recognition model.

[0012] In one embodiment, the burning state parameters include one or more of the image average grayscale, flame effective area ratio, high temperature effective area ratio, flame high temperature ratio, flame center of mass, flame center of mass offset distance, high temperature zone circularity, flame effective area variance per unit time, and image grayscale mean variance per unit time.

[0013] In one embodiment, the classifying the garbage incineration sample images according to the predetermined incineration state categories based on the incineration state parameters includes:

[0014] The EM algorithm is used to perform cluster analysis on the incineration state parameters to obtain three types of indicators of the garbage incineration sample image, namely flame shape, flame temperature and flame flicker. According to the three types of indicators, the garbage incineration sample image is classified according to the predetermined incineration state category.

[0015] In one embodiment, the classification of the garbage incineration sample images according to the three indicators of flame shape, flame temperature and flame flickering and the predetermined incineration state category includes:

[0016] The three indicators of flame shape, flame temperature and flame flicker of the garbage incineration sample image are input into a preset expert system to obtain the incineration state category to which the garbage incineration sample image belongs, wherein the expert system has expert experience in classifying the garbage incineration sample image according to the predetermined incineration state category based on flame shape, flame temperature and flame flicker.

[0017] In one embodiment, the step of using the garbage incineration sample image set as training data and the categories of the garbage incineration sample images as data labels to train the neural network model includes:

[0018] The waste incineration sample image set is divided into a training set, a validation set and a test set in a certain ratio, the training set is used to train the neural network model, the validation set is used to perform hyperparameter tuning on the trained neural network model, and the test set is used to verify the accuracy of the trained neural network model.

[0019] In one embodiment, the burning state categories include uniform burning, incomplete burning, lateral bias burning and longitudinal bias burning.

[0020] In one embodiment, the deep learning-based garbage incineration state identification method further includes: when the garbage incineration states obtained continuously within a certain period of time are consistent, the garbage incineration state within the time period is determined to be the consistent garbage incineration state.

[0021] According to the second aspect, an embodiment provides a garbage incineration control method based on deep learning, including:

[0022] Acquire a garbage incineration image, and input the garbage incineration image into a garbage incineration state recognition model to obtain a garbage incineration state;

[0023] Obtaining incinerator environmental monitoring data and incinerator operation data, and predicting flue gas pollutant concentration and main combustion section temperature field distribution according to the garbage incineration status and incinerator operation data;

[0024] Inputting the incinerator environmental monitoring data, predicted flue gas pollutant concentration and predicted main combustion section temperature field distribution into a reinforcement learning control model to obtain incineration control parameters;

[0025] Sending the incineration control parameters to the incinerator so that the incinerator makes corresponding adjustments;

[0026] The waste incineration state identification model is obtained by the following steps:

[0027] Calculating the burning state parameters of the garbage incineration sample images in the preset garbage incineration sample image set;

[0028] Classifying the garbage incineration sample images according to the predetermined incineration state categories based on the incineration state parameters;

[0029] A neural network model is constructed, the garbage incineration sample image set is used as training data, the categories of the garbage incineration sample images are used as data labels, and the neural network model is trained to obtain the garbage incineration state recognition model.

[0030] In one embodiment, the reinforcement learning control model includes a Dyna-Q algorithm model and a DDPG algorithm model; the DDPG algorithm model is used to calculate the incineration control parameters based on the predicted flue gas pollutant concentration and the predicted temperature field distribution of the main combustion section; the Dyna-Q algorithm model is used to learn the next state that will be entered after outputting different incineration control parameters under different states based on the incinerator environmental monitoring data, the predicted flue gas pollutant concentration and the predicted temperature field distribution of the main combustion section, and provide it to the DDPG algorithm model for learning, the state including the flue gas pollutant concentration and the temperature field distribution of the main combustion section.

[0031] According to the third aspect, an embodiment provides a computer-readable storage medium, wherein the medium includes a program, and the program can be executed by a processor to implement the above-mentioned deep learning-based garbage incineration state identification method and garbage incineration control method.

