A Highly Efficient Combustion Control Method and System Based on Deep Convolutional Neural Networks

Abstract

This invention provides a highly efficient combustion control method and system based on deep convolutional neural networks. It uses an all-around camera inside the furnace to image the flame, and then utilizes a deep convolutional neural network model to learn and diagnose combustion efficiency indicators. This obtains key combustion information without intrusive flame detection inside the furnace, and boasts advantages such as low latency and high reliability. Furthermore, the combustion control method, which uses predicted CO and ash carbon content as the core control benchmarks, has a natural advantage in improving combustion efficiency and maintains CO emissions at a highly controllable level, representing a major direction for future combustion control technology development. Additionally, using existing waste heat boiler outlet oxygen content and chimney CEMS as auxiliary control indicators ensures the entire combustion control mode is flawless and safe. This solution addresses the challenges of large load fluctuations in incinerators, difficulty in controlling instantaneous CO exceedances, and high excess air coefficients, thereby increasing combustion safety and improving combustion efficiency.

Description

Technical Field

[0001] This invention belongs to the field of combustion control, specifically relating to a highly efficient combustion control method and system based on deep convolutional neural networks. Background Technology

[0002] Combustion control is a major challenge in the operation of municipal solid waste incinerators. Combustion status diagnosis in stoker-fired incinerators primarily relies on traditional thermocouples, zirconia oxygen measuring points at the waste heat boiler outlet, carbon monoxide (CO) measuring points on the chimney CEMS, and manual observation via CCD cameras. These diagnostic methods offer very limited effective information. For example, data from in-furnace thermocouples and primary air pressure differences lack broad diagnostic significance and are susceptible to interference. CEMS data suffers from significant time delays, typically 2-3 minutes, increasing the difficulty of combustion control. Manual observation via CCD cameras is highly subjective. These detection methods either have limited representativeness or excessive time delays, hindering timely and effective adjustments when fuel input changes. Typical problems include difficulty in controlling instantaneous CO spikes, high excess air coefficients, and frequent, labor-intensive combustion adjustments, all contributing to poor economic efficiency and safety in stoker-fired incinerators. For instance, to ensure complete combustion, stoker-fired incinerators commonly use an excess air coefficient of 1.4-1.9, significantly higher than the 1.1-1.3 used in pulverized coal, oil-fired, and gas-fired incinerators.

[0003] In addition, existing combustion control technologies based on machine learning algorithms offer limited help in improving combustion efficiency. For example, patent CN 106765199 B focuses on identifying and classifying different combustion states, thus replacing manual labor. Patent CN 113313204 A goes further by predicting pollutant concentrations, but it does not specifically enhance combustion control for improving combustion efficiency. It can only replace manual labor and increase automation, but it does not increase combustion safety or improve combustion efficiency. Summary of the Invention

[0004] This invention provides a high-efficiency combustion control method and system based on deep convolutional neural networks. It predicts the CO and carbon content in the furnace based on flame image deep learning algorithm, and establishes a high-efficiency combustion control loop in the furnace based on the CO and carbon content in the ash. It solves problems such as difficulty in controlling instantaneous CO exceedance, high excess air coefficient, and frequent combustion adjustment with large operation volume, thereby increasing combustion safety and improving combustion efficiency.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A highly efficient combustion control method based on deep convolutional neural networks is proposed for mechanical grate incinerator systems. The method comprises the following steps to achieve real-time combustion control of the incinerator:

[0007] Step 1: Real-time acquisition of flame images inside the incinerator, CO concentration in flue gas inside the incinerator, carbon content in ash discharged from the incinerator, O2 volume at the incinerator outlet, and preset operating parameters for various types of incinerators.

[0008] Step 2: Based on flame images, CO concentration in flue gas, carbon content in ash, O2 content at the incinerator outlet, and preset operating parameters for each type of incinerator, combined with preset design specifications for each type of incinerator, train and construct a combustion prediction model with flame images, O2 content at the incinerator outlet, preset design specifications for each type of incinerator, and preset operating parameters for each type of incinerator as inputs, and CO concentration in flue gas and carbon content in ash as outputs.

[0009] Step 3: Based on the combustion prediction model, obtain the real-time predicted CO concentration in the flue gas inside the incinerator and the carbon content of the ash discharged from the incinerator.

[0010] Step 4: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, the carbon content of the ash discharged from the incinerator, combined with the real-time O2 quantity at the incinerator outlet, the CO quantity in the incinerator chimney, the preset range of O2 quantity at the incinerator outlet, the preset target value of CO concentration in the flue gas inside the incinerator, and the preset target value of carbon content of the ash discharged from the incinerator, the opening value of the blower in the incinerator is obtained in real time. Real-time combustion control of the incinerator is achieved by controlling the opening of the blower.

