Intelligent combustion control system and method for atmosphere furnace

Abstract

The present application discloses an atmosphere furnace intelligent combustion control system and method, which obtains the time queue of air flow value and the time queue of exhaust gas temperature value, and adopts data analysis and processing technology based on deep learning to perform time series correlation feature analysis and time series propagation aggregation of the air flow value and the exhaust gas temperature value, so as to adaptively control the air flow value at the next time point according to the global interaction representation feature between the time series propagation features of the air flow value and the exhaust gas temperature value. In this way, the air flow value can be calculated and adjusted more accurately to achieve the optimal atmosphere ratio, avoid under-oxygen or over-oxygen combustion, and reduce the fluctuation of operating parameters caused by untimely or inaccurate manual adjustment, thereby providing a more intelligent atmosphere furnace intelligent combustion control solution.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more specifically, to an atmosphere furnace intelligent combustion control system and method. Background Art

[0002] As a key heating equipment in the steel industry, atmosphere furnace plays a vital role in various material processing and heat treatment processes. Its combustion efficiency directly affects the energy consumption and production costs of enterprises. Therefore, reducing the energy consumption of atmosphere furnace is of great significance to improving the competitiveness of enterprises and meeting energy conservation and consumption reduction.

[0003] From the perspective of atmosphere furnace combustion control, it is a complex controlled object with nonlinearity, large time lag, and relatively slow time variation. Moreover, as the combustion proceeds, its state and characteristics will continue to change. With the change of working conditions and the adjustment of production links, atmosphere furnace combustion control is accompanied by frequent adjustments. However, traditional combustion control is based on manual adjustment by operators. The operation intensity of operators is relatively high during manual control, and they often lose sight of one thing while focusing on another. Adjustments are not made in time or in a timely manner, and manual control cannot maintain a good atmosphere ratio. There is insufficient or excessive combustion, resulting in low combustion efficiency and large fluctuations in operating parameters. In particular, if the atmosphere ratio is too low, incomplete combustion of the fuel will lead to energy loss, and the exhaust gas discharged from incompletely burned fuel will pollute the surrounding environment. If the atmosphere ratio is too high, the combustion-supporting atmosphere will take away too much heat, which will also cause a large amount of energy loss and affect the operation quality of the unit.

[0004] Therefore, an atmosphere furnace intelligent combustion control system is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an atmosphere furnace intelligent combustion control system and method, which obtains the time queue of air flow value and the time queue of exhaust gas temperature value, and adopts data analysis and processing technology based on deep learning to perform time series correlation feature analysis and time series propagation aggregation of the air flow value and the exhaust gas temperature value, so as to adaptively control the air flow value at the next time point according to the global interaction representation feature between the time series propagation characteristics of the air flow value and the exhaust gas temperature value. In this way, the air flow value can be calculated and adjusted more accurately to achieve the optimal atmosphere ratio, avoid underoxygen or overoxygen combustion, and reduce the fluctuation of operating parameters caused by untimely or inaccurate manual adjustment, thereby providing a more intelligent atmosphere furnace intelligent combustion control solution.

[0006] According to one aspect of the present application, an atmosphere furnace intelligent combustion control system is provided, which includes:

[0007] Atmosphere furnace combustion related data acquisition module, used to obtain the time queue of air flow value and the time queue of exhaust gas temperature value;

[0008] The atmosphere furnace combustion related data time series encoding module is used to input the time series of the air flow value and the time series of the exhaust gas temperature value into the sequence encoder respectively to obtain the sequence of the air flow time series associated feature vector and the sequence of the exhaust gas temperature time series associated feature vector;

[0009] The atmosphere furnace combustion parameter time series propagation aggregation module is used to input the sequence of the air flow time series associated feature vectors and the sequence of the exhaust gas temperature time series associated feature vectors into the time series feature forward aggregation network driven by node energy decay to obtain the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector;

[0010] An air flow-exhaust gas temperature global interaction module, used for decoupling the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector input features into a local interaction fine-grained aggregation network to obtain an air flow-exhaust gas temperature decoupling enhanced interaction global representation vector as an air flow-exhaust gas temperature decoupling enhanced interaction global representation feature;

[0011] The optimization instruction generation module is used to obtain the optimization instruction based on the global representation characteristics of the air flow-exhaust temperature decoupling reinforcement interaction, and the optimization instruction is used to indicate whether the air flow value at the next time point should increase, decrease or remain unchanged.

[0012] According to another aspect of the present application, an atmosphere furnace intelligent combustion control method is provided, which includes:

[0013] Obtaining a time queue of air flow values ​​and a time queue of exhaust gas temperature values;

[0014] Inputting the time queue of the air flow value and the time queue of the exhaust gas temperature value into a sequence encoder respectively to obtain a sequence of air flow time series associated feature vectors and a sequence of exhaust gas temperature time series associated feature vectors;

[0015] Inputting the sequence of the air flow time series associated feature vectors and the sequence of the exhaust gas temperature time series associated feature vectors into a time series feature forward aggregation network driven by node energy decay to obtain an air flow time series propagation aggregation representation vector and an exhaust gas temperature time series propagation aggregation representation vector;

[0016] Input the air flow rate time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector into a feature decoupling local interaction fine-grained aggregation network to obtain an air flow rate-exhaust gas temperature decoupling enhanced interaction global representation vector as an air flow rate-exhaust gas temperature decoupling enhanced interaction global representation feature;

[0017] Based on the global representation characteristics of the air flow-exhaust temperature decoupling enhanced interaction, an optimization instruction is obtained, and the optimization instruction is used to indicate whether the air flow value at the next time point should increase, decrease, or remain unchanged.

