Burning-through point dynamic prediction method based on time cascade width echo state learning system

Abstract

The invention provides a sintering end point dynamic prediction method based on a time cascade width echo state learning system, and relates to the field of production energy conservation and consumption reduction in the steel sintering process. Determining process parameters which directly influence the sintering end point, wherein the process parameters comprise the temperature of a 17 # air box, the temperature of a 18 # air box, the temperature of a 19 # air box, the temperature of a 20 # air box, the temperature of a 21 # air box, the temperature of a 22 # air box, the temperature of BTP, the thickness of a material layer, the trolley speed and the vertical combustion speed; secondly, providing an echo state learning system based on time cascade width as a BTP dynamic prediction method; and finally, according to actual production data, performing BTP dynamic prediction by taking process parameters which influence direct correlation of the BTP as input and the BTP as output. According to the dynamic prediction method provided by the invention, accurate dynamic prediction of the BTP can be realized, and a foundation is laid for realizing green manufacturing and intelligent manufacturing in the iron and steel industry.

Description

Technical Field

[0001] The invention relates to the field of energy saving and consumption reduction in steel sintering process production, and in particular to a sintering end point dynamic prediction method based on a time cascade width echo state learning system. Background Art

[0002] The steel industry is an important basic industry and supporting industry of the national economy. The development level of steel enterprises represents the development level of a country's industrial productivity. How to effectively use energy and reduce pollution and production costs of steel enterprises has become an urgent problem for major steel enterprises.

[0003] The sintering process is an important part of the iron and steel metallurgical process. The quality and output of the sintered ore produced by it not only directly affect the output, quality and energy consumption of the blast furnace ironmaking process, but also play a vital role in achieving good economic and technical indicators and technological progress in the blast furnace ironmaking process. There are many state parameters and operating parameters of the sintering process, which are of great significance to the state of the reaction sintering equipment and the sintering conditions. The sintering endpoint (BTP) is a parameter that has a great relationship with the output, quality and cost of the sintering machine during the sintering process. Controlling the sintering endpoint at a reasonable position is an important link to ensure that the sintering material layer completes various reactions during the movement of the sintering machine, and it can also ensure that the effective sintering area of ​​the sintering machine is fully utilized. If the actual position of the sintering endpoint is less than its set position, the material layer has been over-burned, which means that the effective sintering area of ​​the sintering machine is not fully utilized, and the grate bars on the trolley will also be damaged to a certain extent; if the actual position of the sintering endpoint is greater than its set position, the material layer is under-burned, that is, the material layer is unloaded before it is completely burned through, which reduces the quality and output of the sintered ore. Therefore, how to stabilize the sintering end point to further improve the output and quality of sintered ore is a key issue that steel companies need to solve. It can be seen that in order to improve energy utilization, reduce pollution emissions, and achieve green manufacturing, it is necessary to achieve accurate dynamic prediction of BTP, which is also a key issue to improve energy utilization in the steel industry, reduce pollution emissions, and achieve green manufacturing and intelligent manufacturing. Summary of the invention

[0004] In order to solve the problems of serious pollution and low energy utilization in the sintering process, the present invention provides a sintering endpoint dynamic prediction method based on a time cascade width echo state learning system, which mainly includes the following steps:

[0005] S1: Considering the combustion mechanism of the sintering process, the process parameters related to BTP were determined using the Spearman rank correlation coefficient: 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical combustion speed;

[0006] S2: Establishing and training a BTP dynamic prediction model based on a time cascade width echo state learning system; the BTP dynamic prediction model includes a feature node layer and a reservoir;

[0007] S3: By using actual production data, taking the process parameters affecting BTP as input and BTP as output, the trained BTP dynamic prediction model is used to perform dynamic prediction of BTP.