[0011] As a preferred technical solution of the present invention, in step 1, the flame images inside the incinerator include all flame images on the grate in the incinerator and all flame images of the flue of the waste heat boiler in the incinerator.

[0012] As a preferred technical solution of the present invention, in step 2, the preset design specifications of various types of incinerators include the incinerator processing capacity and the design calorific value of the waste.

[0013] As a preferred technical solution of the present invention, in step 2, the preset operating parameters of each type of incinerator include incinerator load rate and incinerator temperature.

[0014] As a preferred embodiment of the present invention, the fan includes a primary fan and a secondary fan.

[0015] As a preferred embodiment of the present invention, step 4 specifically includes the following steps:

[0016] Step 4.1: Based on the real-time predicted CO concentration in the flue gas inside the incinerator and the carbon content of the ash discharged from the incinerator, combined with the preset target values ​​for CO concentration in the flue gas inside the incinerator, the preset target values ​​for carbon content of the ash discharged from the incinerator, and the empirical parameter k1 value corresponding to the fan opening adjustment, the opening adjustment increment e1 of the primary air fan is obtained. This increment is then input into a fuzzy controller that takes the opening adjustment increment as input and the primary air fan opening adjustment value as output to obtain the primary air fan opening adjustment value. Similarly, the primary air fan temperature adjustment value is obtained by inputting the opening adjustment increment into a fuzzy controller that takes the opening adjustment increment as input and the primary air temperature adjustment value as output.

[0017] Step 4.2: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, combined with the CO amount in the incinerator chimney, the O2 amount at the incinerator outlet, and the preset range of O2 amount at the incinerator outlet, obtain the opening adjustment increment e4 and the increment change amount of the secondary air fan. Then, input the opening adjustment increment and increment change amount as inputs to a fuzzy controller with the opening of the secondary air fan as output to obtain the opening of the secondary air fan.

[0018] As a preferred embodiment of the present invention, the opening adjustment increment e1 of the primary air fan in step 4.1 is obtained by the following formula:

[0019] e1 = max{predicted carbon content of ash discharged from the incinerator - preset target value of carbon content of ash discharged from the incinerator, 0} + max{predicted CO concentration in flue gas in the incinerator - preset target value of CO concentration in flue gas in the incinerator} / 100 * k1;

[0020] In the formula, the target value for the carbon content of the ash discharged from the incinerator is the upper limit of the carbon content set according to laws, regulations and operating requirements; the target value for the CO concentration in the flue gas in the incinerator is the lower limit of the instantaneous CO concentration set according to operating requirements; and k1 is the preset weight adjustment factor.

[0021] As a preferred embodiment of the present invention, step 4.2 is specifically performed as follows:

[0022] Step 4.2.1: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, combined with the preset target value of CO concentration in the flue gas inside the incinerator and the CO amount in the incinerator chimney, the increment e2 is obtained through the following formula:

[0023] e2 = max{Predicted CO concentration in flue gas inside the incinerator - Preset target value of CO concentration in flue gas inside the incinerator, CO amount in the incinerator chimney - Preset target value of CO concentration in flue gas inside the incinerator, 0} / 100;

[0024] In the formula, the preset target value of CO concentration in flue gas in the incinerator is the lower limit of instantaneous CO concentration set according to the operating requirements;

[0025] Step 4.2.2: Based on the O2 quantity at the incinerator outlet and the preset range of O2 quantity at the incinerator outlet, the increment e3 is obtained through the following process:

[0026] The process of obtaining the increment e3 is as follows: When the O2 quantity at the incinerator outlet is less than or equal to the lower limit of the preset range of O2 quantity at the incinerator outlet, then e3 = lower limit of the preset range of O2 quantity at the incinerator outlet - O2 quantity at the incinerator outlet; when the O2 quantity at the incinerator outlet is greater than or equal to the upper limit of the preset range of O2 quantity at the incinerator outlet, then e3 = upper limit of the preset range of O2 quantity at the incinerator outlet - O2 quantity at the incinerator outlet; when the lower limit of the preset range of O2 quantity at the incinerator outlet is less than or equal to the upper limit of the preset range of O2 quantity at the incinerator outlet, then e3 = 0.

[0027] Step 4.2.3: Based on increments e2 and e3, obtain increment e4 and the rate of change of e4 using the following formula, and then input them into the fuzzy controller to obtain the secondary fan opening:

[0028] e4 = e2 + e3 * k2;

[0029] rate of change of e4 = d(e4) / t;

[0030] In the formula, k2 is the preset weight adjustment factor, and t is time.

[0031] A system for an efficient combustion control method based on the aforementioned deep convolutional neural network includes a data acquisition module, a combustion prediction model construction module, a combustion prediction module, and a combustion control module.

