1. An artificial intelligence control method for a waste incineration process is characterized by comprising the following steps:

acquiring incinerator environment monitoring data;

acquiring a waste incineration state and incinerator operation data, and predicting the concentration of smoke pollutants and the temperature field distribution of a main combustion section according to the waste incineration state and the incinerator operation data;

inputting the incinerator environment monitoring data, the predicted concentration of the smoke pollutants and the predicted temperature field distribution of the main combustion section into a reinforcement learning control model to obtain incineration control parameters;

and sending the incineration control parameters to the incinerator to enable the incinerator to make corresponding adjustment.

2. The method of claim 1, wherein predicting flue gas pollutant concentrations and main combustion section temperature field distributions based on the waste incineration conditions and incinerator operating data comprises:

and inputting the garbage incineration state and the incinerator operation data into a preset LSTM model to obtain the predicted smoke pollutant concentration and the predicted main combustion section temperature field distribution.

3. The method of claim 1, wherein the reinforcement learning control model comprises a Dyna-Q algorithm model and a DDPG algorithm model;

the DDPG algorithm model is used for calculating to obtain the incineration control parameters according to the predicted smoke pollutant concentration and the predicted temperature field distribution of the main combustion section; the Dyna-Q algorithm model is used for learning to obtain the next state which is to be entered after different incineration control parameters are output in different states according to the incinerator environment monitoring data, the predicted flue gas pollutant concentration and the predicted main combustion section temperature field distribution, and providing the next state to the DDPG algorithm model for learning, wherein the states comprise the flue gas pollutant concentration and the main combustion section temperature field distribution.

4. The method of claim 1, wherein the incinerator environment monitoring data comprises: one or more of main steam quantity at the outlet of the boiler, CO concentration in flue gas, grate air flow of a drying section, flue gas temperature of a primary combustion chamber, flue gas temperature of a secondary combustion chamber, inlet air flow of a combustion section, primary air quantity, speed of a grate at the first combustion section, oxygen concentration of outlet flue gas, primary air pressure and temperature, negative pressure of a hearth and area of a grate segment.

5. The method of claim 1, wherein the incinerator operating data comprises one or more of a dry section temperature field distribution, a bed thickness, a main firing section current temperature field distribution, a main firing section air volume, a grate sliding speed, a grate turnover frequency.

6. The method of claim 1, wherein the incineration control parameters include a feed rate, an incineration grate operating cycle, and a combustion air distribution ratio.

7. The method of claim 1, wherein the waste incineration status is obtained by:

acquiring a waste incineration image;

inputting the waste incineration image into a waste incineration state identification model to obtain a waste incineration state;

the waste incineration state identification model is obtained through the following steps:

calculating the incineration state parameters of the waste incineration sample images in the preset waste incineration sample image set;

classifying the waste incineration sample images according to the incineration state parameters and preset incineration state categories;

and constructing a neural network model, taking the waste incineration sample image set as training data, taking the category of the waste incineration sample image as a data label, and training the neural network model to obtain the waste incineration state identification model.

8. The method of claim 7, wherein the classifying the waste incineration sample image according to the incineration status parameter by a predetermined incineration status category comprises:

carrying out cluster analysis on the incineration state parameters by adopting an EM algorithm to obtain three indexes of flame shape, flame temperature and flame flicker of the waste incineration sample image, and classifying the waste incineration sample image according to the three indexes of flame shape, flame temperature and flame flicker and according to a preset incineration state category; the incineration state categories include uniform combustion, insufficient combustion, transverse partial combustion and longitudinal partial combustion.

9. The method of claim 8, wherein the classifying the waste incineration sample image according to the three indexes of flame shape, flame temperature and flame flicker according to the predetermined incineration state category comprises:

inputting three indexes of flame shape, flame temperature and flame flicker of the waste incineration sample image into a preset expert system to obtain the incineration state category of the waste incineration sample image, wherein the expert system stores expert experience for classifying the waste incineration sample image according to the preset incineration state category according to the flame shape, the flame temperature and the flame flicker.

10. A computer-readable storage medium, characterized in that the medium has stored thereon a program which is executable by a processor to implement the method according to any one of claims 1 to 9.