[0032] According to the garbage incineration state identification method, incineration control method and computer-readable storage medium based on deep learning of the above-mentioned embodiments, firstly, the incineration state parameters of the garbage incineration sample images in the preset garbage incineration sample image set are calculated, and according to the incineration state parameters, the garbage incineration sample images are classified according to the predetermined incineration state categories, thereby realizing the automatic construction of training samples, which is conducive to building a large sample database to improve the accuracy of garbage incineration state identification; then, the garbage incineration sample image set is used as training data, and the categories of the garbage incineration sample images are used as data labels to train a neural network model to obtain a garbage incineration state identification model, and the garbage incineration image is input into the garbage incineration state identification model to obtain the garbage incineration state, thereby realizing real-time automatic monitoring of the garbage incineration state, which is conducive to improving the accuracy of garbage incineration state identification, thereby improving the stability of garbage incineration effect, bringing higher marginal economic benefits to garbage incineration power plants, and at the same time liberating manpower, reducing labor costs, and improving the working environment of workers. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flowchart of a method for controlling waste incineration according to an embodiment;

[0034] Figure 2 This is the relationship diagram between the incinerator environmental monitoring data and various control quantities;

[0035] Figure 3 A flowchart of a method for identifying a garbage incineration state based on deep learning according to an embodiment;

[0036] Figure 4 A flowchart of a method for constructing a garbage incineration state identification model according to an embodiment;

[0037] Figure 5 A schematic diagram of the structure of a convolutional neural network for building a garbage incineration state recognition model according to an embodiment;

[0038] Figure 6 The figure is an algorithm framework diagram of an artificial intelligence control method for a waste incineration process in an embodiment. DETAILED DESCRIPTION

[0039] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0040] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0041] The serial numbers of the components in this document, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).

[0042] Please refer to Figure 1 In one embodiment of the present application, a method for controlling waste incineration is provided, including steps 101 to 104, which are described in detail below.

[0043] Step 101: Obtain incinerator environmental monitoring data, garbage incineration status and incinerator operation data.

[0044] like Figure 2As shown in the figure, the incinerator environment monitoring data includes the main steam volume at the boiler outlet, CO concentration in the flue gas, air flow rate of the drying section grate, flue gas temperature in the secondary combustion chamber, air flow rate at the combustion section inlet, primary air volume, grate speed of the combustion section, oxygen concentration of the outlet flue gas, primary air pressure and temperature, negative pressure of the furnace and grate area, etc. The specific data to be used can be selected according to the needs. For example, if the feeder speed needs to be controlled, the main steam volume at the boiler outlet, CO concentration in the flue gas, and air flow rate of the drying section grate need to be monitored. If the grate speed of the combustion section needs to be controlled, the flue gas temperature in the secondary combustion chamber, the main steam volume at the boiler outlet, and the air flow rate at the combustion section inlet need to be monitored. If the oxygen volume needs to be controlled, the secondary air volume needs to be controlled. Therefore, it is necessary to monitor the oxygen concentration of the outlet flue gas, the CO concentration in the flue gas, the flue gas temperature of the primary combustion chamber, the main steam volume at the boiler outlet, etc. The incinerator environment is monitored in real time and the monitoring data is continuously updated.

[0045] The incinerator operation data include the temperature field distribution of the drying section, the thickness of the material layer, the current temperature field distribution of the main combustion section, the air volume of the main combustion section, the grate sliding speed, the grate turning frequency, etc., and one or more data can be selected for use according to needs.

[0046] For the garbage incineration status, this application provides a garbage incineration status identification method based on deep learning to obtain the garbage incineration status, please refer to Figure 3 In one embodiment, a method for identifying a waste incineration state comprises the following steps:

[0047] Step 301: Obtaining a garbage incineration image. The garbage incineration image can be obtained by photographing the garbage burning flame in the entire furnace using a CCD (Charge Coupled Device) camera.