[0032] The data acquisition module is used to collect flame images inside the incinerator, CO concentration in the flue gas inside the incinerator, carbon content in the ash discharged from the incinerator, O2 volume at the incinerator outlet, and preset operating parameters for various types of incinerators.

[0033] The combustion prediction model building module is used to train and build a combustion prediction model based on flame images, CO concentration in flue gas, carbon content in ash, O2 content at the incinerator outlet, preset operating parameters for various types of incinerators, and preset design specifications for various types of incinerators. The model takes flame images, O2 content at the incinerator outlet, preset design specifications for various types of incinerators, and preset operating parameters for various types of incinerators as inputs, and CO concentration in flue gas and carbon content in ash as outputs.

[0034] The combustion prediction module is used to obtain real-time predictions of CO concentration in flue gas inside the incinerator and carbon content in ash discharged from the incinerator based on the combustion prediction model.

[0035] The combustion control module is used to obtain the opening value of the blower in the incinerator in real time based on the real-time predicted CO concentration in the flue gas in the incinerator, the carbon content of the ash discharged from the incinerator, the real-time O2 amount at the incinerator outlet, the CO amount in the incinerator chimney, the preset range of O2 amount at the incinerator outlet, the preset target value of CO concentration in the flue gas in the incinerator, and the preset target value of carbon content of the ash discharged from the incinerator.

[0036] The beneficial effects of this invention are as follows: This invention provides a highly efficient combustion control method and system based on deep convolutional neural networks, offering more comprehensive and timely combustion status diagnosis. It uses an all-around camera inside the furnace to image the flame, and then utilizes a deep convolutional neural network model to learn and diagnose combustion efficiency indicators. This obtains key combustion information without intrusive detection of the flame inside the furnace, and has the advantages of low latency and high reliability. Furthermore, the combustion control method, which uses CO and ash carbon content predicted by the neural network model as the core control benchmark parameters, has a natural advantage in improving combustion efficiency compared to adjusting the oxygen content at the waste heat boiler outlet, and always maintains CO emissions in a highly controllable state, representing a major direction for the future development of combustion control technology. In addition, using existing waste heat boiler outlet oxygen content and chimney CEMS as auxiliary control indicators ensures that the entire combustion control mode is flawless and safe. This solution predicts the CO and ash carbon content inside the furnace based on a deep learning algorithm for flame images, and establishes a highly efficient combustion control loop based on the CO and ash carbon content. This solves the problems of large load fluctuations in the incinerator, difficulty in controlling instantaneous CO exceedances, and high excess air coefficients, thereby increasing combustion safety and improving combustion efficiency. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the incinerator and camera system in an embodiment of the present invention;

[0038] Figure 1 In the image, the components are: 1. Incinerator; 2. Primary air fan; 3. Secondary air fan; 4. Waste heat boiler; 41. Boundary line between waste heat boiler and incinerator; 42. First flue of waste heat boiler; 43. Second flue of waste heat boiler; 5. Oxygen meter at the outlet of waste heat boiler; 6. Camera above the first flue of waste heat boiler; 61. Camera field of view; 7. Camera at the tail of incinerator; 8. Computer.

[0039] Figure 2 This is a schematic diagram showing the arrangement of the top camera in an embodiment of the present invention;

[0040] Figure 2 In the middle, a) is the position above 2 / 3 of the height of the flue; b) is the half space near the front wall of the top of a flue;

[0041] Figure 3 This is a deep learning model for flame images in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the combustion prediction model training process in an embodiment of the present invention;

[0043] Figure 5 This is a flowchart illustrating the control process of a primary fan in an embodiment of the present invention.

[0044] Figure 6 This is a flowchart illustrating the control process of the secondary air blower in an embodiment of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to the accompanying drawings. The following embodiments will enable those skilled in the art to more fully understand the present invention, but do not limit the present invention in any way.

[0046] A highly efficient combustion control method based on deep convolutional neural networks is proposed for mechanical grate incinerator systems. The method involves performing the following steps to achieve real-time combustion control of the incinerator.

[0047] In this embodiment, the incinerator used is such as Figure 1 As shown, this is a mechanical grate-type municipal solid waste incinerator, which includes an incinerator, air distribution system, waste heat boiler, and oxygen measuring point at the outlet of the waste heat boiler. Specifically, it includes incinerator 1, primary air fan 2, secondary air fan 3, waste heat boiler 4, boundary line between waste heat boiler and incinerator 41, waste heat boiler first flue 42, waste heat boiler second flue 43, and waste heat boiler outlet oxygen meter 5. The incinerator includes a feeding device, grate bars, ash removal device, furnace arch, etc. The primary air fan system is located below the grate. The primary air drying, combustion, and burnout processes are divided into several air chambers. The secondary air fan is located above the incinerator. The outlet of the incinerator is the waste heat boiler. The waste heat boiler is divided into first flue, second flue, third flue, and horizontal flue according to the number of flues from front to back. The incinerator system is a conventional design.