[0018] Compared with the prior art, the present application provides an atmosphere furnace intelligent combustion control system and method, which obtains the time queue of air flow value and the time queue of exhaust gas temperature value, and adopts data analysis and processing technology based on deep learning to perform time series correlation feature analysis and time series propagation aggregation of the air flow value and the exhaust gas temperature value, so as to adaptively control the air flow value at the next time point according to the global interaction representation feature between the time series propagation features of the air flow value and the exhaust gas temperature value. In this way, the air flow value can be calculated and adjusted more accurately to achieve the optimal atmosphere ratio, avoid underoxygen or overoxygen combustion, and reduce the fluctuation of operating parameters caused by untimely or inaccurate manual adjustment, thereby providing a more intelligent atmosphere furnace intelligent combustion control solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 is a block diagram of an atmosphere furnace intelligent combustion control system according to an embodiment of the present application;

[0021] Figure 2 A data flow diagram of an atmosphere furnace intelligent combustion control system according to an embodiment of the present application;

[0022] Figure 3 It is a flow chart of the atmosphere furnace intelligent combustion control method according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0025] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0026] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0027] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0028] Traditional combustion control is based on manual adjustment by operators. Operators have high operating intensity during manual control, and often lose sight of one thing while focusing on another. Adjustments are made too late or in a timely manner. Manual control cannot maintain a good atmosphere ratio, and there are problems such as insufficient or excessive oxygen combustion, resulting in low combustion efficiency, and large fluctuations in operating parameters. In particular, if the atmosphere ratio is too low, incomplete fuel combustion will lead to energy loss, and exhaust gas discharged from incompletely burned fuel will pollute the surrounding environment. If the atmosphere ratio is too high, the combustion-supporting atmosphere will take away too much heat, which will also cause a large amount of energy loss and affect the unit's operating quality. Therefore, an intelligent combustion control system for an atmosphere furnace is desired.

[0029] In the technical solution of the present application, an atmosphere furnace intelligent combustion control system is proposed. Figure 1 It is a block diagram of an atmosphere furnace intelligent combustion control system according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of atmosphere furnace intelligent combustion control system according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the atmosphere furnace intelligent combustion control system 300 includes: an atmosphere furnace combustion related data acquisition module 310, which is used to obtain the time queue of air flow values ​​and the time queue of exhaust gas temperature values; an atmosphere furnace combustion related data timing encoding module 320, which is used to input the time queue of air flow values ​​and the time queue of exhaust gas temperature values ​​into the sequence encoder respectively to obtain the sequence of air flow timing associated feature vectors and the sequence of exhaust gas temperature timing associated feature vectors; an atmosphere furnace combustion parameter timing propagation aggregation module 330, which is used to input the sequence of air flow timing associated feature vectors and the sequence of exhaust gas temperature timing associated feature vectors into the timing feature forward driven by node energy attenuation. Aggregation network to obtain air flow time series propagation aggregation representation vector and exhaust temperature time series propagation aggregation representation vector; air flow-exhaust temperature global interaction module 340, used to decouple the air flow time series propagation aggregation representation vector and the exhaust temperature time series propagation aggregation representation vector input features into the local interaction fine-grained aggregation network to obtain the air flow-exhaust temperature decoupling and enhanced interaction global representation vector as the air flow-exhaust temperature decoupling and enhanced interaction global representation feature; optimization instruction generation module 350, used to obtain optimization instructions based on the air flow-exhaust temperature decoupling and enhanced interaction global representation feature, and the optimization instructions are used to indicate that the air flow value at the next time point should increase, decrease or remain unchanged.

[0030] In particular, the atmosphere furnace combustion related data acquisition module 310 and the atmosphere furnace combustion related data timing encoding module 320 are used to obtain the time queue of air flow values ​​and the time queue of exhaust gas temperature values; and the time queue of air flow values ​​and the time queue of exhaust gas temperature values ​​are respectively input into a sequence encoder to obtain a sequence of air flow timing associated feature vectors and a sequence of exhaust gas temperature timing associated feature vectors. In a specific example of the present application, the time queue of air flow values ​​and the time queue of exhaust gas temperature values ​​are respectively input into a sequence encoder based on a 1D-CNN model to obtain a sequence of air flow timing associated feature vectors and a sequence of exhaust gas temperature timing associated feature vectors. Taking into account that the time queue of the air flow value and the time queue of the exhaust gas temperature value both have implicit correlation features within a local time segment, and the one-dimensional convolutional network has a good perception of local features in the time series, based on this, in the technical solution of the present application, the time queue of the air flow value and the time queue of the exhaust gas temperature value are respectively input into a sequence encoder based on the 1D-CNN model to capture and extract the temporal feature information of the local time, and obtain a sequence of air flow time series correlation feature vectors and a sequence of exhaust gas temperature time series correlation feature vectors.

[0031] In particular, the atmosphere furnace combustion parameter time series propagation aggregation module 330 is used to input the sequence of the air flow time series correlation feature vector and the sequence of the exhaust gas temperature time series correlation feature vector into the node energy decay driven time series feature forward aggregation network to obtain the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector. Considering that the different time series correlation features in the sequence of the air flow time series correlation feature vector and the sequence of the exhaust gas temperature time series correlation feature vector have different influences in the entire time period, therefore, in order to better adjust the features at different time points, so as to better integrate and aggregate the features in the entire time segment, in the technical solution of the present application, the sequence of the air flow time series correlation feature vector and the sequence of the exhaust gas temperature time series correlation feature vector are input into the node energy decay driven time series feature forward aggregation network to obtain the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector. In particular, the node energy decay driven time series feature forward aggregation network introduces an energy decay mechanism to correspondingly strengthen the features of each time point, thereby improving the multi-representation of the entire time feature. That is, the energy decay mechanism simulates the process of information propagation in a time series. As time goes by, the importance of early information gradually weakens, while the impact of recent information becomes more significant. This mechanism helps the model pay more attention to recent data and thus better capture the latest changing trends.