[0048] Step 302: Input the garbage incineration image into the garbage incineration state recognition model to obtain the garbage incineration state. In order to make the recognition result more accurate, the garbage incineration state output for a period of time can be monitored. When the garbage incineration states obtained continuously for a period of time are consistent, the garbage incineration state in the time period is determined to be the consistent garbage incineration state. The garbage incineration state recognition model in this application can be constructed by a neural network model. For the specific construction process, please refer to Figure 4 , including steps 311 to 313, which are described in detail below.

[0049] Step 311: Calculate the burning state parameters of the garbage incineration sample images in the preset garbage incineration sample image set. The burning state parameters include one or more of the image average grayscale, flame effective area rate, high temperature effective area rate, flame high temperature rate, flame centroid, flame centroid offset distance, high temperature area circularity, unit time flame effective area variance and unit time image grayscale mean variance. The calculation methods of these parameters are introduced below.

[0050] Since the burning state parameters are calculated based on the grayscale image, it is first necessary to convert the obtained R, G, B three-channel color image into a single-channel grayscale image. The calculation formula is as follows:

[0051] g(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y),

[0052] Where g(x,y) is the grayscale value of the grayscale image at (x,y), R(x,y) is the pixel value in the red channel of the original color image at (x,y), G(x,y) is the pixel value in the green channel of the original color image at (x,y), and B(x,y) is the pixel value in the blue channel of the original color image at (x,y).

[0053] (1) Average grayscale of the image

[0054] The average grayscale of an image can be used to evaluate the overall temperature level of a burning flame. To obtain the average grayscale value of an image, we traverse all the pixels in the entire image and accumulate the sum of the grayscale values ​​of all the pixels. The average grayscale value is the quotient of the sum of the grayscale values ​​and the number of pixels. The steps are as follows:

[0055] 1) Traverse the pixels of the image to be calculated, and add up the grayscale values ​​of each pixel to get the total grayscale value sum;

[0056] 2) Calculate the total number of pixels n in the image;

[0057] 3) Find the average gray value g of the image mean =sum / n.

[0058] (2) Flame effective area ratio

[0059] The flame effective area ratio is used to evaluate the area of ​​the flame filling the image space in the grayscale image, and represents the coverage of the burning flame in the monitoring area. The calculation formula is as follows:

[0060]

[0061] Where S v is the flame effective area ratio, g this the grayscale threshold for judging the flame point. Points greater than this threshold are considered flame points. i is the number of rows of pixels, j is the number of columns of pixels, and L(x) is a step function defined as

[0062]

[0063] (3) High temperature effective area ratio

[0064] The high temperature effective area rate mainly reflects the area of ​​the burning flame with higher temperature in the entire image space. The higher the temperature, the higher the gray value of the image. The calculation formula is:

[0065]

[0066] Where S h is the flame effective area ratio, It is the grayscale threshold for judging high-temperature flame points. Points greater than this threshold are regarded as high-temperature flame points.

[0067] (4) High temperature rate

[0068] The high temperature rate mainly reflects the ratio of the area with higher combustion flame temperature to the effective flame area, which reflects the fullness of the high temperature area. The higher the value, the larger the high temperature area in the furnace, the higher the fullness, and the better the combustion state. The high temperature rate is equal to the high temperature effective area rate S h and flame effective area ratio S v ratio.

[0069] (5) Flame center of mass

[0070] The flame centroid reflects the spatial position information of the combustion, and can fully and accurately reflect whether the flame in the area is in a reasonable position, and is used to diagnose combustion problems. Its calculation formula is as follows:

[0071]

[0072]

[0073] in, is the horizontal coordinate of the flame center of mass, is the ordinate of the flame center of mass.