[0048] Step 1: Real-time acquisition of flame images inside the incinerator, CO concentration in the flue gas inside the incinerator, carbon content in the ash discharged from the incinerator, O2 volume at the incinerator outlet, and preset operating parameters for various types of incinerators.

[0049] In step 1, the flame images inside the incinerator include all flame images on the grate in the incinerator and all flame images of the flue of the waste heat boiler in the incinerator.

[0050] In this embodiment, flame images inside the incinerator can be captured by a camera device, and the camera must adhere to the following design and installation principles:

[0051] (1) At least one or more cameras should be installed in the rear arch of the incinerator, such as Figure 1The camera at the tail of the incinerator shown in Figure 7 must have an observation range that includes the entire flame area on the grate in the incinerator, including all combustion areas in both the horizontal and vertical directions.

[0052] (2) Install at least one camera on the upper part of the flue of the waste heat boiler, such as Figure 1 The camera above the flue of the waste heat boiler shown in Figure 6 must have an observation range that includes at least two-thirds of the flame area below the height of the flue. The camera should be positioned above two-thirds of the height of the front wall of the flue. Figure 1 The camera's field of view is shown in Figure 62; or the top position near the front wall, such as... Figure 2 As shown, 'a' represents the position above 2 / 3 of the flue height, and 'b' represents the top half of the flue near the front wall, i.e., the side of the flue centerline slightly off-center from the front wall. In the width direction of the incinerator, the camera is positioned near the centerline to observe the flames on both sides equally.

[0053] (3) Other cameras besides the above cameras can provide more observable range or observation information of the flame.

[0054] In this embodiment, the camera can be a CCD or ICCD camera to image the visible light emitted by the flame, or it can also include an additional infrared camera to image the flame of the material layer.

[0055] Step 2: Based on flame images, CO concentration in flue gas, carbon content in ash, O2 content at the incinerator outlet, and preset operating parameters for each type of incinerator, combined with preset design specifications for each type of incinerator, train and construct a combustion prediction model that takes flame images, O2 content at the incinerator outlet, preset design specifications for each type of incinerator, and preset operating parameters for each type of incinerator as inputs, and CO concentration in flue gas and carbon content in ash as outputs.

[0056] The preset design specifications for each type of incinerator include the incinerator's processing capacity and the design calorific value of the waste. The preset operating parameters for each type of incinerator include the incinerator load rate and the incinerator temperature.

[0057] In this embodiment, the training dataset corresponding to the combustion prediction model is obtained by using the aforementioned camera device to image the combustion conditions inside the furnace under various operating conditions. The operating conditions here are the flame images inside the incinerator collected in real time within a preset historical time period, the CO concentration in the flue gas inside the incinerator, the carbon content of the ash discharged from the incinerator, the O2 amount at the outlet of the incinerator, and the preset operating parameters of various types of incinerators, which constitute the training samples. A set of flame images was acquired. Simultaneously, the CO concentration in the flue gas was measured in real-time at the flue or outlet of the waste heat boiler. High-precision, short-delay, and wide-range measurement technologies, such as a CO laser gas analyzer based on TDLAS (Tunable Laser Diode), were employed to improve measurement accuracy. Ash and slag discharged from the incinerator under corresponding combustion conditions were collected, and the carbon content (also called ash calorific value loss) was measured. A training dataset was established by mapping the flame images, O2 levels at the waste heat boiler outlet, load rate, furnace temperature, and incinerator design specifications (including but not limited to processing capacity and design waste calorific value) under the same operating condition to the measured CO concentration and ash carbon content. Typical operating conditions should include different loads, waste calorific values, O2 content, and CO concentrations, and should be repeated multiple times to collect a sufficient amount of sample data.

[0058] The combustion prediction model includes a feature extraction module and a prediction module. The feature extraction module is used to extract features from the flame images in step 2 to obtain image feature vectors corresponding to each flame image, which are then input into the prediction module. In this embodiment, feature extraction is performed using a deep convolutional neural network (DCNN) model, such as... Figure 3 The deep learning model for flame images shown extracts features from flame images captured by the above cameras. It uses the flame images captured by the above cameras, the O2 concentration at the incinerator outlet, the incinerator design specifications, and the incinerator operating parameters as training data, and CO concentration and ash carbon content as data labels to train the preset combustion model, resulting in a combustion prediction model for predicting CO and ash carbon content in the incinerator. In this embodiment, multiple convolutional and pooling layers in the deep convolutional neural network model are used for feature extraction to improve prediction capability. Since the angle and environment of each camera are different, and the flame location seen is also different, different convolutional kernels and pooling methods are needed to enhance processing capability, and adjustments need to be made based on the training results. Therefore, different convolutional kernels and pooling methods can be used for images from different cameras to improve the feature extraction capability of different cameras. After feature extraction from images from different cameras, their corresponding image feature vectors X are obtained. A X B …X M .