[0032] In an embodiment of the present application, taking the sequence of air flow time series associated feature vectors as an example, first, based on the mean and standard deviation of each air flow time series associated feature vector in the sequence of air flow time series associated feature vectors, the energy significant descriptors of each air flow time series associated feature vector are calculated to obtain a sequence of air flow time series energy significant descriptors; that is, the energy feature of each feature is quantified by the energy descriptor of each feature. Then, the timestamps of each air flow time series associated feature vector in the sequence of air flow time series associated feature vectors are extracted to obtain a sequence of air flow timestamps; by extracting the timestamp of each feature, the time information of the feature is obtained. Then, the air flow time series correlation feature vector corresponding to the current time point in the sequence of the air flow time series correlation feature vector is used as the current node feature vector and the air flow time series correlation feature vectors corresponding to other time points are used as the historical node feature vectors to obtain a sequence of the current air flow time series correlation feature vector and the historical air flow time series correlation feature vector; and the energy time decay factor of each historical air flow time series correlation feature vector in the sequence of the historical air flow time series correlation feature vector relative to the current air flow time series correlation feature vector is calculated to obtain a sequence of air flow energy time decay factors; here, the energy decay factor of each node is calculated based on the current node feature and the historical node feature, reflecting the characteristics of decay over time. Further, the sequence of the air flow time series energy significant descriptors is used as a positive adjustment factor and the sequence of the air flow energy time decay factor is used as a negative adjustment factor, and the sequence of the historical air flow time series associated feature vectors is subjected to time series propagation aggregation to obtain a historical air flow time series propagation aggregation representation feature vector; that is, the energy significant descriptor and the energy decay factor are used to perform weighted aggregation on the historical node feature vectors to highlight the information related to the current node energy significance to obtain a historical node propagation aggregation representation feature vector. Subsequently, the historical air flow time series propagation aggregation representation feature vector and the current air flow time series associated feature vector are fused to obtain the air flow time series propagation aggregation representation vector.

[0033] Among them, based on the mean and standard deviation of each air flow time series association feature vector in the sequence of the air flow time series association feature vector, the process of calculating the energy significant descriptors of each air flow time series association feature vector to obtain the sequence of air flow time series energy significant descriptors includes: respectively calculating the mean and standard deviation of the air flow time series association feature vector to obtain the air flow time series association feature mean and the air flow time series association feature standard deviation; calculating the positional difference between the air flow time series association feature vector and the air flow time series association feature mean to obtain the air flow time series association difference vector; calculating the fourth power of each eigenvalue in the air flow time series association difference vector to obtain the air flow time series association modulation difference vector; calculating the expected value of the air flow time series association modulation difference vector to obtain the air flow time series association expected value; dividing the air flow time series association expected value by the fourth power of the air flow time series association feature standard deviation to obtain the air flow time series energy significant descriptor.

[0034] More specifically, the process of calculating the energy time decay factor of each historical air flow time series associated feature vector in the sequence of the historical air flow time series associated feature vector relative to the current air flow time series associated feature vector to obtain a sequence of air flow energy time decay factors includes: respectively subtracting the air flow timestamp corresponding to the current air flow time series associated feature vector from the air flow timestamp corresponding to each historical air flow time series associated feature vector and rounding down to obtain a sequence of historical air flow time span values; calculating the square of the historical air flow time span value sequence divided by the time decay inverse scaling parameter to obtain a historical air flow time span square modulation value, and then multiplying the historical air flow time span square modulation value by the decay rate positive scaling parameter to obtain a sequence of air flow time series span energy decay coefficients; using the air flow time series span energy The energy attenuation coefficient of each air flow time series span in the sequence of attenuation coefficients is used as an exponential power, and the natural exponential function value with the natural constant e as the base is calculated to obtain the sequence of the air flow energy time attenuation factors; and, using the sequence of the air flow time series energy significant descriptors as positive adjustment factors and the sequence of the air flow energy time attenuation factors as negative adjustment factors, the sequence of the historical air flow time series associated feature vectors is subjected to time series propagation aggregation to obtain the historical air flow time series propagation aggregation representation feature vector, the process includes: dividing the sequence of the air flow time series energy significant descriptors by the sequence of the air flow energy time attenuation factors by position to obtain the sequence of air flow adjustment factors, and calculating the position-weighted sum of the sequence of the air flow adjustment factors and the sequence of the historical air flow time series associated feature vectors to obtain the historical air flow time series propagation aggregation representation feature vector.

[0035] In summary, in the above embodiment, the sequence of the air flow timing-related feature vectors and the sequence of the exhaust gas temperature timing-related feature vectors are input into the node energy decay-driven timing feature forward aggregation network to obtain the air flow timing propagation aggregation representation vector and the exhaust gas temperature timing propagation aggregation representation vector, including: inputting the sequence of the air flow timing-related feature vectors into the node energy decay-driven timing feature forward aggregation network, and processing them with the following forward aggregation formula to obtain the air flow timing propagation aggregation representation vector; wherein the forward aggregation formula is: ;in, is the sequence of the air flow time series associated feature vectors, and are the first and The air flow time series correlation feature vector, represents the number of feature vectors in the sequence of the air flow time series associated feature vectors, is the first in the sequence of the air flow time series associated feature vector The eigenvalues ​​of each position in the air flow time series associated eigenvector, To calculate the expected value, and Respectively The mean and variance of the air flow time series correlation feature vector, It is energy-significant descriptors of the air flow time series correlation feature vectors, is the number of eigenvalues ​​in each air flow time series associated eigenvector, It is The energy-significant descriptor of the air flow time series correlation feature vector, It is The energy-significant descriptor of the air flow time series correlation feature vector, and Distribution represents and timestamp of the air flow time series associated feature vector, To round down, is the time decay inverse scaling parameter, is the positive scaling parameter of the decay rate, and is the weight hyperparameter, Indicated by natural constant The exponential function value with base , is the air flow time series propagation aggregation representation vector. Similarly, the encoding method of the sequence of the exhaust gas temperature time series associated implicit feature vector is as shown in the above encoding.

