CN-121995769-A - Combustion optimization control method and system of semi-coke furnace based on machine vision

CN121995769ACN 121995769 ACN121995769 ACN 121995769ACN-121995769-A

Abstract

The application discloses a machine vision-based combustion optimization control method and system for a semi-coke furnace. The method comprises the steps of respectively generating a flame dynamic texture map and a furnace temperature field time-space evolution matrix by collecting a combustion chamber image sequence. The two are input into a coupling analysis network, deep correlation between flame texture and temperature field evolution is excavated, and a coupling characteristic tensor representing a combustion state is output. And inquiring a pre-generated combustion mode knowledge graph based on the tensor to match the most similar reference control strategy. And finally, calculating and outputting a control instruction through the multivariable coordinator by combining the real-time operation data. The method provided by the application realizes the deep coupling analysis and intelligent decision of flame vision and temperature field, improves the sensing precision and control adaptability of the combustion state, and optimizes the combustion efficiency and stability.

Inventors

WANG FUPING
YAN JIMIN
WANG TING
LIU YUANLI
QIU XIAOYUN
JI YANFEI

Assignees

神木泰和煤化工有限公司

Dates

Publication Date: 20260508
Application Date: 20260313

Claims (10)

1. The combustion optimization control method of the semi-coke furnace based on machine vision is characterized by comprising the following steps of: s1, generating a flame dynamic texture map through an image processing algorithm based on an image sequence of an acquired semi-coke oven combustion chamber in a set time period; S2, separating an infrared thermal imaging image subsequence from the image sequence, performing region gridding segmentation on the infrared thermal imaging image subsequence, calculating a statistical temperature value in each grid unit and a gradient of the statistical temperature value changing along with time, and constructing a temperature field time-space evolution matrix of a hearth based on the statistical temperature value and the gradient; S3, inputting the flame dynamic texture map and the temperature field space-time evolution matrix into a pre-trained coupling analysis network together, analyzing an implicit association mode between flame texture dynamic change and temperature field space-time evolution, and outputting a coupling characteristic tensor representing the current combustion coupling state; s4, performing similarity matching in a combustion mode knowledge graph based on the coupling characteristic tensor to acquire a reference control strategy most similar to the coupling characteristic tensor, wherein the combustion mode knowledge graph stores mapping relations between the coupling characteristic tensor and the optimal control strategy under various typical working conditions; S5, calculating the acquired real-time fuel supply flow data, primary air volume data and secondary air volume data of the semi-coke furnace through a multivariable coordinator based on the reference control strategy to obtain the adjustment quantity of the fuel quantity and the air volume set value, wherein the combustion state of the adjustment quantity approximates to the target coupling characteristic; s6, generating a control instruction set containing an execution time sequence according to the fuel quantity and the adjustment quantity of the air quantity set value so as to drive the semi-coke furnace executing mechanism to realize combustion optimization control.
2. The machine vision-based semi-coke oven combustion optimization control method of claim 1, wherein the image sequence comprises a visible light image reflecting flame morphology and an infrared thermal imaging image reflecting a temperature field; The step S1 comprises the following steps: S11, carrying out synchronous space-time registration on the image sequence, enabling the visible light image and the infrared thermal imaging image to be aligned on space pixel coordinates and a time stamp, and generating a space-time synchronous image sequence; s12, separating a visible light image subsequence based on the space-time synchronous image sequence, transferring to a specific color space for separating brightness information and chromaticity information, extracting dynamic texture features related to combustion stability, and forming a flame dynamic texture map.
3. The machine vision-based semi-coke oven combustion optimization control method according to claim 2, wherein the step S11 comprises: S111, respectively detecting characteristic points of the visible light image and the infrared thermal imaging image, and extracting stable angular points or edge characteristics; s112, based on the stable angular points or the edge characteristics, obtaining space transformation model parameters between the visible light image and the infrared thermal imaging image through a characteristic matching algorithm; s113, carrying out space resampling on the infrared thermal imaging image based on the space transformation model parameters so as to align the visual angles and the scales of the infrared thermal imaging image and the visible light image; And S114, marking a time stamp for the image sequence, and carrying out inter-frame interpolation alignment on the visible light image and the infrared thermal imaging image based on the time stamp so that the images with the same time stamp reflect the combustion state of the corresponding time stamp.
4. The machine vision-based semi-coke oven combustion optimization control method according to claim 2, wherein the step S12 comprises: S121, carrying out differential operation on continuous multi-frame images of the visible light image subsequence on a chromaticity channel to obtain a dynamic differential image sequence reflecting flame color change; S122, performing three-dimensional wavelet transformation on the dynamic differential image sequence, and extracting wavelet coefficient energy of a specific frequency band, wherein the specific frequency band corresponds to a characteristic frequency range of flame jitter; S123, arranging the wavelet coefficient energy in time sequence and unfolding along a space dimension to form the flame dynamic texture map, wherein the horizontal axis of the flame dynamic texture map is time, the vertical axis of the flame dynamic texture map is a space position, and the map value is texture liveness.