[0059] The image feature vector, combined with parameters such as the O2 output at the incinerator outlet, preset design specifications for various types of incinerators, and preset operating parameters for various types of incinerators, forms the operating feature vector X.S The input data for the prediction module consists of CO in the furnace and carbon content in the ash. The output parameters of the prediction module are CO in the furnace and carbon content in the ash. The prediction module includes, but is not limited to, prediction models and methods such as the fully connected layer of DCNN, support vector machine, and least squares support vector regression. In this embodiment, a combination model of DCNN and support vector machine is used, where DCNN is used to extract image features and support vector machine is used to establish the relationship between the input data and the prediction object. The fully connected layer in DCNN is not used for prediction; instead, the support vector machine prediction method is used because it is more efficient.

[0060] The training process of the combustion prediction model is as follows: Figure 4 As shown, the first step is to acquire sample data under different typical operating conditions, including images, incinerator design specifications, key operating parameters, and measured CO concentration and ash carbon content. Then, DCNN feature extraction is performed on the images to obtain feature vectors. The prediction module is then trained by combining incinerator specifications, operating data, and measurement data. The model is then tested and evaluated. Once qualified, the target model, i.e., the combustion prediction model, is obtained. The data can be imported into DCS or combustion control ACC to participate in the combustion process control.

[0061] Step 3: Based on the combustion prediction model, obtain the real-time predicted CO concentration in the flue gas inside the incinerator and the carbon content of the ash discharged from the incinerator.

[0062] Step 4: Based on real-time predicted CO concentration in the flue gas inside the incinerator, carbon content in the ash discharged from the incinerator, combined with real-time O2 output at the incinerator outlet, CO output at the incinerator chimney, preset O2 output range at the incinerator outlet, preset target value for CO concentration in the flue gas inside the incinerator, and preset target value for carbon content in the ash discharged from the incinerator, the opening degree of the blowers in the incinerator is obtained in real time. Real-time combustion control of the incinerator is achieved by controlling the opening degree of the blowers. Figure 1 The computer in the system controls the temperature and opening degree of the blower to achieve real-time combustion control of the incinerator.

[0063] The chimney is the final stage in flue gas purification, the outlet where the flue gas is finally released into the atmosphere; all such projects have chimneys. According to industry standards, all municipal solid waste incineration projects must have an online continuous emission monitoring system (CEMS) installed at the chimney to monitor various parameters of the flue gas emitted into the air, such as flue gas volume and pollutant concentrations, including CO. However, CEMS data has a relatively large delay, typically 2-3 minutes, making it unsuitable for real-time control; it can be used for auxiliary control.

[0064] Step 4, the specific steps are as follows:

[0065] Step 4.1: Based on the real-time predicted CO concentration in the flue gas inside the incinerator and the carbon content of the ash discharged from the incinerator, combined with the preset target values ​​for CO concentration in the flue gas inside the incinerator, the preset target values ​​for carbon content of the ash discharged from the incinerator, and the empirical parameter k1 value corresponding to the fan opening adjustment, the primary air fan opening adjustment increment e1 is obtained. This increment is then input to a fuzzy controller that takes the opening adjustment increment as input and the primary air fan opening adjustment value as output to obtain the primary air fan opening adjustment value; similarly, it is input to a fuzzy controller that takes the opening adjustment increment as input and the primary air temperature adjustment value as output to obtain the primary air temperature adjustment value. For example... Figure 5 As shown.

[0066] In step 4.1, the opening adjustment increment e1 of the primary air fan is obtained by the following formula:

[0067] e1 = max{predicted carbon content of ash discharged from the incinerator - preset target value of carbon content of ash discharged from the incinerator, 0} + max{predicted CO concentration in flue gas in the incinerator - preset target value of CO concentration in flue gas in the incinerator} / 100 * k1;

[0068] In the formula, the preset target value for the carbon content of the ash discharged from the incinerator is the upper limit of the carbon content set according to laws, regulations and operational requirements, generally 3 to 5%; the preset target value for the CO concentration in the flue gas in the incinerator is the lower limit of the instantaneous CO concentration set according to operational requirements. Exceeding this limit means that operational operations are required to adjust the CO, generally 20 to 50 mg / Nm3, but can also exceed this range; k1 is a preset weight adjustment factor, generally taken in the range of 0 to 5, but can also exceed this range. The larger the value, the greater the proportion of CO adjustment.

[0069] Step 4.2: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, combined with the CO amount in the incinerator chimney, the O2 amount at the incinerator outlet, and the preset range of O2 amount at the incinerator outlet, the incremental adjustment e4 and the change in increment of the secondary air fan are obtained. This is then input into a fuzzy controller that takes the incremental adjustment e4 and the change in increment as input and the secondary air fan opening as output to obtain the secondary air fan opening. For example... Figure 6 As shown.