[0036] In particular, the air flow-exhaust temperature global interaction module 340 is used to input the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector into a feature decoupled local interaction fine-grained aggregation network to obtain an air flow-exhaust temperature decoupled enhanced interaction global representation vector as an air flow-exhaust temperature decoupled enhanced interaction global representation feature. Considering that there is a temporal interaction relationship in the time dimension between the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector. In order to be able to understand the interactive influence of each local time period in a more fine-grained manner, in the technical solution of the present application, the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector are input into a feature decoupled local interaction fine-grained aggregation network to obtain an air flow-exhaust temperature decoupled enhanced interaction global representation vector. That is, the feature decoupled local interaction fine-grained aggregation network is a network that has a good ability to capture and perceive fine-grained semantic features between features.

[0037] In an embodiment of the present application, the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector are input into a feature decoupled local interaction fine-grained aggregation network to obtain an air flow-exhaust gas temperature decoupling enhanced interaction global representation vector as an air flow-exhaust gas temperature decoupling enhanced interaction global representation feature, including: first, feature decoupling the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector to obtain a sequence of air flow local time series propagation aggregation representation vectors and a sequence of exhaust gas temperature local time series propagation aggregation representation vectors, so as to enhance the ability to capture local time series feature interactions. Next, the local semantic association interaction vectors between each group of corresponding air flow local time series propagation aggregation representation vectors and exhaust gas temperature local time series propagation aggregation representation vectors in the sequence of the air flow local time series propagation aggregation representation vectors and the sequence of the exhaust gas temperature local time series propagation aggregation representation vectors are calculated to obtain a sequence of air flow-exhaust gas temperature local time series semantic association interaction vectors; and the fine-grained attention weights of each air flow-exhaust gas temperature local time series semantic association interaction vector in the sequence of the air flow-exhaust gas temperature local time series semantic association interaction vectors are calculated to obtain a sequence of air flow-exhaust gas temperature local time series fine-grained attention weights; that is, by calculating the local semantic interaction vectors between the sequences of two local vectors and calculating the fine-grained attention weights of each local semantic interaction vector, the important key information in the interaction vector is highlighted. Further, the sequence of the air flow-exhaust gas temperature local time series fine-grained attention weights is input into the random deactivation module to obtain a sequence of the air flow-exhaust gas temperature local time series fine-grained attention weights after deactivation processing, so as to achieve weight sparsification. In particular, the random dropout process retains the most important feature interaction information by randomly reducing a portion of the weights. Finally, the sequence of fine-grained attention weights of the air flow-exhaust temperature local temporal series after the deactivation process is used as the weight distribution, and the weighted sum of the sequence of the air flow-exhaust temperature local temporal series semantic association interaction vectors is calculated to obtain the air flow-exhaust temperature decoupled reinforcement interaction global representation vector.

[0038] Among them, the process of calculating the fine-grained attention weights of each air flow-exhaust temperature local temporal semantic association interaction vector in the sequence of air flow-exhaust temperature local temporal semantic association interaction vectors to obtain a sequence of air flow-exhaust temperature local temporal fine-grained attention weights includes: multiplying each air flow-exhaust temperature local temporal semantic association interaction vector in the sequence of air flow-exhaust temperature local temporal semantic association interaction vectors with a weight matrix, and then adding the obtained vector to the bias vector by position to obtain a sequence of air flow-exhaust temperature local temporal semantic association interaction modulation vectors; inputting the sequence of air flow-exhaust temperature local temporal semantic association interaction modulation vectors into a tanh function for activation processing to obtain a sequence of air flow-exhaust temperature local temporal semantic association interaction activation vectors; multiplying each air flow-exhaust temperature local temporal semantic association interaction activation vector in the sequence of air flow-exhaust temperature local temporal semantic association interaction activation vectors with a weight coefficient vector to obtain a sequence of air flow-exhaust temperature local temporal fine-grained attention weights.

[0039] More specifically, the process of using the sequence of fine-grained attention weights of the air flow-exhaust temperature local time series after the deactivation process as the weight distribution and calculating the weighted sum of the sequence of the air flow-exhaust temperature local time series semantic association interaction vectors to obtain the air flow-exhaust temperature decoupling reinforcement interaction global representation vector includes: respectively calculating the position difference, position dot multiplication and position addition between the air flow local time series propagation aggregation representation vector and the exhaust temperature local time series propagation aggregation representation vector to obtain the air flow-exhaust temperature local time series propagation aggregation difference vector, ... representation vector, the air flow-exhaust temperature local time series propagation aggregation representation vector, the air flow-exhaust temperature local time series propagation aggregation representation vector, the air flow-exhaust temperature local time series propagation aggregation representation vector, the air flow-exhaust temperature local time series propagation aggregation representation vector, the air flow- The air flow-exhaust temperature local time series propagation aggregation point product vector and the air flow-exhaust temperature local time series propagation aggregation sum vector are cascaded and one-dimensional convolution encoding is performed to obtain the air flow-exhaust temperature local time series multi-dimensional fusion vector; the air flow-exhaust temperature local time series multi-dimensional fusion vector is subjected to local window-based maximum pooling processing to obtain the air flow-exhaust temperature local time series semantic association interaction vector.