Background

In recent years, the number of waste incineration power generation projects in our country continues to increase, and under the background that the demand for waste incineration power generation is gradually saturated and the price of electricity subsidies and moves away from the slope, how to operate the waste incineration power generation projects with high quality and high efficiency through technical means becomes an urgent problem to be solved.

At present, the waste incineration technology in China has the problem of unstable incineration. The waste incineration is a process of carrying out combustion reaction on oxygen in air and combustible components in the waste to convert the combustible components into inorganic residues, and simultaneously recovering heat energy released during combustion and converting the heat energy into electric energy. However, the garbage in China has the factors of large component change, high moisture content, unstable calorific value and the like, so that the garbage incineration process is more complicated, and the incineration state is difficult to stabilize.

An ACC (Automatic Combustion Control) Control system for waste incineration adopted at the present stage has a short board which cannot adapt to nonlinear complex working conditions, and stable Combustion of waste is difficult to realize. The ACC control system mainly generates stable steam flow by changing air quantity and adjusting grate speed, and simultaneously controls the range of oxygen content in flue gas to maintain stable combustion. However, the control of the air volume and the control of the grate speed are mutually influenced, and each control object is not a simple single-input single-output single-loop regulating system but a complex control system with multiple inputs and multiple outputs. In addition, the PID control method adopted in the ACC control system mainly comprises the steps that a worker carries out manual adjustment by monitoring boiler data, the control requirement can be met to a certain extent only by depending on the experience of a skilled worker, and the labor is consumed.

Disclosure of Invention

The application provides an artificial intelligence control method of a waste incineration process, which optimizes waste incineration control parameters and realizes automatic control of the waste incineration process by establishing a reinforcement learning control model so as to solve the problems that the conventional waste incineration control method is difficult to realize stable combustion of waste and consumes manpower.

According to a first aspect, an embodiment provides an artificial intelligence control method for a waste incineration process, comprising:

acquiring incinerator environment monitoring data;

acquiring a waste incineration state and incinerator operation data, and predicting the concentration of smoke pollutants and the temperature field distribution of a main combustion section according to the waste incineration state and the incinerator operation data;

inputting the incinerator environment monitoring data, the predicted concentration of the smoke pollutants and the predicted temperature field distribution of the main combustion section into a reinforcement learning control model to obtain incineration control parameters;

and sending the incineration control parameters to the incinerator to enable the incinerator to make corresponding adjustment.

In one embodiment, the predicting the flue gas pollutant concentration and the main combustion section temperature field distribution according to the waste incineration state and the incinerator operation data comprises:

and inputting the garbage incineration state and the incinerator operation data into a preset LSTM model to obtain the predicted smoke pollutant concentration and the predicted main combustion section temperature field distribution.

In one embodiment, the reinforcement learning control model comprises a Dyna-Q algorithm model and a DDPG algorithm model; the DDPG algorithm model is used for calculating to obtain the incineration control parameters according to the predicted smoke pollutant concentration and the predicted temperature field distribution of the main combustion section; the Dyna-Q algorithm model is used for learning to obtain the next state which is to be entered after different incineration control parameters are output in different states according to the incinerator environment monitoring data, the predicted flue gas pollutant concentration and the predicted main combustion section temperature field distribution, and providing the next state to the DDPG algorithm model for learning, wherein the states comprise the flue gas pollutant concentration and the main combustion section temperature field distribution.

In one embodiment, the incinerator environment monitoring data comprises: one or more of main steam quantity at the outlet of the boiler, CO concentration in flue gas, grate air flow of a drying section, flue gas temperature of a primary combustion chamber, flue gas temperature of a secondary combustion chamber, inlet air flow of a combustion section, primary air quantity, speed of a grate at the first combustion section, oxygen concentration of outlet flue gas, primary air pressure and temperature, negative pressure of a hearth and area of a grate segment.

In one embodiment, the incinerator operating data comprises one or more of drying section temperature field distribution, bed thickness, current temperature field distribution of the main combustion section, air volume of the main combustion section, grate sliding speed, and grate turning frequency.

In one embodiment, the incineration control parameters include a feed rate, an incineration grate operating period, and a combustion air distribution ratio.