[0070] In step 4.2, the specific process is as follows:

[0071] Step 4.2.1: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, combined with the preset target value of CO concentration in the flue gas inside the incinerator and the CO amount in the incinerator chimney, the increment e2 is obtained through the following formula:

[0072] e2 = max{Predicted CO concentration in flue gas inside the incinerator - Preset target value of CO concentration in flue gas inside the incinerator, CO amount in the incinerator chimney - Preset target value of CO concentration in flue gas inside the incinerator, 0} / 100;

[0073] In the formula, the preset target value of CO concentration in flue gas in the incinerator is the lower limit of instantaneous CO concentration set according to the operating requirements;

[0074] Step 4.2.2: Based on the O2 quantity at the incinerator outlet and the preset range of O2 quantity at the incinerator outlet, the increment e3 is obtained through the following process:

[0075] The process of obtaining the increment e3 is as follows: When the O2 quantity at the incinerator outlet is less than or equal to the lower limit of the preset O2 quantity range, then e3 = lower limit of the preset O2 quantity range - O2 quantity at the incinerator outlet; when the O2 quantity at the incinerator outlet is greater than the upper limit of the preset O2 quantity range, then e3 = upper limit of the preset O2 quantity range - O2 quantity at the incinerator outlet; when the lower limit of the preset O2 quantity range is less than or equal to the upper limit of the preset O2 quantity range, then e3 = 0. In the formula, the lower limit and upper limit of the preset O2 quantity range are the lower and upper limits of the normal O2 range that do not require adjustment. These limits vary depending on the incinerator, waste characteristics, and operational requirements, and are empirical parameters. Generally, the lower limit ranges from 3% to 5%, and the upper limit ranges from 5% to 7%, but these ranges can also be exceeded.

[0076] Step 4.2.3: Based on increments e2 and e3, obtain increment e4 and the rate of change of e4 using the following formula, and then input them into the fuzzy controller to obtain the secondary fan opening:

[0077] e4 = e2 + e3 * k2;

[0078] rate of change of e4 = d(e4) / t;

[0079] In the formula, k2 is the preset weight adjustment factor; k2 is the weight adjustment factor. The larger the value, the greater the proportion of O2 adjustment. It is generally taken as 0 to 2, but can also exceed this range; t is time.

[0080] Fuzzy control theory is a control method based on fuzzy set theory, fuzzy language, and fuzzy logic. Fuzzy control technology uses control rules to describe the relationship between system variables. It is a nonlinear intelligent control method with the advantages of simplifying system complexity and being applicable to nonlinear, time-varying, and incomplete model systems. It is very suitable for the air distribution control of waste incinerators.

[0081] Based on the above scheme, a system for efficient combustion control based on deep convolutional neural networks is also proposed, including a data acquisition module, a combustion prediction model construction module, a combustion prediction module, and a combustion control module.

[0082] The data acquisition module is used to collect flame images inside the incinerator, CO concentration in the flue gas inside the incinerator, carbon content in the ash discharged from the incinerator, O2 volume at the incinerator outlet, and preset operating parameters for various types of incinerators.

[0083] In this embodiment, the device for acquiring flame images inside the incinerator includes (1) at least one or more cameras installed on the rear arch of the incinerator, the camera's observation range needing to include the entire flame range on the grate, including all combustion areas in both the horizontal and vertical directions; (2) at least one camera installed on the upper part of a flue of the waste heat boiler, the camera's observation range needing to include at least the entire flame range below 2 / 3 of the flue's height, the camera's position should be above 2 / 3 of the height of the front wall of the flue, or within half of the space near the front wall at the top, such as... Figure 2 As shown; (3) Other cameras besides the above cameras can provide more observable range or observation information of the flame.

[0084] The device for collecting CO concentration in flue gas inside the incinerator employs high-precision, short-delay, and wide-range measurement technologies, such as a CO laser gas analyzer based on TDLAS (tunable laser diode), to improve measurement accuracy.

[0085] The combustion prediction model building module is used to train and build a combustion prediction model based on flame images, CO concentration in flue gas, carbon content in ash, O2 content at the incinerator outlet, preset operating parameters for various types of incinerators, and preset design specifications for various types of incinerators. The model takes flame images, O2 content at the incinerator outlet, preset design specifications for various types of incinerators, and preset operating parameters for various types of incinerators as inputs, and CO concentration in flue gas and carbon content in ash as outputs.

[0086] The combustion prediction module is used to obtain real-time predictions of CO concentration in flue gas inside the incinerator and carbon content in ash discharged from the incinerator based on the combustion prediction model.