[0008] The beneficial effects brought by the technical solution provided by the present invention are:

[0009] Based on the sintering process mechanism analysis and the Spearman rank correlation coefficient, the present invention can determine the process parameters that directly affect BTP: 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical combustion speed; the present invention proposes a time cascade width echo state learning system, through which a BTP dynamic prediction model is established and trained; according to the trained model, the dynamic characteristics of sintering production can be determined in real time, and BTP dynamic prediction can be realized, thereby laying the foundation for realizing green manufacturing and intelligent manufacturing in the steel industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0011] Figure 1 is a flow chart of a method for dynamic prediction of sintering endpoint based on a time cascade width echo state learning system in an embodiment of the present invention;

[0012] Figure 2 This is a block diagram of the BTP prediction model in an embodiment of the present invention;

[0013] Figure 3 is a comparison chart of the BTP dynamic prediction result and the actual value in the embodiment of the present invention;

[0014] Figure 4 is a prediction error result diagram of the BTP dynamic prediction result in an embodiment of the present invention;

[0015] Figure 5 It is a schematic diagram of the operation of the hardware device in the embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0017] Example 1

[0018] Please refer to Figure 1 , Figure 1 : is a flow chart of a method for dynamic prediction of sintering endpoint based on a time cascade width echo state learning system in an embodiment of the present invention, the method comprising:

[0019] S1: The sintering process is a kind of production process with complex chemical and physical changes. The process parameters that affect BTP are mainly related to production. Considering the combustion mechanism of the sintering process, the process parameters related to BTP are determined using the Spearman rank correlation coefficient: 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical combustion speed; while 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical burning speed are directly related to the production status. Therefore, 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical burning speed are used as inputs of the BTP dynamic prediction model.

[0020] S2: Establish and train a BTP dynamic prediction model based on the time cascade width echo state learning system;

[0021] Please refer to Figure 2 , Figure 2 The BTP prediction model principle block diagram in this embodiment, the BTP dynamic prediction model includes characteristic node layers and reservoirs, and its structure can better reflect the dynamic characteristics of the sintering process. The proposed time cascade width echo state learning system can obtain the dynamic characteristics related to the production state in real time, laying the foundation for establishing an accurate and effective BTP prediction model.

[0022] The process of establishing a BTP dynamic model is as follows:

[0023] S2.1: Collect the production data corresponding to BTP and process parameters affecting BTP to form a training sample data set {(X,Y)|X∈R N×11 ,Y∈R N×1}, where X={x1,...,x o ,...,x N} and Y={y1,...,y o ,...,y N}, x o is the oth training sample, y ois the oth BTP, N represents the total number of samples in the training sample data set; in the time cascade width echo state learning system, the initial output set of the feature node is determined as:

[0024]

[0025] Where F1 is the output of the first set of feature nodes; ψ() is the first activation function; is the weight vector of the first group; is the deviation vector of the first group;

[0026] The output F2 of the second set of feature nodes is generated based on the output F1 of the previous set of nodes, expressed as:

[0027]

[0028] in, is the weight vector of the second group; is the bias vector of the second group;

[0029] The temporal cascading mechanism facilitates the transfer of feature node data from one group to the next;

[0030] Similarly, define the output of all n groups of feature nodes and determine the output F of the last feature node n It is expressed as:

[0031]

[0032] Among them, F n-1 It is the output of the n-1th group of feature nodes; is the weight vector of the nth group; is the deviation vector of the nth group;

[0033] The interconnection between feature nodes in the feature node layer is as follows:

[0034] F n =[F1,F2,...,F n ]

[0035] S2.2: Output F of feature node layer n is forwarded to the initial reservoir in the reservoir, where each reservoir not only contributes its unique output but also passes its current state to the subsequent reservoir; assuming there are m reservoirs, the state of the jth reservoir at time t is H j (t) The update equation is:

[0036]

[0037] Among them, ζ(·) is the second excitation function, and denote the input weight matrix and weight matrix of the j-th reservoir, respectively. represents the connection weight between the j-1th reservoir and the jth reservoir, H j (t-1) represents the state of the jth reservoir at time t-1;