In one embodiment, the waste incineration status is obtained by:

acquiring a waste incineration image;

inputting the waste incineration image into a waste incineration state identification model to obtain a waste incineration state;

the waste incineration state identification model is obtained through the following steps:

calculating the incineration state parameters of the waste incineration sample images in the preset waste incineration sample image set;

classifying the waste incineration sample images according to the incineration state parameters and preset incineration state categories;

and constructing a neural network model, taking the waste incineration sample image set as training data, taking the category of the waste incineration sample image as a data label, and training the neural network model to obtain the waste incineration state identification model.

In one embodiment, the classifying the waste incineration sample image according to the incineration state parameter and a predetermined incineration state category includes:

carrying out cluster analysis on the incineration state parameters by adopting an EM algorithm to obtain three indexes of flame shape, flame temperature and flame flicker of the waste incineration sample image, and classifying the waste incineration sample image according to the three indexes of flame shape, flame temperature and flame flicker and according to a preset incineration state category; the incineration state categories include uniform combustion, insufficient combustion, transverse partial combustion and longitudinal partial combustion.

In one embodiment, the classifying the waste incineration sample image according to three indexes, namely flame shape, flame temperature and flame flicker, and according to a predetermined incineration state class comprises:

inputting three indexes of flame shape, flame temperature and flame flicker of the waste incineration sample image into a preset expert system to obtain the incineration state category of the waste incineration sample image, wherein the expert system stores expert experience for classifying the waste incineration sample image according to the preset incineration state category according to the flame shape, the flame temperature and the flame flicker.

According to a second aspect, an embodiment provides a computer readable storage medium having a program stored thereon, the program being executable by a processor to implement the method of the first aspect described above.

According to the artificial intelligence control method and the computer readable storage medium of the waste incineration process of the embodiment, the incinerator environment monitoring data, the waste incineration state parameters and the incinerator operation data are obtained, the smoke pollutant concentration and the temperature field distribution of the main combustion section are predicted according to the waste incineration state and the incinerator operation data, then the incinerator environment monitoring data, the predicted smoke pollutant concentration and the predicted temperature field distribution of the main combustion section are input into an enhanced learning control model to obtain the incineration control parameters, the incineration control parameters are sent to the incinerator, the incinerator is correspondingly adjusted, the waste incineration control parameters are optimized, the method and the system can adapt to the complex waste incineration process, the automatic control of the waste incineration process is achieved, the stability of waste incineration is improved, meanwhile, manpower is released, and the labor cost is reduced.

Drawings

FIG. 1 is a flowchart of an artificial intelligence control method of a waste incineration process according to an embodiment;

FIG. 2 is a graph showing the relationship between environmental monitoring data of an incinerator and various control amounts;

FIG. 3 is a flowchart of an embodiment of a garbage incineration status identification method based on deep learning;

FIG. 4 is a flowchart of a method for constructing a waste incineration state identification model according to an embodiment;

FIG. 5 is a schematic structural diagram of a convolutional neural network for constructing a garbage incineration state identification model according to an embodiment;

FIG. 6 is a block diagram of an algorithm of an artificial intelligence control method for a garbage incineration process according to an embodiment.

Detailed Description

The present invention will be described in further detail with reference to the following detailed description and accompanying drawings. Wherein like elements in different embodiments are numbered with like associated elements. In the following description, numerous details are set forth in order to provide a better understanding of the present application. However, those skilled in the art will readily recognize that some of the features may be omitted or replaced with other elements, materials, methods in different instances. In some instances, certain operations related to the present application have not been shown or described in detail in order to avoid obscuring the core of the present application from excessive description, and it is not necessary for those skilled in the art to describe these operations in detail, so that they may be fully understood from the description in the specification and the general knowledge in the art.

Furthermore, the features, operations, or characteristics described in the specification may be combined in any suitable manner to form various embodiments. Also, the various steps or actions in the method descriptions may be transposed or transposed in order, as will be apparent to one of ordinary skill in the art. Thus, the various sequences in the specification and drawings are for the purpose of describing certain embodiments only and are not intended to imply a required sequence unless otherwise indicated where such sequence must be followed.

The numbering of the components as such, e.g., "first", "second", etc., is used herein only to distinguish the objects as described, and does not have any sequential or technical meaning. The term "connected" and "coupled" when used in this application, unless otherwise indicated, includes both direct and indirect connections (couplings).