5. The machine vision-based semi-coke oven combustion optimization control method according to claim 1, wherein the step S2 comprises: s21, dividing each frame of image in the infrared thermal imaging image subsequence into a regular grid according to a hearth physical structure; s22, calculating average temperature values of all infrared pixel points of each grid in each frame of image; S23, based on the infrared thermal imaging image subsequence, recording a temperature time sequence of the average temperature value of each grid changing along with time, and calculating the temperature change rate of each grid at adjacent time points based on the temperature time sequence to obtain a temperature gradient; S24, arranging average temperature values of all grids at the same moment into a first matrix according to grid positions, arranging temperature gradients of all grids at the same moment into a second matrix according to grid positions, and stacking the first matrix and the second matrix based on a time dimension to form a space-time evolution matrix.
6. The machine vision-based semi-coke oven combustion optimization control method of claim 1, wherein the coupling analysis network comprises a texture feature coding path and a temperature feature coding path; in the step S3, the flame dynamic texture map and the temperature field space-time evolution matrix are input to a pre-trained coupling analysis network together, and a step of analyzing an implicit association mode between flame texture dynamic change and temperature field space-time evolution comprises the following steps: s31, carrying out convolution processing on the flame dynamic texture map through the texture feature coding passage to extract multi-scale space-time texture features; S32, carrying out convolution processing on the temperature field space-time evolution matrix through the temperature characteristic coding passage, and extracting multi-scale space-time temperature distribution and change characteristics; s33, carrying out outer product interaction processing on the multi-scale space-time texture features, the multi-scale space-time temperature distribution and the change features at a fusion layer of the coupling analysis network to obtain a high-order interaction feature map; And S34, carrying out global pooling and full connection transformation on the high-order interaction feature map to obtain a coupling feature tensor for representing the combustion coupling state.
7. The machine vision-based semi-coke oven combustion optimization control method according to claim 1, wherein the combustion mode knowledge graph is constructed by historical operation data, and nodes in the combustion mode knowledge graph comprise coupling characteristic tensor samples extracted under various historical working conditions and control parameter combinations corresponding to the coupling characteristic tensor samples; the step S4 comprises the following steps: s41, performing similarity matching on the coupling characteristic tensor and coupling characteristic tensor samples of all nodes in the combustion mode knowledge graph; s42, screening all neighbor nodes with similarity exceeding a set threshold value; s43, generating a reference control strategy based on the control parameter combination corresponding to all the neighbor nodes.
8. The machine vision-based semi-coke oven combustion optimization control method of claim 7, wherein the multivariate coordinator is built in with a simplified dynamic model configured to describe a dynamic mapping relationship between fuel quantity, primary air quantity, secondary air quantity and coupling characteristic tensor; The multivariable coordinator is also defined with an objective function for minimizing the difference between the predicted combustion state of the simplified dynamic model and the expected target in the prediction time domain and punishing excessive control quantity change; the multivariable coordinator is also used for setting process constraint conditions, wherein the process constraint conditions at least comprise an air quantity upper limit and a lower limit, a fuel valve opening limit and an air-fuel ratio safety range; The step S5 comprises the following steps: S51, taking a control parameter combination in the reference control strategy as an expected target, taking current fuel supply flow data, primary air volume data and secondary air volume data as initial states, and inputting the initial states into the simplified dynamic model; S52, obtaining a control quantity change sequence which enables the objective function to be optimal under the process constraint condition, and taking an instant control quantity change value in the control quantity change sequence as an adjustment quantity of the fuel quantity and the air quantity set value.
9. The machine vision-based semi-coke oven combustion optimization control method according to claim 1, wherein the step S6 comprises: S61, respectively generating independent adjustment instructions for a fuel supply valve, a primary air door and a secondary air door based on analysis results of the fuel quantity and the adjustment quantity of the air quantity set value; S62, respectively distributing an execution time starting point and an execution duration time for each independent adjustment instruction based on a sequential logic sequence, wherein the sequential logic sequence is that the primary air quantity is higher than the fuel quantity, and the fuel quantity is higher than the secondary air quantity; s63, dividing the adjustment quantity of each independent adjustment instruction into a plurality of adjustment step sizes, and setting an execution interval for each adjustment step size to form a smooth slope type control signal; S64, packaging all independent adjustment instructions and encapsulating the independent adjustment instructions into the control instruction set.
10. A machine vision-based semi-coke oven combustion optimization control system comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that the processor, when executing the computer program, realizes the steps of the machine vision-based semi-coke oven combustion optimization control method according to any one of the preceding claims 1 to 9.