[0087] The combustion control module is used to obtain the opening value of the blower in the incinerator in real time based on the real-time predicted CO concentration in the flue gas in the incinerator, the carbon content of the ash discharged from the incinerator, the real-time O2 amount at the incinerator outlet, the CO amount in the incinerator chimney, the preset range of O2 amount at the incinerator outlet, the preset target value of CO concentration in the flue gas in the incinerator, and the preset target value of carbon content of the ash discharged from the incinerator.

[0088] This invention designs a highly efficient combustion control method and system based on deep convolutional neural networks, providing more comprehensive and timely combustion status diagnosis. It uses an all-around camera inside the furnace to image the flame, and then utilizes a deep convolutional neural network model to learn and diagnose combustion efficiency indicators. This obtains key combustion information without intrusive flame detection inside the furnace, and boasts advantages such as low latency and high reliability. Furthermore, the combustion control method, which uses CO and ash carbon content predicted by the neural network model as core control benchmark parameters, has a natural advantage in improving combustion efficiency compared to adjusting the oxygen content at the waste heat boiler outlet, and maintains CO emissions at a highly controllable level, representing a major direction for future combustion control technology development. Additionally, using existing waste heat boiler outlet oxygen content and chimney CEMS as auxiliary control indicators ensures the entire combustion control mode is flawless and safe. This scheme predicts CO and ash carbon content inside the furnace based on a deep learning algorithm for flame images, and establishes a highly efficient combustion control loop based on these values. This solves the problems of large load fluctuations in the incinerator, difficulty in controlling instantaneous CO exceedances, and high excess air coefficients, thereby increasing combustion safety and improving combustion efficiency.

[0089] The above are merely preferred embodiments of the present invention, but do not limit the patent scope of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of the present invention specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the patent protection scope of the present invention.

Claims

1. A highly efficient combustion control method based on a deep convolutional neural network, characterized in that, For a stoker-type mechanical grate incinerator system, the following steps are performed to achieve real-time combustion control of the incinerator: Step 1: Real-time acquisition of flame images inside the incinerator, CO concentration in flue gas inside the incinerator, carbon content in ash discharged from the incinerator, O2 volume at the incinerator outlet, and preset operating parameters for various types of incinerators. Step 2: Based on flame images, CO concentration in flue gas, carbon content in ash, O2 content at the incinerator outlet, and preset operating parameters for each type of incinerator, combined with preset design specifications for each type of incinerator, train and construct a combustion prediction model with flame images, O2 content at the incinerator outlet, preset design specifications for each type of incinerator, and preset operating parameters for each type of incinerator as inputs, and CO concentration in flue gas and carbon content in ash as outputs. Step 3: Based on the combustion prediction model, obtain the real-time predicted CO concentration in the flue gas inside the incinerator and the carbon content of the ash discharged from the incinerator. Step 4: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, the carbon content of the ash discharged from the incinerator, combined with the real-time O2 quantity at the incinerator outlet, the CO quantity in the incinerator chimney, the preset range of O2 quantity at the incinerator outlet, the preset target value of CO concentration in the flue gas inside the incinerator, and the preset target value of carbon content of the ash discharged from the incinerator, the opening value of the blower in the incinerator is obtained in real time. Real-time combustion control of the incinerator is achieved by controlling the opening of the blower.

2. The efficient combustion control method based on a deep convolutional neural network according to claim 1, characterized in that, In step 1, the flame images inside the incinerator include all flame images on the grate in the incinerator and all flame images of the flue of the waste heat boiler in the incinerator.

3. The efficient combustion control method based on a deep convolutional neural network according to claim 1, characterized in that, In step 2, the preset design specifications for each type of incinerator include the incinerator's processing capacity and the incinerator's design calorific value of the waste.

4. The efficient combustion control method based on a deep convolutional neural network according to claim 1, characterized in that, In step 2, the preset operating parameters for each type of incinerator include incinerator load rate and incinerator temperature.

5. The efficient combustion control method based on a deep convolutional neural network according to claim 1, characterized in that, The fan includes a primary fan and a secondary fan.

6. The efficient combustion control method based on a deep convolutional neural network according to claim 1, characterized in that, Step 4, the specific steps are as follows: Step 4.1: Based on the real-time predicted CO concentration in the flue gas inside the incinerator and the carbon content of the ash discharged from the incinerator, combined with the preset target values ​​for CO concentration in the flue gas inside the incinerator, the preset target values ​​for carbon content of the ash discharged from the incinerator, and the preset weight adjustment factor k1 value corresponding to the fan opening adjustment, the opening adjustment increment e1 of the primary air fan is obtained. This increment is then input into a fuzzy controller that takes the opening adjustment increment as input and the primary air fan opening adjustment value as output to obtain the primary air fan opening adjustment value. Similarly, the primary air fan temperature adjustment value is obtained by inputting the opening adjustment increment into a fuzzy controller that takes the opening adjustment increment as input and the primary air temperature adjustment value as output. Step 4.2: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, combined with the CO amount in the incinerator chimney, the O2 amount at the incinerator outlet, and the preset range of O2 amount at the incinerator outlet, obtain the opening adjustment increment e4 and the increment change amount of the secondary air fan. Then, input the opening adjustment increment and increment change amount as inputs to a fuzzy controller with the opening of the secondary air fan as the output to obtain the opening of the secondary air fan.