Referring to fig. 1, an artificial intelligence control method for a waste incineration process in an embodiment of the present application includes steps 101 to 104, which are described in detail below.

Step 101: acquiring incinerator environment monitoring data, a garbage incineration state and incinerator operation data.

As shown in FIG. 2, the environmental monitoring data of the incinerator includes the main steam flow at the outlet of the boiler, the CO concentration in the flue gas, the grate air flow at the drying section, the flue gas temperature of the secondary combustion chamber, the inlet air flow of the combustion section, the primary air flow, the grate speed at the first combustion section, the oxygen concentration of the flue gas at the outlet, the primary air pressure and temperature, the negative pressure of the furnace and the area of the grate segments, etc., and the specific data can be selected according to the requirements, for example, the main steam flow at the outlet of the boiler, the CO concentration in the flue gas and the grate air flow at the drying section need to be monitored when the feeder speed needs to be controlled, the flue gas temperature of the secondary combustion chamber, the main steam flow at the outlet of the boiler and the grate air flow at the inlet of the combustion section need to be monitored when the grate speed of the combustion section needs to be controlled, and therefore the oxygen concentration of the flue gas at the outlet, the CO concentration in the flue gas, the CO concentration of the flue gas, the secondary air flow, etc., need to be monitored, The temperature of the flue gas of the primary combustion chamber, the main steam quantity at the outlet of the boiler, and the like. And monitoring the environment of the incinerator in real time, and continuously updating monitoring data.

The operation data of the incinerator comprises temperature field distribution of a drying section, material layer thickness, current temperature field distribution of a main combustion section, air quantity of the main combustion section, sliding speed of a fire grate, turning frequency of the fire grate and the like, and one or more data can be selected for use according to needs.

For the waste incineration state, the conventional method relies on manual work for identification, relies on the subjective experience of workers, has unsatisfactory accuracy, and is easy to cause unstable incineration effect, the application provides a waste incineration state identification method based on deep learning, please refer to fig. 3, and the waste incineration state identification method in one embodiment comprises the following steps:

step 301: and acquiring a waste incineration image. The garbage incineration image can be obtained by shooting the garbage incineration flame in the whole furnace chamber through a Charge Coupled Device (CCD) camera.

Step 302: and inputting the waste incineration image into a waste incineration state identification model to obtain a waste incineration state. In order to make the identification result more accurate, the waste incineration state output for a period of time can be monitored, and when the waste incineration states obtained continuously within a certain period of time are all consistent, the waste incineration state within the period of time is judged to be the consistent waste incineration state. The waste incineration state identification model in the application can be constructed through a neural network model, and a specific construction process refers to fig. 4, which includes steps 311-313, which are described in detail below.

Step 311: and calculating the incineration state parameters of the waste incineration sample images in the preset waste incineration sample image set. The parameters of the incineration state include one or more of average image gray scale, effective flame area rate, effective high-temperature area rate, high-temperature flame rate, flame centroid offset distance, circularity of high-temperature region, effective flame area variance per unit time, and mean image gray scale variance per unit time, and the calculation methods of these parameters are described below respectively.

Since the incineration state parameters are calculated according to the gray-scale image, the obtained R, G, B three-channel color image needs to be converted into a single-channel gray-scale image firstly, and the calculation formula is as follows:

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

where G (x, y) is the grayscale image grayscale value at (x, y), R (x, y) is the pixel value in the original color image red channel at (x, y), G (x, y) is the pixel value in the original color image green channel at (x, y), and B (x, y) is the pixel value in the original color image blue channel at (x, y).