Description

Combustion optimization control method and system of semi-coke furnace based on machine vision Technical Field The invention relates to the technical field of intelligent control of industrial kilns, in particular to a combustion optimization control method and system of a semi-coke furnace based on machine vision. Background In the combustion control of semi-coke production, the prior art generally relies on independent monitoring means. The flame visual monitoring system captures the changes of flame form, color and brightness through a camera and is used for qualitatively judging combustion stability, and the infrared temperature measuring system measures the temperature of a hearth through scanning or fixed point positions to acquire local or whole temperature distribution information. The two systems often run in parallel, and the data only carry out simple comparison or threshold alarm at the control level, so that deep fusion analysis of visual characteristics and a dynamic evolution process of a temperature field cannot be realized mechanically. The separated monitoring leads to the judgment of the combustion state to stay on the surface or local part, and the coupling state between the fuel, the air quantity and the complex combustion chemical reaction cannot be accurately reflected. At the control strategy generation level, existing methods rely mostly on building accurate mathematical models of combustion, or setting a large number of "if-then" rules based on expert experience. The mathematical model is seriously dependent on prior parameters such as fuel characteristics, hearth structures and the like, has poor adaptability when the raw materials fluctuate and the working conditions change frequently, and has high maintenance cost. Although the expert rule base has certain flexibility, it is difficult to exhaust all complex working conditions, conflicts among rules are difficult to coordinate, control actions tend to be lagged and stiff, and real-time and flexible optimization based on essential characteristics of combustion states cannot be realized. How to automatically extract the characteristics capable of representing the combustion coupling state from the multi-source heterogeneous data and rapidly match the approximately optimal control actions according to the characteristics is a difficult problem faced by the current technology. Disclosure of Invention The invention aims to provide a machine vision-based combustion optimization control method and system for a semi-coke oven, which are used for solving the problems in the background technology. In order to achieve the above purpose, the invention provides a combustion optimization control method of a semi-coke oven based on machine vision, which comprises the following steps: s1, generating a flame dynamic texture map through an image processing algorithm based on an image sequence of an acquired semi-coke oven combustion chamber in a set time period; S2, separating an infrared thermal imaging image subsequence from the image sequence, performing region gridding segmentation on the infrared thermal imaging image subsequence, calculating a statistical temperature value in each grid unit and a gradient of the statistical temperature value changing along with time, and constructing a temperature field time-space evolution matrix of a hearth based on the statistical temperature value and the gradient; S3, inputting the flame dynamic texture map and the temperature field space-time evolution matrix into a pre-trained coupling analysis network together, analyzing an implicit association mode between flame texture dynamic change and temperature field space-time evolution, and outputting a coupling characteristic tensor representing the current combustion coupling state; s4, performing similarity matching in a combustion mode knowledge graph based on the coupling characteristic tensor to acquire a reference control strategy most similar to the coupling characteristic tensor, wherein the combustion mode knowledge graph stores mapping relations between the coupling characteristic tensor and the optimal control strategy under various typical working conditions; S5, calculating the acquired real-time fuel supply flow data, primary air volume data and secondary air volume data of the semi-coke furnace through a multivariable coordinator based on the reference control strategy to obtain the adjustment quantity of the fuel quantity and the air volume set value, wherein the combustion state of the adjustment quantity approximates to the target coupling characteristic; s6, generating a control instruction set containing an execution time sequence according to the fuel quantity and the adjustment quantity of the air quantity set value so as to drive the semi-coke furnace executing mechanism to realize combustion optimization control. Preferably, the image sequence comprises a visible light image reflecting flame morphology and an infrared thermal imaging