[0040] In particular, the optimization instruction generation module 350 is used to obtain optimization instructions based on the global representation characteristics of the air flow-exhaust temperature decoupling enhancement interaction, and the optimization instructions are used to indicate that the air flow value at the next time point should increase, decrease or remain unchanged. In a specific example of the present application, a time queue of air flow values ​​and a time queue of exhaust gas temperature values ​​are obtained; the time queue of air flow values ​​and the time queue of exhaust gas temperature values ​​are respectively input into a sequence encoder to obtain a sequence of air flow timing-related feature vectors and a sequence of exhaust gas temperature timing-related feature vectors; the sequence of air flow timing-related feature vectors and the sequence of exhaust gas temperature timing-related feature vectors are input into a node energy decay-driven timing feature forward aggregation network to obtain an air flow timing propagation aggregation representation vector and an exhaust gas temperature timing propagation aggregation representation vector; the air flow timing propagation aggregation representation vector and the exhaust gas temperature timing propagation aggregation representation vector are input into a feature decoupling local interaction fine-grained aggregation network to obtain an air flow-exhaust gas temperature decoupling enhanced interaction global representation vector as an air flow-exhaust gas temperature decoupling enhanced interaction global representation feature; based on the air flow-exhaust gas temperature decoupling enhanced interaction global representation feature, an optimization instruction is obtained, and the optimization instruction is used to indicate that the air flow value at the next time point should increase, decrease, or remain unchanged. That is, the global representation features of air flow-exhaust gas temperature decoupling enhanced interaction obtained by local interaction of the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector are classified and processed to adaptively control the air flow value at the next time point. In this way, the air flow value can be calculated and adjusted more accurately to achieve the optimal atmosphere ratio, avoid under-oxygen or over-oxygen combustion, and reduce the fluctuation of operating parameters caused by untimely or inaccurate manual adjustment, thereby providing a more intelligent atmosphere furnace intelligent combustion control system.

[0041] In a preferred example, the applicant of the present application takes into account the forward aggregation characteristics of the one-dimensional time series local correlation characteristics of the air flow time series propagation aggregation representation vector and the exhaust temperature time series propagation aggregation representation vector, which respectively represent the air flow value and the exhaust temperature value, driven by the energy attenuation of the global time domain time series nodes. When performing feature decoupled local interactive fine-grained aggregation on them, the full time domain forward aggregation difference caused by the source time series distribution difference will further cause the local interactive aggregation fine-grained alignment conflict based on feature time series decoupling, thereby causing the loss of aggregation key information and affecting the expression effect of the air flow-exhaust temperature decoupling enhanced interactive global representation vector.

[0042] Based on this, in the preferred example, when the air flow-exhaust gas temperature decoupling enhanced interaction global representation vector is input into the classifier-based air flow optimizer to obtain an optimization instruction, the air flow-exhaust gas temperature decoupling enhanced interaction global representation vector is optimized. Specifically, the optimization process includes the following steps:

[0043] Clustering all eigenvalues ​​of the air flow-exhaust temperature decoupling enhanced interaction global representation vector based on the distance between eigenvalues, such as the difference absolute value distance, and determining the ratio of the number of features of clustered features to the number of features of the air flow-exhaust temperature decoupling enhanced interaction global representation vector;

[0044] Arranging the clustering features into an air flow-exhaust temperature decoupling enhanced interaction global clustering vector, and dividing the second norm of the air flow-exhaust temperature decoupling enhanced interaction global clustering vector by the second norm of the air flow-exhaust temperature decoupling enhanced interaction global representation vector to obtain a first air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value;

[0045] Dividing a first power value of a norm of the air flow-exhaust temperature decoupling enhanced interaction global clustering vector with the scale value as an exponent by a second power value of a norm of the air flow-exhaust temperature decoupling enhanced interaction global representation vector with the scale value as an exponent to obtain a second air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value;

[0046] For each eigenvalue of the air flow-exhaust temperature decoupling enhanced interaction global clustering vector, multiply it by the inverse of the difference between the first air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value and the second air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value to obtain an optimized eigenvalue of the air flow-exhaust temperature decoupling enhanced interaction global clustering vector;

[0047] For each eigenvalue outside the cluster in the air flow-exhaust temperature decoupling enhanced interaction global representation vector, multiply it by the inverse of the sum of the first air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value and the second air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value to obtain an optimized out-of-class eigenvalue of the air flow-exhaust temperature decoupling enhanced interaction global representation vector; and

[0048] The optimized eigenvalues ​​of the air flow-exhaust temperature decoupling enhanced interaction global clustering vector and the optimized out-of-class eigenvalues ​​of the air flow-exhaust temperature decoupling enhanced interaction global representation vector are combined to form an optimized air flow-exhaust temperature decoupling enhanced interaction global representation vector.

[0049] That is, the optimization process is expressed as: ;in, is the global representation vector of the air flow-exhaust temperature decoupling enhancement interaction, is the number of features of the global representation vector of the air flow-exhaust temperature decoupling enhancement interaction, is the global clustering vector of the air flow-exhaust temperature decoupling reinforcement interaction, is the number of features of the global clustering vector of the air flow-exhaust temperature decoupling enhancement interaction, represents the clustering feature set corresponding to the global clustering vector of the air flow-exhaust temperature decoupling enhancement interaction, and Represents the binary norm and the uninorm of the vector respectively. Power, is the air flow-exhaust temperature decoupling enhanced interaction global clustering related weight value.