(1) Average gray level of image

The image mean gray scale can be used to estimate the overall temperature level of the combustion flame. The average gray value of the image is obtained by traversing all pixel points of the whole image, accumulating and obtaining the gray value sum of all the pixel points, and the average gray value is the quotient of the gray value sum and the number of the pixel points, and the steps are as follows:

1) traversing pixels of an image to be calculated, and accumulating and summing the gray value of each pixel point to obtain a gray value sum;

2) calculating the total number n of pixel points of the image;

3) calculating the average gray value g of the image_mean＝sum/n。

(2) Effective area ratio of flame

The effective area ratio of the flame is used for evaluating the area of the flame filled in the image space in the gray level image and representing the coverage degree of the combustion flame in the monitoring area. The calculation formula is as follows:

wherein S_vG is the desired effective area ratio of the flame_thJudging the gray threshold of the flame point, regarding the point larger than the threshold as the flame point, i is the number of rows of pixel points, j is the number of columns of pixel points, L (x) is a step function, and is defined as

(3) High temperature effective area ratio

The high-temperature effective area rate mainly reflects the area of the whole image space occupied by the region with higher combustion flame temperature, and the higher the temperature is, the higher the gray value of the image is. The calculation formula is as follows:

wherein S_hIn order to obtain the desired effective area ratio of the flame,the gray level threshold value of the high-temperature flame point is judged, and the point which is larger than the threshold value is regarded as the high-temperature flame point.

(4) High temperature rate

The high temperature rate mainly reflects the ratio of the area with higher combustion flame temperature to the effective area of the flame, and reflects the fullness of the high temperature area, and the higher the value of the fullness, the larger the area with higher temperature 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_hAnd the effective area ratio S of the flame_vThe ratio of (a) to (b).

(5) Center of mass of flame

The flame centroid reflects the spatial position information of combustion, can comprehensively and accurately reflect whether the flame in the area is at a reasonable position or not, and is used for diagnosing the combustion problem. The calculation formula is as follows:

wherein the content of the first and second substances,is the abscissa of the center of mass of the flame,is the flame centroid ordinate.

(6) Offset distance of flame centroid

The flame position is constantly moving as old combustibles burn out and new combustibles are ignited, but the movement is non-hopping and is generally continuous, i.e. the flame meets the relative stability of the motion, and the overall movement of the flame can be described by calculating the flame centroid offset distance. The offset distance d of the flame centroid refers to the flame centroidReference centroid (x) tangent to four corners of flame image_s,y_s) The offset distance is calculated by the formula:

(7) high temperature zone circularity

The high temperature zone roundness characterizes the geometric profile of the combustion flame, which reflects the quality of the aerodynamic field. The length of the contour line of the high-temperature effective area is taken as the perimeter L^hAnd the area of the region surrounded by the contour line is used as the projection area of the flame image high-temperature region. Assuming a circle having the same area as the projected area of the flame image high temperature region, called equivalent circle, the diameter of the circle is equivalent circle diameter D, calculating equivalent circumference L according to the equivalent circle diameter^dDivided by the perimeter L of the flame image high temperature zone^hThe obtained value was taken as circularityThe calculation formula is as follows:

L^d＝πD

(8) variance of effective area of flame per unit time

The variance of the effective flame area in unit time reflects the change condition of the flame area in unit time, and the calculation formula is as follows:

wherein D_sFor the variance of the effective area of the flame per unit time, S_viFor the ith calculated flame effective area ratio, S_vmeanThe average value of the flame effective area rate in the statistical time period is obtained, and n is the total statistical sample size, namely the measurement times.

(9) Mean variance of image gray levels per unit time

The mean variance of the gray level of the image in unit time reflects the fluctuation condition of the flame temperature, and the calculation formula is as follows:

wherein D_gFor the mean variance of the gray level of the image per unit time, g_meaniFor the image gray level mean calculated at the i-th time,the average value of the image gray level mean values in the statistical time period is obtained.

Step 312: and classifying the waste incineration sample images according to the incineration state parameters and the preset incineration state types. The incineration state can be divided into four types, namely even combustion, insufficient combustion, transverse partial combustion and longitudinal partial combustion.

In the embodiment, an EM (Expectation-maximization) algorithm and an expert system are adopted to classify the waste incineration sample images. Firstly, carrying out cluster analysis on the incineration state parameters by using an EM algorithm to obtain three indexes of flame shape, flame temperature and flame flicker of the waste incineration sample image, wherein the three indexes of the flame shape, the flame temperature and the flame flicker are the mass centers of the three cluster clusters. The EM algorithm is an algorithm that finds the maximum likelihood estimate or maximum a posteriori estimate of the parameters in a probabilistic model, where the probabilistic model depends on unobservable hidden variables, which are the centroids of the clusters: three indexes of flame shape, flame temperature and flame flicker. The EM algorithm mainly includes two steps: the first step is to calculate expectation (step E), and the maximum likelihood estimated value of the implicit variable (namely the centroid) is calculated by utilizing the existing estimated value of the implicit variable; and the second step is maximization (M steps), the nearest mass center of each sample image is obtained through calculation, and the sample images are clustered to the nearest mass center, so that the parameter estimation value of the probability model is obtained. And (3) the parameter estimation values found in the M step are used for the calculation of the next E step, the process is continuously and alternately carried out until the mass center is not changed, so that the clustering with 3 mass centers is completed, and finally three indexes of the flame shape, the flame temperature and the flame flicker are extracted.

