Optimal Target Moisture Prediction Method for Sinter Mix

Abstract

The present invention provides a method for predicting the optimal target moisture of sintering mixture. A large amount of sintering historical data screened by the sintering comprehensive permeability evaluation index is used to perform offline training on the optimized model to learn the complex relationship between the input and output parameters. At the same time, the model is optimized by performing small-scale training on the relevant production data saved every day to overcome the influence brought by the change of raw material composition, realizing the function of accurately predicting the suitable moisture rate of the sintering mixture online. It reduces manual participation, improves the current situation of traditional manual setting based on experience, and improves the automation and intelligent level of sintering. The present invention is based on the WOA-BP neural network. By utilizing the global optimization ability of the whale algorithm, it overcomes the drawback that the traditional BP neural network is prone to fall into local solutions, optimizes the BP neural network basic model with the WOA algorithm, and improves the stability and convergence speed of the network model.

Description

Technical Field

[0001] The present invention belongs to the technical field of sintering production, and particularly relates to a method for predicting the optimal target moisture of sintering mixture. Background Art

[0002] The majority of the raw materials charged into blast furnaces in our country are provided through sintering production. Determining the target moisture content of sintering mixture is a crucial link in the sintering production process. Appropriate moisture in the mixture can achieve the highest pelletizing rate and the maximum permeability of the material layer, and the sintering permeability directly or indirectly affects the yield and quality of sinter. Therefore, it is particularly important to select the optimal target moisture of sintering mixture according to the changes in sintering production operation parameters.

[0003] Since the appropriate moisture rates of different raw materials vary, in actual production, the ore blending structure of sintering mixture often changes. Therefore, when the ore blending structure of sintering mixture changes, the sintering target moisture should be adjusted in a timely manner to make the mixture reach the appropriate granulation moisture to meet the production conditions of subsequent sintering processes.

[0004] So far, the target moisture rate of sintering mixture is mainly set manually based on experience, and the moisture content is controlled by the deviation between the set target value and the actual value. However, the raw material components and ratios under different ore blending structures are different, the wet capacities of different raw materials are different, and the operating habits of each operator are also different. Therefore, it is very difficult to accurately infer the appropriate granulation moisture rate under different ore blending structures only based on manual experience. If there is a large error in the moisture rate of sintering mixture, it will directly affect product quality indicators such as sintering speed, yield, and tumbler strength of sinter.

[0005] In addition, some sintering plants determine the appropriate granulation moisture by conducting tests when the ore blending structure of sintering mixture changes significantly. Sintering cup tests are carried out on sintering mixture under different granulation moisture contents, and the granulation moisture content when the sintering index is optimal is used as the appropriate granulation moisture content of the sintering mixture. However, due to the long test cycle and large test volume of this test method, and the ore blending structure of sintering mixture in sintering production changes at any time, this method cannot accurately determine the appropriate granulation moisture content of sintering mixture with different ore blending structures. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for predicting the optimal target moisture of sintering mixture for online predicting the appropriate moisture rate of sintering mixture.

[0007] The technical solution adopted by the present invention to solve the above technical problem is: a method for predicting the optimal target moisture of sintering mixture, comprising the following steps:

[0008] S1: Obtain the ore blending structure information, the moisture information of the mixture, and the sintering permeability index information of the mixture after sintering at the sintering site;

[0009] S2: Process the data information before sintering according to the sintering situation, perform time series alignment processing on the data obtained in step S1, eliminate abnormal data, construct a sintering permeability evaluation index, and screen out the ore blending structure information and the mixture moisture data sample group with good sintering permeability as the optimal samples;

[0010] S3: Determine the initial structure of the BP neural network, use the WOA whale algorithm to optimize the BP neural network structure, and perform offline training on the optimized BP neural network through the screened data samples to obtain the optimal sintering target moisture prediction model based on the WOA-BP neural network that meets the boundary conditions;

[0011] S4: Online and real-time predict the optimal target moisture under the current ore blending structure through the optimized sintering target moisture prediction model obtained by offline training, and continuously perform small-scale training optimization on the network parameters to adapt to the changes in the production environment.