[0050] Here, in order to avoid the loss of key suffix semantic information relative to the original feature set as a whole due to aggregation conflict caused by the air flow-exhaust temperature decoupling and enhanced interaction global representation vector being based on aggregated features, the clustering ratio of the number of features of the air flow-exhaust temperature decoupling and enhanced interaction global clustering vector relative to the number of features of the air flow-exhaust temperature decoupling and enhanced interaction global representation vector is used as a decision function to perform an adversarial decision on the set absolute representation of the one-norm of the air flow-exhaust temperature decoupling and enhanced interaction global clustering vector and the air flow-exhaust temperature decoupling and enhanced interaction global representation vector, and the air flow-exhaust temperature decoupling and enhanced interaction global clustering vector and the air flow-exhaust temperature decoupling and enhanced interaction global representation vector are respectively compared. The clustering inherent conflict representation of the two norms of the air flow-exhaust temperature decoupling enhanced interaction global representation vector is positively and negatively interacted to construct the optimized air flow-exhaust temperature decoupling enhanced interaction global representation vector based on the aggregated features and the original feature set as a whole. The solid alignment guardrail is used to achieve the mitigation of the harmful information loss intention of the optimized air flow-exhaust temperature decoupling enhanced interaction global representation vector based on the aggregated risk transferability, and the expression effect of the optimized air flow-exhaust temperature decoupling enhanced interaction global representation vector is improved, thereby improving the accuracy of the optimization instructions obtained by the air flow-exhaust temperature decoupling enhanced interaction global representation vector input based on the classifier-based air flow optimizer. In this way, the air flow value can be calculated and adjusted more accurately to achieve the optimal atmosphere ratio, avoid underoxygen or overoxygen combustion, and reduce the fluctuation of operating parameters caused by untimely or inaccurate manual adjustment, thereby providing a more intelligent atmosphere furnace intelligent combustion control system.

[0051] As described above, the atmosphere furnace intelligent combustion control system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with an atmosphere furnace intelligent combustion control algorithm. In a possible implementation, the atmosphere furnace intelligent combustion control system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the atmosphere furnace intelligent combustion control system 300 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the atmosphere furnace intelligent combustion control system 300 can also be one of the many hardware modules of the wireless terminal.

[0052] Alternatively, in another example, the atmosphere furnace intelligent combustion control system 300 and the wireless terminal may also be separate devices, and the atmosphere furnace intelligent combustion control system 300 may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0053] Furthermore, an atmosphere furnace intelligent combustion control method is also provided.

[0054] Figure 3 FIG. 1 is a flow chart of an atmosphere furnace intelligent combustion control method according to an embodiment of the present application. Figure 3 As shown, according to the embodiment of the present application, the intelligent combustion control method of the atmosphere furnace includes the following steps: S1, obtaining a time queue of air flow values ​​and a time queue of exhaust gas temperature values; S2, inputting the time queue of the air flow values ​​and the time queue of the exhaust gas temperature values ​​into a sequence encoder respectively to obtain a sequence of air flow time series associated feature vectors and a sequence of exhaust gas temperature time series associated feature vectors; S3, inputting the sequence of air flow time series associated feature vectors and the sequence of exhaust gas temperature time series associated feature vectors into a node energy decay driven time series feature forward aggregation network to obtain air flow time series associated feature vectors. S4, decoupling the air flow time series propagation aggregation representation vector and the exhaust temperature time series propagation aggregation representation vector into the feature decoupling local interaction fine-grained aggregation network to obtain the air flow-exhaust temperature decoupling enhanced interaction global representation vector as the air flow-exhaust temperature decoupling enhanced interaction global representation feature; S5, based on the air flow-exhaust temperature decoupling enhanced interaction global representation feature, obtaining an optimization instruction, which is used to indicate whether the air flow value at the next time point should increase, decrease or remain unchanged.

[0055] In summary, the intelligent combustion control method of the atmosphere furnace according to the embodiment of the present application is explained, which obtains the time queue of the air flow value and the time queue of the exhaust gas temperature value, and adopts data analysis and processing technology based on deep learning to perform time series correlation feature analysis and time series propagation aggregation of the air flow value and the exhaust gas temperature value, so as to adaptively control the air flow value at the next time point according to the global interaction representation feature between the time series propagation features of the air flow value and the exhaust gas temperature value. In this way, the air flow value can be calculated and adjusted more accurately to achieve the optimal atmosphere ratio, avoid underoxygen or overoxygen combustion, and reduce the fluctuation of operating parameters caused by untimely or inaccurate manual adjustment, thereby providing a more intelligent atmosphere furnace intelligent combustion control solution.

[0056] In the technical solution of this application, the technical route and requirements for safe furnace burning research are as follows:

[0057] Vault temperature protection function: When the vault temperature exceeds the vault safety setting value, the adaptive model automatically reduces the opening of the gas regulating valve to ensure safe production of the blast furnace hot blast stove.

[0058] Exhaust gas temperature protection function: The adaptive model burns to the set exhaust gas temperature within the set time. If the hot blast furnace is not replaced due to the blast furnace, the adaptive model automatically reduces the opening of the air regulating valve and the gas regulating valve to ensure the safe production of the blast furnace hot blast furnace.

[0059] The adaptive model needs to design a safe opening limit setting function for the normal operation of the hot air furnace for each control loop to minimize the operating frequency and amplitude of the regulating valve and ensure the service life of the valve and burner.