After three indexes of flame shape, flame temperature and flame flicker are obtained, the three indexes are input into a preset expert system to obtain the incineration state type of the waste incineration sample image, the expert system stores expert experience for classifying the waste incineration sample image according to the preset incineration state type according to the flame shape, the flame temperature and the flame flicker, and the expert system can classify the sample image according to the input flame shape, the flame temperature and the flame flicker index according to the expert experience. Therefore, the images of the waste incineration samples have respective class labels, so that the automatic construction of a training sample set is realized, and a large sample database is constructed to improve the accuracy of waste incineration state identification.

Step 313: and constructing a neural network model, taking the waste incineration sample image set as training data, taking the category of the waste incineration sample image as a data label, and training the neural network model to obtain a waste incineration state identification model.

The constructed neural network model may be a convolutional neural network model, and referring to fig. 5, the convolutional neural network includes convolutional layers, batch normalization layers, pooling layers, fully-connected networks, and the like. The waste incineration sample image is convolved, batch standardized and pooled to obtain a feature vector, and the feature vector is input into a full-connection network to obtain 4 types of probabilities for distinguishing. The convolution is a method for effectively extracting image features, and generally a square convolution kernel is used to slide on an input image according to a specified step length, traverse each pixel point in the input image, perform convolution calculation, extract features in the image, and obtain a feature map.

The batch standardization is to standardize a small batch of data output by each layer of the network, and the realization formula is as follows:

whereinRepresenting the ith pixel point in a characteristic diagram output by the kth convolution kernel before batch normalization,representing the average value of all pixel points in the batch characteristic graph output by the kth convolution kernel before batch normalization,representing the standard deviation of all pixel points in the batch characteristic diagram output by the kth convolution kernel before batch normalization, wherein batch is an integer larger than 0,representative pairAnd carrying out batch standardization on the pixel points.

Pooling has the effect of reducing the number of features (dimensionality reduction), where maximal pooling can extract picture texture and mean pooling can preserve background features.

The waste incineration sample image needs to be preprocessed before being input into a neural network, and standard deviation standardization and whitening are mainly performed on the waste incineration sample image in sequence. The purpose of the standardization of the data standard deviation is to enable the data to fall into a specific interval through the scaling of the data, so that the problem that the data with larger magnitude of order is too heavy in the algorithm and the data with smaller magnitude of order is too small in weight can be effectively avoided; whitening is used to reduce the correlation between features and improve the stability of the algorithm.

The method comprises the steps of dividing a waste incineration sample image set into a training set, a verification set and a test set according to a certain proportion during training, firstly training a neural network model by using the training set, sending waste incineration sample images in the training set into the neural network according to batches to obtain a prediction result of a waste incineration state, then comparing the prediction result with a corresponding data label to calculate a loss value, carrying out iterative correction on parameters of the neural network through a back propagation algorithm according to the loss value until a preset stop condition is met, wherein the stop condition can be that the prediction error is smaller than a threshold value, the training time meets a requirement or the prediction result is converged, and the like, and meanwhile estimating the generalization error of the neural network model during or after training by using the verification set to update the hyper-parameters of the neural network. And after the training is finished, sending the garbage incineration sample images in the test set into a neural network to obtain a predicted garbage incineration state, and then comparing the predicted garbage incineration state with the corresponding data labels to verify the accuracy of the trained neural network model. And obtaining a final waste incineration state identification model after a plurality of times of training and super-parameter tuning.