[0074] (6) Flame center of mass offset distance

[0075] As the old combustion material is burned out and the new combustion material is ignited, the flame position keeps moving, but this movement is non-jumpy and is continuous as a whole, that is, the flame meets the relative stability of movement. The overall movement of the flame can be described by calculating the flame mass center offset distance. The flame mass center offset distance d refers to the flame mass center Relative to the reference centroid of the four-corner tangent circle of the flame image (x s,y s ) is offset, and the calculation formula is:

[0076]

[0077] (7) Circularity in high temperature zone

[0078] The circularity of the high temperature area characterizes the geometric shape of the burning flame, which reflects the quality of the aerodynamic field. The contour length of the high temperature effective area is taken as the perimeter L. h The area surrounded by the contour line is taken as the projection area of ​​the high temperature area of ​​the flame image. Assume that there is a circle with the same area as the projection area of ​​the high temperature area of ​​the flame image, which is called the equivalent circle. The diameter of this circle is the equivalent circle diameter D. The equivalent circle circumference L is calculated based on the equivalent circle diameter d , divided by the perimeter L of the high temperature area of ​​the flame image h The obtained value is taken as the circularity The calculation formula is:

[0079] L d =πD

[0080]

[0081] (8) Variance of flame effective area per unit time

[0082] The variance of flame effective area per unit time reflects the change of flame area per unit time, and its calculation formula is as follows:

[0083]

[0084] Where D s is the variance of flame effective area per unit time, S vi is the flame effective area ratio calculated for the i-th time, S vmean is the mean value of the flame effective area rate during the statistical time period, and n is the total sample size of the statistics, that is, the number of measurements.

[0085] (9) Image grayscale mean variance per unit time

[0086] The mean variance of the grayscale value of the image per unit time reflects the fluctuation of the flame temperature, and its calculation formula is as follows:

[0087]

[0088] Where D g is the grayscale mean variance of the image per unit time, g meani is the grayscale mean of the image calculated for the i-th time, is the mean of the grayscale mean values ​​of the image sought within the statistical time period.

[0089] Step 312: Classify the garbage incineration sample images according to the predetermined incineration state categories based on the incineration state parameters. The incineration state categories can be divided into four types: uniform combustion, incomplete combustion, horizontal biased combustion, and vertical biased combustion.

[0090] In this embodiment, the EM (Expectation-maximization) algorithm and the expert system are used to classify the garbage incineration sample images. First, the EM algorithm is used to perform cluster analysis on the incineration state parameters to obtain three types of indicators of flame shape, flame temperature and flame flicker of the garbage incineration sample images. The three indicators of flame shape, flame temperature and flame flicker are the centroids of the three clusters. The EM algorithm is an algorithm for finding the maximum likelihood estimation or maximum a posteriori estimation of parameters in a probability model, in which the probability model depends on unobservable hidden variables. The hidden variables here are the centroids of the clusters: flame shape, flame temperature and flame flicker. The EM algorithm mainly includes two steps: the first step is to calculate the expectation (E step), using the existing estimated value of the hidden variable (i.e., the centroid) to calculate its maximum likelihood estimate; the second step is maximization (M step), calculating the nearest centroid of each sample image, and clustering the sample image to the nearest centroid to obtain the parameter estimation value of the probability model. The parameter estimates found in the M step are used in the next E step calculation, and this process is repeated until the centroid no longer changes. In this way, a cluster with three centroids is completed, and finally the three indicators of flame shape, flame temperature and flame flicker are extracted.

[0091] After obtaining the three indicators of flame shape, flame temperature and flame flicker, they are input into a pre-set expert system to obtain the burning state category to which the garbage incineration sample image belongs. The expert system stores expert experience in classifying the garbage incineration sample images according to the predetermined burning state category based on flame shape, flame temperature and flame flicker. Based on the expert experience, the expert system can classify the sample images according to the input flame shape, flame temperature and flame flicker indicators. In this way, the garbage incineration sample images have their own category labels, thereby realizing the automatic construction of the training sample set, which is conducive to the construction of a large sample database to improve the accuracy of garbage incineration state recognition.