[0060] When the gas / air pressure is lower than the lower limit, a voice alarm will be issued. When the gas / air pressure is lower than the lower limit, the system will automatically close the valve. Air interlock gas. When the air volume is insufficient, the adaptive model will automatically reduce the gas to prevent safety accidents caused by excessive gas. When the fan trips, the adaptive model will also automatically interlock and close the gas valve.

[0061] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. An atmosphere furnace intelligent combustion control system, characterized in that: include: Atmosphere furnace combustion related data acquisition module, used to obtain the time queue of air flow value and the time queue of exhaust gas temperature value; The atmosphere furnace combustion related data time series encoding module is used to input the time series of the air flow value and the time series of the exhaust gas temperature value into the sequence encoder respectively to obtain the sequence of the air flow time series associated feature vector and the sequence of the exhaust gas temperature time series associated feature vector; The atmosphere furnace combustion parameter time series propagation aggregation module is used to input the sequence of the air flow time series associated feature vectors and the sequence of the exhaust gas temperature time series associated feature vectors into the time series feature forward aggregation network driven by node energy decay to obtain the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector; An air flow-exhaust gas temperature global interaction module, used for decoupling the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector input features into a local interaction fine-grained aggregation network to obtain an air flow-exhaust gas temperature decoupling enhanced interaction global representation vector as an air flow-exhaust gas temperature decoupling enhanced interaction global representation feature; An optimization instruction generation module, used to obtain an optimization instruction based on the global representation feature of the air flow-exhaust temperature decoupling reinforcement interaction, wherein the optimization instruction is used to indicate that the air flow value at the next time point should increase, decrease, or remain unchanged; Wherein, the atmosphere furnace combustion parameter time series propagation aggregation module includes: an air flow energy significant descriptor calculation unit, configured to calculate the energy significant descriptors of each air flow time series associated feature vector in the sequence of air flow time series associated feature vectors based on the mean and standard deviation of each air flow time series associated feature vector in the sequence of air flow time series associated feature vectors to obtain a sequence of air flow time series energy significant descriptors; An air flow timestamp extraction unit, used for extracting the timestamp of each air flow time series associated feature vector in the sequence of air flow time series associated feature vectors to obtain a sequence of air flow timestamps; An air flow time series associated node feature extraction unit, used to use the air flow time series associated feature vector corresponding to the current time point in the sequence of the air flow time series associated feature vector as the current node feature vector and use the air flow time series associated feature vectors corresponding to other time points as the historical node feature vectors to obtain a sequence of the current air flow time series associated feature vector and the historical air flow time series associated feature vector; An air flow energy time decay factor calculation unit, used to calculate the energy time decay factor of each historical air flow time series associated feature vector in the sequence of historical air flow time series associated feature vectors relative to the current air flow time series associated feature vector to obtain a sequence of air flow energy time decay factors; A historical air flow time series propagation unit, configured to perform time series propagation aggregation on the sequence of historical air flow time series associated feature vectors using the sequence of the air flow time series energy significant descriptors as a positive adjustment factor and the sequence of the air flow energy time decay factors as a negative adjustment factor to obtain a historical air flow time series propagation aggregation representation feature vector; The air flow time series propagation aggregation representation unit is used to fuse the historical air flow time series propagation aggregation representation feature vector and the current air flow time series association feature vector to obtain the air flow time series propagation aggregation representation vector.

2. The atmosphere furnace intelligent combustion control system according to claim 1 is characterized in that: The atmosphere furnace combustion-related data timing encoding module is used to: input the time queue of the air flow value and the time queue of the exhaust gas temperature value into a sequence encoder based on the 1D-CNN model respectively to obtain a sequence of the air flow timing-related feature vectors and a sequence of the exhaust gas temperature timing-related feature vectors.

3. The atmosphere furnace intelligent combustion control system according to claim 2 is characterized in that: The air flow energy significant descriptor calculation unit is used to: Calculating the mean and standard deviation of the air flow time series correlation feature vector respectively to obtain the air flow time series correlation feature mean and the air flow time series correlation feature standard deviation; Calculating the position difference between the air flow time series correlation feature vector and the air flow time series correlation feature mean to obtain an air flow time series correlation difference vector; Calculating the fourth power of each eigenvalue in the air flow time series correlation difference vector to obtain an air flow time series correlation modulation difference vector; Calculating the expected value of the air flow timing-related modulation difference vector to obtain the air flow timing-related expected value; The air flow time series associated expected value is divided by the fourth power of the air flow time series associated characteristic standard deviation to obtain the air flow time series energy significant descriptor.

4. The atmosphere furnace intelligent combustion control system according to claim 3 is characterized in that: The air flow energy time attenuation factor calculation unit is used to: Subtracting the air flow timestamp corresponding to the current air flow time series associated feature vector from the air flow timestamp corresponding to each historical air flow time series associated feature vector and rounding down to obtain a sequence of historical air flow time span values; After calculating the square of the sequence of the historical air flow time span values ​​divided by the time attenuation inverse scaling parameter by position to obtain the historical air flow time span square modulation value, the historical air flow time span square modulation value is multiplied by the decay rate positive scaling parameter by position to obtain the sequence of air flow time series span energy decay coefficients; Taking each air flow time series span energy attenuation coefficient in the sequence of the air flow time series span energy attenuation coefficient as an exponential power, a natural exponential function value with a natural constant e as a base is calculated to obtain the sequence of the air flow energy time attenuation factor.