The convolutional neural network for identifying the waste incineration state can be established by using a Tensorflow framework and a keras library, and four modules are mainly used, namely a model, a model. The model module is mainly responsible for defining a network structure, including the number of layers of convolution layers, the number and the size of convolution kernels, whether batch standardization processing is carried out or not, selection of an activation function and a pooling method, the number of layers of a full-connection network, the number of neurons, the activation function and the like; the module is responsible for configuring a training method, and comprises defining an optimizer, a loss function, a model evaluation index and the like; the fit module is responsible for training parameters, including defining a training set and a verification set, the number of samples input into the neural network each time, the number of iteration rounds and the like; summary module is responsible for parameter extraction and predictive effect visualization. In practical application, each module adopts different setting methods according to actual conditions.

The application provides a waste incineration state identification method based on deep learning utilizes neural network model to establish waste incineration state identification model, can obtain the waste incineration state with waste incineration image input waste incineration state identification model in to realize the real-time automatic monitoring of waste incineration state, be favorable to improving the accuracy of waste incineration state discernment, liberate the manpower simultaneously, reduce the human cost, improve the operational environment of labourer.

The following description proceeds to steps 102 to 104.

Step 102: and predicting the concentration of the smoke pollutants and the temperature field distribution of the main combustion section according to the waste incineration state and the incinerator operation data. The distribution of the concentration of the smoke pollutants and the temperature field of the main combustion section of the next time sequence is predicted, so that the adjustment in advance is facilitated, the change of the incineration process is adapted, and the incineration stability is improved.

In this embodiment, an LSTM (Long Short-term memory) model is used to predict the flue gas pollutant concentration and the main combustion section temperature field distribution of the next time sequence. The LSTM is a special RNN (Recurrent neural network) neural network algorithm, can solve the problems of gradient elimination and gradient explosion in the long sequence training process, and is suitable for trend prediction of long sequences. Compared with the common neural network, the circular neural network structure is more consistent with the actual construction of the biological neural network. The LSTM may also receive its own information while receiving other neuron information, thereby forming a network structure with loops to produce a "memory" effect. Like other neural networks, the recurrent neural network can be parameter trained by a back propagation algorithm in reverse order of time.

In the application process, the data can be divided into time sequences according to actual experience, for example, every 15 minutes is used as a time sequence, the garbage incineration states and the incinerator operation data of a plurality of time sequences are input into a preset LSTM model, and the predicted smoke pollutant concentration and the predicted main combustion section temperature field distribution are obtained.

Step 103: and inputting the incinerator environment monitoring data, the predicted concentration of the smoke pollutants and the predicted temperature field distribution of the main combustion section into a reinforcement learning control model to obtain incineration control parameters.

Reinforcement learning refers to a method in which an agent (agent) achieves the maximum return or achieves a specific goal through learning a strategy in an interaction process with an environment, and reinforcement learning does not require any data to be given in advance, but obtains learning information and updates model parameters by receiving rewards or feedback of the environment to actions. Use in this application and strengthen learning control model and constantly study according to burning the result burning in-process to adapt to complicated msw incineration process, obtain the optimal control parameter that burns.

Referring to fig. 6, the reinforcement learning control model in an embodiment of the present application includes a Dyna-Q algorithm model and a DDPG algorithm model. The DDPG algorithm was developed on a Deterministic Gradient (Deterministic Policy Gradient) algorithm. The method comprises a strategy network used for generating actions, a value network used for judging the quality of the actions, and drawing the success experience of DQN, and a sample pool and a fixed target network are used, so that the method is an Actor-criticic method combined with a deep network. The method has the characteristic of single-step updating of the Actor-Critic algorithm, and improves the convergence of the algorithm by using the success experience of the DQN algorithm double-network structure and experience playback. Compared with an algorithm based on a value function, such as DQN, the DDPG is suitable for the task of outputting continuous actions and can be used for scenes of action dimension explosion. The Dyna-Q algorithm is a Model-based algorithm (Model-based algorithm) in reinforcement learning. The Dyna-Q algorithm provides environmental experience for environments that are complex and inconvenient to directly deploy a training reinforcement learning algorithm by fitting environmental changes with a neural network.