[0092] Step 313: construct a neural network model, use the garbage incineration sample image set as training data, use the category of the garbage incineration sample image as a data label, train the neural network model, and obtain a garbage incineration state recognition model.

[0093] The constructed neural network model can be a convolutional neural network model, please refer to Figure 5, the convolutional neural network includes convolutional layers, batch normalization layers, pooling layers and fully connected networks. After convolution, batch normalization and pooling, the garbage incineration sample image obtains a feature vector. After the feature vector is input into the fully connected network, the probabilities of 4 categories are obtained for discrimination. Among them, convolution is an effective method to extract image features. Generally, a square convolution kernel is used to slide on the input image at a specified step size, traverse each pixel in the input image, perform convolution calculations, extract features in the image, and obtain a feature map.

[0094] Batch normalization is to standardize a small batch of data output by each layer of the network. The formula for its implementation is:

[0095]

[0096] in Represents the i-th pixel in a feature map output by the k-th convolution kernel before batch normalization. Represents the average value of all pixels in the batch feature map output by the kth convolution kernel before batch normalization. Represents the standard deviation of all pixels in the batch feature map output by the kth convolution kernel before batch normalization. Batch is an integer greater than 0. Representatives Pixels after batch normalization.

[0097] The role of pooling is to reduce the number of features (dimensionality reduction), among which maximum pooling can extract image texture and mean pooling can retain background features.

[0098] Before inputting the garbage incineration sample images into the neural network, they need to be preprocessed, mainly by standardizing and whitening the garbage incineration sample images in sequence. The purpose of data standard deviation standardization is to make the data fall into a specific range by scaling the data, which can effectively avoid the problem that the weight of data with a large order of magnitude is too large in the algorithm and the weight of data with a small order of magnitude is too small; whitening is used to reduce the correlation between features and improve the stability of the algorithm.

[0099] During training, the garbage incineration sample image set can be divided into a training set, a validation set and a test set according to a certain ratio. First, the training set is used to train the neural network model. The garbage incineration sample images in the training set are sent to the neural network in batches to obtain the prediction results of the garbage incineration state, and then compared with the corresponding data labels to calculate the loss value. According to the loss value, the parameters of the neural network are iteratively corrected through the back propagation algorithm until the preset stop condition is met. The stop condition can be that the prediction error is less than the threshold, the training time reaches the requirement, or the prediction result has converged. At the same time, the validation set is used to estimate the generalization error of the neural network model during or after training, and the hyperparameters of the neural network are updated. After the training is completed, the garbage incineration sample images in the test set are sent to the neural network to obtain the predicted garbage incineration state, which is then compared with the corresponding data labels to verify the accuracy of the trained neural network model. After several trainings and hyperparameter tuning, the final garbage incineration state recognition model is obtained.

[0100] The Tensorflow framework and keras library can be used to build a convolutional neural network for garbage incineration state identification. Four modules are mainly used, namely model, model.compile, model.fit and model.summary. The model module is mainly responsible for defining the network structure, including the number of convolutional layers, the number and size of convolution kernels, whether to perform batch normalization, the selection of activation functions, the pooling method, the number of layers of the fully connected network, the number of neurons and the activation function, etc.; the model.compile module is responsible for configuring the training method, including defining the optimizer, loss function, model evaluation indicators, etc.; the model.fit module is responsible for training parameters, including defining the training set and validation set, the number of samples input into the neural network each time, the number of iterations, etc.; the model.summary module is responsible for parameter extraction and prediction effect visualization. In practical applications, each of the above modules adopts different setting methods according to the actual situation.

[0101] The deep learning-based garbage incineration state identification method provided in the present application utilizes a neural network model to construct a garbage incineration state identification model. The garbage incineration state can be obtained by inputting a garbage incineration image into the garbage incineration state identification model, thereby realizing real-time automatic monitoring of the garbage incineration state, which is beneficial to improving the accuracy of garbage incineration state identification, while liberating manpower, reducing labor costs, and improving the working environment of workers.