[0038] By updating the states between multiple reservoirs, the output matrix of all reservoirs is obtained as follows:

[0039] H m =[H1,H2,...,H m ]

[0040] S2.3: Combine the output of the feature node layer with the output of all reservoir layers to obtain:

[0041] A=[F1,F2,...,F|H1,H2,...,H m ]=[F n |H m ],

[0042] S2.4: The BTP dynamic prediction model formula obtained based on the time cascade width echo state learning system is:

[0043] Y=[F n |H m ]W,

[0044] Among them, Y represents BTP and W represents the weight matrix connecting the hidden layer nodes to the output layer.

[0045] The calculation formula of the weight matrix W connecting the hidden layer nodes to the output layer is:

[0046] W=(A T A+λI) -1 A T Y,

[0047] where I represents the identity matrix with appropriate dimension, λ represents a non-negative regularization constant, and Y represents the BTP.

[0048] Use the pseudo-inverse algorithm to solve the weight matrix W. When λ approaches 0,

[0049]

[0050] Among them, A + represents the pseudo-inverse matrix of A;

[0051] By adding A + Substituting into the calculation formula of the weight matrix W, the weight matrix W can be obtained.

[0052] The process parameter samples in the test data set in step S2 are input into the BTP dynamic prediction model to obtain the predicted value of BTP, and the predicted value of BTP is compared with the actual value of BTP corresponding to the sintering parameter sample to obtain the following: Figure 3 The results shown by Figure 3 It can be seen that the predicted value of BTP can follow the actual value well. Figure 3 The prediction error between the BTP prediction value and the actual value is as follows: Figure 4 As shown by Figure 3 and Figure 4 It can be seen that the prediction error of the BTP model is within [-0.4, 0.2], and most of the prediction results are within [-0.2, 0.2]. Therefore, the BTP dynamic prediction model can accurately perform dynamic prediction of BTP, meet the production requirements of the actual sintering process, and is conducive to promoting green manufacturing and intelligent manufacturing of the sintering process.

[0053] S3: Obtain the actual production data of 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed, vertical combustion speed and BTP. By using the actual production data, taking the process parameters affecting BTP as input and BTP as output, the trained BTP dynamic prediction model is used to perform dynamic prediction of BTP.

[0054] Example 2

[0055] A sintering process carbon consumption adaptive weighted width echo state learning system modeling device 501, such as Figure 5 As shown, it includes: a processor 502 and a storage device 503; the processor 502 loads and executes the instructions and data in the storage device 503 to implement the sintering process carbon consumption adaptive weighted width echo state learning system modeling method.