5. The atmosphere furnace intelligent combustion control system according to claim 4 is characterized in that: The historical air flow time series propagation unit is used to: divide the sequence of the air flow time series energy significance descriptors and the sequence of the air flow energy time attenuation factors by position to obtain a sequence of air flow adjustment factors, and calculate the position weighted sum of the sequence of the air flow adjustment factors and the sequence of the historical air flow time series associated feature vectors to obtain the historical air flow time series propagation aggregation representation feature vector.

6. The atmosphere furnace intelligent combustion control system according to claim 5 is characterized in that: The air flow-exhaust gas temperature global interaction module includes: a feature decoupling unit, configured to perform feature decoupling on the air flow time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector to obtain a sequence of air flow local time series propagation aggregation representation vectors and a sequence of exhaust gas temperature local time series propagation aggregation representation vectors; An air flow-exhaust gas temperature local time series semantic association interaction unit, used to calculate the local semantic association interaction vector between each group of corresponding air flow local time series propagation aggregation representation vectors and exhaust gas temperature local time series propagation aggregation representation vectors in the sequence of the air flow local time series propagation aggregation representation vectors and the sequence of the exhaust gas temperature local time series propagation aggregation representation vectors to obtain a sequence of air flow-exhaust gas temperature local time series semantic association interaction vectors; An air flow-exhaust gas temperature local temporal series fine-grained attention weight calculation unit, used to calculate the fine-grained attention weight of each air flow-exhaust gas temperature local temporal series semantic association interaction vector in the sequence of air flow-exhaust gas temperature local temporal series semantic association interaction vectors to obtain a sequence of air flow-exhaust gas temperature local temporal series fine-grained attention weights; A deactivation unit, used for inputting the sequence of air flow-exhaust gas temperature local time series fine-grained attention weights into a random deactivation module to obtain a sequence of air flow-exhaust gas temperature local time series fine-grained attention weights after deactivation processing; An air flow-exhaust temperature decoupling and enhanced global interaction unit, used to calculate the weighted sum of the sequence of air flow-exhaust temperature local temporal sequence semantic association interaction vectors after the deactivation process as the weight distribution to obtain the air flow-exhaust temperature decoupling and enhanced interaction global representation vector; The air flow-exhaust gas temperature local time series fine-grained attention weight calculation unit is used to: After multiplying each air flow-exhaust temperature local time series semantic association interaction vector in the sequence of air flow-exhaust temperature local time series semantic association interaction vectors by a weight matrix, the obtained vector is added to the bias vector by position to obtain a sequence of air flow-exhaust temperature local time series semantic association interaction modulation vectors; Inputting the sequence of air flow-exhaust gas temperature local temporal semantic association interaction modulation vectors into a tanh function for activation processing to obtain a sequence of air flow-exhaust gas temperature local temporal semantic association interaction activation vectors; Each air flow-exhaust temperature local temporal semantic association interaction activation vector in the sequence of the air flow-exhaust temperature local temporal semantic association interaction activation vectors is multiplied by a weight coefficient vector to obtain a sequence of the air flow-exhaust temperature local temporal fine-grained attention weights.

7. The atmosphere furnace intelligent combustion control system according to claim 6 is characterized in that: The air flow-exhaust gas temperature local time series semantic association interaction unit is used to: Respectively calculating the position difference, position dot product and position addition between the air flow local time series propagation aggregate representation vector and the exhaust gas temperature local time series propagation aggregate representation vector to obtain an air flow-exhaust gas temperature local time series propagation aggregate difference vector, an air flow-exhaust gas temperature local time series propagation aggregate dot product vector and an air flow-exhaust gas temperature local time series propagation aggregate sum vector; The air flow-exhaust gas temperature local time series propagation aggregation difference vector, the air flow-exhaust gas temperature local time series propagation aggregation point product vector and the air flow-exhaust gas temperature local time series propagation aggregation sum vector are cascaded and then one-dimensional convolutional encoded to obtain an air flow-exhaust gas temperature local time series multi-dimensional fusion vector; The air flow-exhaust gas temperature local time series multi-dimensional fusion vector is subjected to a local window-based maximum pooling process to obtain the air flow-exhaust gas temperature local time series semantic association interaction vector.

8. The atmosphere furnace intelligent combustion control system according to claim 7 is characterized in that: The optimization instruction generation module is used to: input the air flow-exhaust temperature decoupling enhancement interaction global representation vector into the classifier-based air flow optimizer to obtain the optimization instruction, and the optimization instruction is used to indicate that the air flow value at the next time point should increase, decrease or remain unchanged.

9. An atmosphere furnace intelligent combustion control method, using the atmosphere furnace intelligent combustion control system according to claim 1, characterized in that: include: Obtaining a time queue of air flow values ​​and a time queue of exhaust gas temperature values; Inputting the time queue of the air flow value and the time queue of the exhaust gas temperature value into a sequence encoder respectively to obtain a sequence of air flow time series associated feature vectors and a sequence of exhaust gas temperature time series associated feature vectors; Inputting the sequence of the air flow time series associated feature vectors and the sequence of the exhaust gas temperature time series associated feature vectors into a time series feature forward aggregation network driven by node energy decay to obtain an air flow time series propagation aggregation representation vector and an exhaust gas temperature time series propagation aggregation representation vector; Input the air flow rate time series propagation aggregation representation vector and the exhaust gas temperature time series propagation aggregation representation vector into a feature decoupling local interaction fine-grained aggregation network to obtain an air flow rate-exhaust gas temperature decoupling enhanced interaction global representation vector as an air flow rate-exhaust gas temperature decoupling enhanced interaction global representation feature; Based on the global representation characteristics of the air flow-exhaust temperature decoupling enhanced interaction, an optimization instruction is obtained, and the optimization instruction is used to indicate whether the air flow value at the next time point should increase, decrease, or remain unchanged.

Images