[0102] The following is a further description of steps 102 to 104.

[0103] Step 102: Predict the flue gas pollutant concentration and the main combustion section temperature field distribution according to the waste incineration state and the incinerator operation data. Predicting the flue gas pollutant concentration and the main combustion section temperature field distribution in the next time series is conducive to making adjustments in advance, adapting to changes in the incineration process, and improving the stability of incineration.

[0104] In this embodiment, the LSTM (Long Short-Term Memory) model is used to predict the flue gas pollutant concentration and the temperature field distribution of the main combustion section of the next time series. LSTM is a special RNN (Recurrent Neural Network) neural network algorithm that can solve the problems of gradient disappearance and gradient explosion in the long sequence training process, and is suitable for trend prediction of long sequences. Compared with ordinary neural networks, the recurrent neural network structure is more in line with the actual structure of biological neural networks. While receiving information from other neurons, LSTM can also receive its own information, thereby forming a network structure with a loop to produce a "memory" effect. Like other neural networks, the recurrent neural network can perform parameter training by a back-propagation algorithm in reverse time order.

[0105] During the application process, the data can be divided into time series based on actual experience. For example, every 15 minutes is taken as a time series, and the garbage incineration status and incinerator operation data of several time series are input into the preset LSTM model to obtain the predicted flue gas pollutant concentration and the predicted main combustion section temperature field distribution.

[0106] Step 103: Input the incinerator environmental monitoring data, the predicted flue gas pollutant concentration and the predicted main combustion section temperature field distribution into a reinforcement learning control model to obtain the incineration control parameters.

[0107] Reinforcement learning refers to a method in which an agent learns strategies to maximize rewards or achieve specific goals during its interaction with the environment. Reinforcement learning does not require any data to be given in advance, but rather obtains learning information and updates model parameters by receiving rewards or feedback from the environment for actions. In this application, the reinforcement learning control model is used to continuously learn according to the incineration results during the incineration process to adapt to the complex waste incineration process and obtain the optimal incineration control parameters.

[0108] Please refer to Figure 6, the reinforcement learning control model in one embodiment of the present application includes a Dyna-Q algorithm model and a DDPG algorithm model. The DDPG algorithm is developed on the Deterministic Policy Gradient algorithm. It contains a policy network for generating actions, a value network for judging the quality of actions, and draws on the successful experience of DQN, using a sample pool and a fixed target network. It is an Actor-Critic method combined with a deep network. It not only has the characteristics of single-step update of the Actor-Critic algorithm, but also improves the convergence of the algorithm by drawing on the successful experience of the dual network structure and experience playback of the DQN algorithm. Compared with DQN, an algorithm based on value functions, DDPG is suitable for tasks that output continuous actions and can be used in scenarios with explosive action dimensions. The Dyna-Q algorithm is a Model-based algorithm in reinforcement learning. The Dyna-Q algorithm uses neural networks to fit environmental changes to provide environmental experience for complex environments that are not convenient for direct deployment and training of reinforcement learning algorithms.

[0109] In this embodiment, the DDPG algorithm model receives the predicted flue gas pollutant concentration and the predicted main combustion section temperature field distribution output by LSTM, finds the optimal action combination, and outputs continuous combustion control parameters, which may include feeding speed, combustion grate operation cycle, and combustion air distribution ratio. The Dyna-Q algorithm model receives the incinerator environmental monitoring data, the predicted flue gas pollutant concentration, and the predicted main combustion section temperature field distribution, learns the next state to be entered after outputting different combustion control parameters under different states, and provides it to the DDPG algorithm model for learning. The state here includes the flue gas pollutant concentration and the main combustion section temperature field distribution. The incinerator environmental monitoring data to a certain extent represents the combustion results obtained by the incinerator when burning garbage according to the incineration control parameters. Therefore, in fact, the Dyna-Q algorithm model simulates a virtual environmental model based on the data generated by the interaction between the DDPG algorithm model and the actual environment. It is used to simulate the combustion results that will be produced by the incinerator under different states and different incineration control parameters, and use the simulated combustion results as feedback for the DDPG algorithm, which makes the algorithm parameter training more stable while reducing the number of interactions between the DDPG algorithm and the actual environment, saving the algorithm training cost.