[0012] According to the above solution, in step S1, the ore blending structure information includes the chemical components of each raw material and the feeding quality measured by the belt scale under the silo; the mixture moisture information includes the measurement value of the microwave moisture meter at the discharge port of the secondary mixer; the sintering permeability index information is the negative pressure value, the air volume, the suction area, and the material layer thickness at each air box.

[0013] According to the above solution, in step S2, the specific steps are as follows:

[0014] S21: Perform time series alignment processing on the data obtained in step S1. Based on the negative pressure measurement point of the last air box in the sintering process, delay the feeding value of each raw material and the mixture moisture measurement value to this time node to ensure that the measurement variables in different processes are the characteristics of the same section of material; eliminate the abnormal data groups including the shutdown stage, the equipment abnormal stage, and the data with too large jitter amplitude to obtain the historical data group under normal production conditions;

[0015] S22: Construct a sintering permeability index based on the negative pressure value of each air box and the sintering machine material layer permeability information to evaluate the sintering situation, screen out the ore blending structure information and the mixture moisture data sample group with good sintering permeability, including the data set of suitable granulation moisture, and use this data set as the optimal sample and perform data segmentation processing.

[0016] Furthermore, in step S22, the specific steps are as follows:

[0017] S221: Let m be the number of air boxes of the sintering machine; a iis the weighting coefficient corresponding to the i-th bellow, and its magnitude is determined according to the degree to which the sintering stage at each bellow position affects the sintering quality, and satisfies Q i is the air volume passing through the material layer at the i-th bellow, F i is the air extraction area at the i-th bellow, h i is the material layer thickness at the i-th bellow, (ΔP) i is the pressure difference between the upper and lower parts of the material layer at the i-th bellow, and n is the gas characteristic constant; the sintering permeability index is calculated according to the following formula:

[0018]

[0019] Select the historical feeding amount and mixture moisture data samples corresponding to the top 40% with good sintering permeability according to the sintering permeability index;

[0020] S222: Divide the filtered data samples into a training set, a validation set, and a test set according to the ratio of 70%:15%:15%.

[0021] According to the above scheme, in step S3, the specific steps are as follows:

[0022] S31: Determine the topological structure of the BP neural network and initialize it, including determining the network input layer parameters, output layer parameters, the number of neurons in the hidden layer, the activation function, and the learning accuracy; normalize the filtered optimal sample data and use it as the training sample of the BP network to obtain the training error of the BP neural network; assign the initial weights and initial thresholds between each neuron;

[0023] S32: Optimize the initial weights and initial thresholds of the BP neural network using the WOA whale algorithm. The specific steps are as follows:

[0024] S321: Initialize the WOA whale algorithm, set the population size of the whales, the initial minimum weight and maximum weight, the maximum number of iterations, and the convergence factor, convert the initial weights and initial thresholds of the BP neural network into the position vectors of the whales, generate random operators, and select the encircling mechanism or the spiral method to update the positions of the whale individuals;

[0025] S322: Use the training error of the BP neural network as the fitness function of the whale algorithm to evaluate the positions of the whale individuals, and repeat step S321 to optimize the positions of the whale individuals until the evaluation requirements are met or the maximum number of iterations is reached to obtain the optimal weights and optimal thresholds;

[0026] S323: Decode the optimal weights and optimal thresholds and output them to the BP neural network to obtain the WOA-BP neural network model with the optimal structure;

[0027] S33: Train the BP neural network model optimized by the WOA whale algorithm using the optimal data samples, continuously perform the forward propagation of the input data and the backpropagation of the output result error until the output result error meets the calculation accuracy or reaches the maximum number of iterations.