[0056] Example 3

[0057] A storage device stores instructions and data for implementing the sintering process carbon consumption adaptive weighted width echo state learning system modeling method.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A dynamic prediction method for sintering endpoint based on a time cascade width echo state learning system, characterized in that: The following steps are involved: S1: Considering the combustion mechanism of the sintering process, the process parameters related to BTP were determined using the Spearman rank correlation coefficient: 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical combustion speed; S2: Establishing and training a BTP dynamic prediction model based on a time cascade width echo state learning system; the BTP dynamic prediction model includes a feature node layer and a reservoir; The process of establishing a BTP dynamic model is as follows: S2.1: Collect the production data corresponding to BTP and process parameters affecting BTP to form a training sample data set {(X,Y)|X∈R N×11 ,Y∈R N×1 }, Where X={x1,...,x o ,...,x N } and Y={y1,...,y o ,...,y N }, x o is the oth training sample, y o is the oth BTP, N represents the total number of samples in the training sample data set; in the time cascade width echo state learning system, the initial output set of the feature node is determined as: Where F1 is the output of the first set of feature nodes; ψ() is the first activation function; is the weight vector of the first group; is the deviation vector of the first group; The output F2 of the second set of feature nodes is generated based on the output F1 of the previous set of nodes, expressed as: in, is the weight vector of the second group; is the deviation vector of the second group; The temporal cascading mechanism facilitates the transfer of feature node data from one group to the next; Similarly, define the output of all n groups of feature nodes and determine the output F of the last feature node n It is expressed as: Among them, F n-1 It is the output of the n-1th group of feature nodes; is the weight vector of the nth group; is the deviation vector of the nth group; The interconnection between feature nodes in the feature node layer is as follows: F n =[F1,F2,...,F n ] S2.2: Output F of feature node layer n The initial reservoir is forwarded to the reservoir, and each reservoir passes its current state to the subsequent reservoir. Through the state update between multiple reservoirs, the output matrix of all reservoirs is obtained as follows: H m =[H1,H2,...,H m ] S2.3: Combine the output of the feature node layer with the output of all reservoir layers to obtain: A=[F1,F2,...,F|H1,H2,...,H m ]=[F n |H m ], S2.4: The BTP dynamic prediction model formula obtained based on the time cascade width echo state learning system is: Y=[F n |H m ]W, Among them, Y represents BTP, W represents the weight matrix connecting the hidden layer nodes to the output layer; S3: By using actual production data, taking the process parameters affecting BTP as input and BTP as output, the trained BTP dynamic prediction model is used to perform dynamic prediction of BTP.

2. The method for dynamic prediction of sintering endpoint based on time cascade width echo state learning system according to claim 1, characterized in that: In step S1, based on the carbon combustion mechanism of the sintering process, it can be known that the process parameters affecting BTP mainly involve production quality and carbon combustion reaction, mainly including raw material parameters, state parameters and operation parameters of the sintering process; then using the Spearman rank correlation coefficient, the process parameters affecting BTP are determined by data analysis: 17# bellows temperature, 18# bellows temperature, 19# bellows temperature, 20# bellows temperature, 21# bellows temperature, 22# bellows temperature, BTP temperature, material layer thickness, trolley speed and vertical combustion speed.

3. The method for dynamic prediction of sintering endpoint based on time cascade width echo state learning system according to claim 1, characterized in that: In step S2.2, the state H of the jth reservoir at time t is j (t) The update equation is: Among them, ζ(·) is the second excitation function, and denote the input weight matrix and weight matrix of the j-th reservoir, respectively. represents the connection weight between the j-1th reservoir and the jth reservoir, H j (t-1) represents the state of the jth reservoir at time t-1, and m represents the number of reservoirs.

4. The method for dynamic prediction of sintering endpoint based on time cascade width echo state learning system according to claim 1, characterized in that: In step S2.4, the weight matrix W connecting the hidden layer nodes to the output layer is determined by ridge regression.

5. The method for dynamic prediction of sintering endpoint based on time cascade width echo state learning system according to claim 4, characterized in that: The calculation formula of the weight matrix W connecting the hidden layer nodes to the output layer is: W=(A T A+λI) -1 A T Y, where I represents the identity matrix with appropriate dimension, λ represents a non-negative regularization constant, and Y represents the BTP.

6. The method for dynamic prediction of sintering endpoint based on time cascade width echo state learning system according to claim 5, characterized in that: Use the pseudo-inverse algorithm to solve the weight matrix W. When λ approaches 0, Among them, A + represents the pseudo-inverse matrix of A; By adding A + Substituting into the calculation formula of the weight matrix W, the weight matrix W can be obtained.

7. A storage device, characterized in that: The storage device stores instructions and data for implementing the sintering endpoint dynamic prediction method based on the time cascade width echo state learning system as described in any one of claims 1 to 6.

8. A sintering endpoint dynamic prediction device based on a time cascade width echo state learning system, characterized in that: include: Processor and storage device; the processor loads and executes instructions and data in the storage device to implement the sintering endpoint dynamic prediction method based on the time cascade width echo state learning system as described in any one of claims 1 to 6.

Images