[0110] Step 104: Send the incineration control parameters to the incinerator so that the incinerator makes corresponding adjustments.

[0111] According to the garbage incineration control method of the above embodiment, the garbage incineration state is obtained by a deep learning-based method to realize real-time automatic monitoring of the garbage incineration state, which is beneficial to improving the accuracy of garbage incineration state identification, thereby improving the stability of garbage incineration, bringing higher marginal economic benefits to garbage incineration power plants, and at the same time liberating manpower and improving the working environment of workers; at the same time, the incinerator environmental monitoring data and the incinerator operation data are obtained, and the LSTM model is used for simulation prediction according to the garbage incineration state and the incinerator operation data to obtain the predicted flue gas pollutant concentration and the main combustion section temperature field distribution, and then the incinerator environmental monitoring data, the predicted flue gas pollutant concentration and the predicted main combustion section temperature field distribution are input into a reinforcement learning control model to obtain the incineration control parameters, and the incineration control parameters are sent to the incinerator to make the incinerator make corresponding adjustments, so that the garbage incineration control parameters are optimized, can adapt to complex garbage incineration processes, realize the automatic control of the garbage incineration process, improve the stability of garbage incineration, and at the same time liberate manpower and reduce labor costs.

[0112] This document is described with reference to various exemplary embodiments. However, those skilled in the art will recognize that changes and modifications may be made to the exemplary embodiments without departing from the scope of this document. For example, various operating steps and components for performing the operating steps may be implemented in different ways (e.g., one or more steps may be deleted, modified, or incorporated into other steps) depending on the specific application or considering any number of cost functions associated with the operation of the system.

[0113] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. In addition, as understood by those skilled in the art, the principles of this article can be reflected in a computer program product on a computer-readable storage medium, which is pre-installed with a computer-readable program code. Any tangible, non-temporary computer-readable storage medium can be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD to ROM, DVD, Blu Ray disks, etc.), flash memory and / or the like. These computer program instructions can be loaded onto a general-purpose computer, a special-purpose computer or other programmable data processing device to form a machine, so that these instructions executed on a computer or other programmable data processing device can generate a device that implements a specified function. These computer program instructions can also be stored in a computer-readable memory, which can instruct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory can form a manufactured product, including an implementation device that implements a specified function. Computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operating steps are performed on a computer or other programmable device to generate a computer-implemented process, so that the instructions executed on a computer or other programmable device can provide steps for implementing a specified function.

[0114] Although the principles of this invention have been shown in various embodiments, many modifications of structures, arrangements, proportions, elements, materials and components particularly suitable for specific environments and operational requirements can be used without departing from the principles and scope of this invention. The above modifications and other changes or amendments will be included in the scope of this invention.

[0115] The foregoing specific description has been described with reference to various embodiments. However, those skilled in the art will recognize that various modifications and changes can be made without departing from the scope of the present disclosure. Therefore, the consideration of the present disclosure will be illustrative rather than restrictive, and all these modifications will be included in its scope. Similarly, the advantages, other advantages and solutions to the problems of various embodiments have been described above. However, the benefits, advantages, solutions to the problems and any elements that can produce these, or make them more clear, should not be interpreted as critical, necessary or necessary. The term "include" and any other variants used in this article are all non-exclusive inclusions, so that the process, method, article or device including the list of elements not only includes these elements, but also includes other elements that are not explicitly listed or do not belong to the process, method, system, article or device. In addition, the term "coupled" and any other variants used in this article refer to physical connections, electrical connections, magnetic connections, optical connections, communication connections, functional connections and / or any other connections.