7. The efficient combustion control method based on a deep convolutional neural network according to claim 6, characterized in that, In step 4.1, the opening adjustment increment e1 of the primary air fan is obtained by the following formula: e1 = max{predicted carbon content of ash discharged from the incinerator - preset target value of carbon content of ash discharged from the incinerator, 0} + max{predicted CO concentration in flue gas in the incinerator - preset target value of CO concentration in flue gas in the incinerator} / 100 * k1; In the formula, the target value for the carbon content of the ash discharged from the incinerator is the upper limit value of the carbon content set according to laws, regulations and operational requirements; The preset target value for CO concentration in flue gas inside the incinerator is the lower limit of instantaneous CO concentration set according to operating requirements; k1 is a preset weight adjustment factor.

8. The efficient combustion control method based on a deep convolutional neural network according to claim 6, characterized in that, In step 4.2, the specific process is as follows: Step 4.2.1: Based on the real-time predicted CO concentration in the flue gas inside the incinerator, combined with the preset target value of CO concentration in the flue gas inside the incinerator and the CO amount in the incinerator chimney, the increment e2 is obtained through the following formula: e2 = max{Predicted CO concentration in flue gas inside the incinerator - Preset target value of CO concentration in flue gas inside the incinerator, CO amount in the incinerator chimney - Preset target value of CO concentration in flue gas inside the incinerator, 0} / 100; In the formula, the preset target value of CO concentration in flue gas in the incinerator is the lower limit of instantaneous CO concentration set according to the operating requirements; Step 4.2.2: Based on the O2 quantity at the incinerator outlet and the preset range of O2 quantity at the incinerator outlet, the increment e3 is obtained through the following process: The process of obtaining the increment e3 is as follows: When the O2 quantity at the incinerator outlet is less than or equal to the lower limit of the preset range of O2 quantity at the incinerator outlet, then e3 = lower limit of the preset range of O2 quantity at the incinerator outlet - O2 quantity at the incinerator outlet; when the O2 quantity at the incinerator outlet is greater than or equal to the upper limit of the preset range of O2 quantity at the incinerator outlet, then e3 = upper limit of the preset range of O2 quantity at the incinerator outlet - O2 quantity at the incinerator outlet; when the lower limit of the preset range of O2 quantity at the incinerator outlet is less than or equal to the upper limit of the preset range of O2 quantity at the incinerator outlet, then e3 = 0. Step 4.2.3: Based on increments e2 and e3, obtain increment e4 and the rate of change of e4 using the following formula, and then input them into the fuzzy controller to obtain the secondary fan opening: e4 = e2 + e3 * k2; rate of change of e4 = d(e4) / t; In the formula, k2 is the preset weight adjustment factor, and t is time.

9. A system for an efficient combustion control method based on any one of the deep convolutional neural networks described in claims 1-8, characterized in that, It includes a data acquisition module, a combustion prediction model building module, a combustion prediction module, and a combustion control module. The data acquisition module is used to collect flame images inside the incinerator, CO concentration in the flue gas inside the incinerator, carbon content in the ash discharged from the incinerator, O2 volume at the incinerator outlet, and preset operating parameters for various types of incinerators. The combustion prediction model building module is used to train and build a combustion prediction model based on flame images, CO concentration in flue gas, carbon content in ash, O2 content at the incinerator outlet, preset operating parameters for various types of incinerators, and preset design specifications for various types of incinerators. The model takes flame images, O2 content at the incinerator outlet, preset design specifications for various types of incinerators, and preset operating parameters for various types of incinerators as inputs, and CO concentration in flue gas and carbon content in ash as outputs. The combustion prediction module is used to obtain real-time predictions of CO concentration in flue gas inside the incinerator and carbon content in ash discharged from the incinerator based on the combustion prediction model. The combustion control module is used to obtain the opening value of the blower in the incinerator in real time based on the real-time predicted CO concentration in the flue gas in the incinerator, the carbon content of the ash discharged from the incinerator, the real-time O2 amount at the incinerator outlet, the CO amount in the incinerator chimney, the preset range of O2 amount at the incinerator outlet, the preset target value of CO concentration in the flue gas in the incinerator, and the preset target value of carbon content of the ash discharged from the incinerator.

Images