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 the LSTM, finds an optimal action combination, and outputs continuous incineration control parameters, which may include a feeding speed, an incineration grate operating period, a combustion air distribution ratio, and the like. The Dyna-Q algorithm model receives incinerator environment monitoring data, predicted smoke pollutant concentration and predicted main combustion section temperature field distribution, learns the next state which is to be entered after different incineration control parameters are output in different states, and provides the next state to the DDPG algorithm model for learning, wherein the states comprise the smoke pollutant concentration and the main combustion section temperature field distribution. The incinerator environment monitoring data represents a combustion result obtained by incinerating garbage by the incinerator according to the incineration control parameters to a certain extent, therefore, in practice, the Dyna-Q algorithm model simulates a virtual environment model according to data generated by interaction of the DDPG algorithm model and the actual environment, the virtual environment model is used for simulating combustion results which will be generated by the incinerator under different states and different incineration control parameters, and the simulated combustion results are used as feedback of the DDPG algorithm, so that algorithm parameter training is more stable, interaction times of the DDPG algorithm and the actual environment are reduced, and algorithm training cost is saved.

Step 104: and sending the incineration control parameters to the incinerator to enable the incinerator to make corresponding adjustment.

According to the artificial intelligence control method for the waste incineration process of the embodiment, the waste incineration state is obtained through the deep learning-based method, the real-time automatic monitoring of the waste incineration state is realized, and the accuracy of waste incineration state identification is improved, so that the stability of waste incineration is improved, higher marginal economic benefits are brought to a waste incineration power plant, manpower is liberated, and the working environment of workers is improved; the method comprises the steps of simultaneously acquiring incinerator environment monitoring data and incinerator operation data, performing simulation prediction by adopting an LSTM model according to a garbage incineration state and incinerator operation data to obtain predicted smoke pollutant concentration and main combustion section temperature field distribution, inputting the incinerator environment monitoring data, the predicted smoke pollutant concentration and the predicted main combustion section temperature field distribution into a reinforcement learning control model to obtain incineration control parameters, and sending the incineration control parameters to the incinerator to make corresponding adjustment on the incinerator, so that the garbage incineration control parameters are optimized, the method can adapt to a complex garbage incineration process, automatic control of the garbage incineration process is realized, the stability of the garbage incineration is improved, manpower is liberated, and the labor cost is reduced.

Reference is made herein 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 hereof. For example, the various operational steps, as well as the components used to perform the operational steps, may be implemented in differing ways depending upon the particular application or consideration of any number of cost functions associated with operation of the system (e.g., one or more steps may be deleted, modified or incorporated into other steps).

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. Additionally, as will be appreciated by one skilled in the art, the principles herein may be reflected in a computer program product on a computer readable storage medium, which is pre-loaded with computer readable program code. Any tangible, non-transitory computer-readable storage medium may be used, including magnetic storage devices (hard disks, floppy disks, etc.), optical storage devices (CD-to-ROM, DVD, Blu-Ray discs, etc.), flash memory, and/or the like. These computer program instructions may be loaded onto a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions which execute on the computer or other programmable data processing apparatus create means for implementing the functions specified. These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including means for implementing the function specified. The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified.

While the principles herein have been illustrated in various embodiments, many modifications of structure, arrangement, proportions, elements, materials, and components particularly adapted to specific environments and operative requirements may be employed without departing from the principles and scope of the present disclosure. The above modifications and other changes or modifications are intended to be included within the scope of this document.

The foregoing detailed description has been described with reference to various embodiments. However, one skilled in the art will recognize that various modifications and changes may be made without departing from the scope of the present disclosure. Accordingly, the disclosure is to be considered in an illustrative and not a restrictive sense, and all such modifications are intended to be included within the scope thereof. Also, advantages, other advantages, and solutions to problems have been described above with regard to various embodiments. However, the benefits, advantages, solutions to problems, and any element(s) that may cause any element(s) to occur or become more pronounced are not to be construed as a critical, required, or essential feature or element of any or all the claims. As used herein, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, system, article, or apparatus. Furthermore, the term "coupled," and any other variation thereof, as used herein, refers to a physical connection, an electrical connection, a magnetic connection, an optical connection, a communicative connection, a functional connection, and/or any other connection.

Those skilled in the art will recognize that many changes may be made to the details of the above-described embodiments without departing from the underlying principles of the invention. Accordingly, the scope of the invention should be determined only by the claims.