[0116] Those skilled in the art will appreciate that many changes may be made to the details of the above-described embodiments without departing from the basic principles of the invention. Therefore, the scope of the present invention should be determined solely by the claims.

Claims

1. A garbage incineration control method based on deep learning, characterized in that: include: Obtain images of waste incineration; Inputting the garbage incineration image into a garbage incineration state recognition model to obtain the garbage incineration state; Obtaining incinerator environmental monitoring data and incinerator operation data, and predicting flue gas pollutant concentration and main combustion section temperature field distribution according to the garbage incineration status and incinerator operation data; Inputting the incinerator environmental monitoring data, predicted flue gas pollutant concentration and predicted main combustion section temperature field distribution into a reinforcement learning control model to obtain incineration control parameters; The incineration control parameters are sent to the incinerator so that the incinerator makes corresponding adjustments; the reinforcement learning control model includes a Dyna-Q algorithm model and a DDPG algorithm model; the DDPG algorithm model is used to calculate the incineration control parameters according to the predicted flue gas pollutant concentration and the predicted temperature field distribution of the main combustion section; the Dyna-Q algorithm model is used to learn the next state to be entered after outputting different incineration control parameters under different states according to the incinerator environmental monitoring data, the predicted flue gas pollutant concentration and the predicted temperature field distribution of the main combustion section, and provide it to the DDPG algorithm model for learning, the state includes the flue gas pollutant concentration and the temperature field distribution of the main combustion section; The waste incineration state identification model is obtained by the following steps: Calculating the burning state parameters of the garbage incineration sample images in the preset garbage incineration sample image set, wherein the burning state parameters include one or more of the image average grayscale, flame effective area rate, high temperature effective area rate, flame high temperature rate, flame centroid, flame centroid offset distance, high temperature area circularity, flame effective area variance per unit time, and image grayscale mean variance per unit time; Classifying the garbage incineration sample images according to the predetermined incineration state categories based on the incineration state parameters; A neural network model is constructed, the garbage incineration sample image set is used as training data, the categories of the garbage incineration sample images are used as data labels, and the neural network model is trained to obtain the garbage incineration state recognition model.

2. The method according to claim 1, characterized in that The classifying the garbage incineration sample images according to the predetermined incineration state categories based on the incineration state parameters includes: The EM algorithm is used to perform cluster analysis on the incineration state parameters to obtain three types of indicators of the garbage incineration sample image, namely flame shape, flame temperature and flame flicker. According to the three types of indicators, the garbage incineration sample image is classified according to the predetermined incineration state category.

3. The method according to claim 2, characterized in that The method of classifying the garbage incineration sample images according to the three indicators of flame shape, flame temperature and flame flicker and the predetermined incineration state category includes: The three indicators of flame shape, flame temperature and flame flicker of the garbage incineration sample image are input into a preset expert system to obtain the incineration state category to which the garbage incineration sample image belongs, wherein the expert system has expert experience in classifying the garbage incineration sample image according to the predetermined incineration state category based on flame shape, flame temperature and flame flicker.

4. The method according to claim 1, characterized in that The method of using the garbage incineration sample image set as training data and the categories of the garbage incineration sample images as data labels to train the neural network model includes: The waste incineration sample image set is divided into a training set, a validation set and a test set in a certain ratio, the training set is used to train the neural network model, the validation set is used to perform hyperparameter tuning on the trained neural network model, and the test set is used to verify the accuracy of the trained neural network model.

5. The method according to claim 1, characterized in that The burning state categories include uniform burning, incomplete burning, lateral bias burning and longitudinal bias burning.

6. The method according to claim 1, characterized in that Also includes: When the garbage incineration states obtained continuously within a certain period of time are all consistent, it is determined that the garbage incineration state within the time period is the consistent garbage incineration state.

7. A computer-readable storage medium, characterized in that: The method comprises a program which can be executed by a processor to implement the method according to any one of claims 1 to 6.

Images