[0028] Further, in step S31, the specific steps are as follows:

[0029] S311: Select the ore blending raw material components including homogenized ore, solvent, fuel, dust, dolomite, quicklime, and cold return ore as the input layer x i , i = (1, 2, 3, 4, 5, 6, 7); Set the output layer node as the moisture after mixing;

[0030] S312: Let m be the number of neurons in the hidden layer, n be the number of nodes in the input layer, l be the number of nodes in the output layer, and α be any number between 1 and 10; Calculate the number of neurons in the hidden layer of the BP neural network according to the following formula:

[0031]

[0032] S313: Select the function sigmod() as the transfer function between the input layer and the hidden layer, select the function tansig() as the transfer function between the hidden layer and the output layer, set the number of network learning times as 1000, the learning rate as 0.02, and the minimum training error as 0.00001;

[0033] Let w ij be the connection weight between the i-th node in the input layer and the j-th node in the hidden layer, and b j be the threshold on the j-th node in the hidden layer; Then the output y j of the j-th node in the hidden layer is:

[0034]

[0035] Let w jk be the connection weight between the j-th node in the hidden layer and the k-th node in the output layer, and b k be the threshold on the k-th node in the output layer; Then the output y k of the output layer node is:

[0036]

[0037] Let y' k be the actual value of the data sample; Normalize the optimal sample obtained in step S2 and use it as the training sample of the BP network, and the mean square error function of the network output and the actual sample is:

[0038]

[0039] S314: Select a random number between 0 and 1 as the initial weight and initial threshold of the network neuron connections.

[0040] Further, in the step S321, the specific steps are as follows:

[0041] Convert the initial weight and initial threshold of the BP neural network into the position vector X of the whale * , calculate the fitness of the whale individual, and update the position of the humpback whale; let X (t) be the position vector of the t-th generation, be the optimal position vector of the t-th generation; let r1 and r2 be random numbers belonging to the interval [0, 1]; T max represents the maximum value of the iteration times, and t represents the current iteration times; during the iteration process, a linearly decreases in the interval [2, 0]; A and C are coefficient vectors; let the initial iteration times of the population be 1, and generate a random number p between [0, 1]; if the distance coefficient |A| < 1 and p < 0.5, then the whale updates its position by the way of shrinking and surrounding to perform surrounding predation. Let D be the distance between the optimal position vector of the t-th generation and the position vector of the t-th generation when the whale adopts the shrinking type of surrounding, as shown in formulas (1) to (5);

[0042] A = 2a·r1 - a (1),

[0043] C = 2r2 (2),

[0044] a = 2 - 2t / T max (3),

[0045]

[0046]

[0047] X (t+1) is the position vector of the (t + 1)-th generation, b is a constant; l is a random number between 0 and 1; if the distance coefficient |A| < 1 and p ≥ 0.5, then the whale updates its position by the spiral way to perform surrounding predation. Let D' be the distance between the optimal position vector of the t-th generation and the position vector of the t-th generation when the whale adopts spiral rising, as shown in formulas (6), (7);

[0048]

[0049] X (t+1) = X * (t) + D'e bl cos(2πl) (7);

[0050] Let X randIt is the position vector of any individual in the current generation; if the distance coefficient |A| ≥ 1, it is the stage of exploring food, and the whale updates its own position by continuous search, as shown in formulas (8) and (9);

[0051] D = |CX rand -X (t) |(8),

[0052] X (t+1) = X rand -A·D (9).

[0053] According to the above scheme, in step S4, the specific steps for small-scale training and optimization of network parameters are as follows:

[0054] Process the network parameters according to steps S2 to S3 and retain the relevant data of daily production operations.

[0055] A computer storage medium stores a computer program executable by a computer processor, and the computer program executes an optimal target moisture prediction method for sintering mixture.

[0056] The beneficial effects of the present invention are as follows:

[0057] 1. For the optimal target moisture prediction method of sintering mixture of the present invention, the optimized model is offline trained with a large amount of sintering historical data screened by the sintering comprehensive air permeability evaluation index to learn the complex relationship between the input quantity and output quantity parameters. At the same time, the model is small-scale trained and optimized by saving the relevant production data of each day to overcome the influence brought by the change of raw material composition, realizing the function of accurately predicting the suitable moisture rate of sintering mixture online. It reduces manual participation and improves the sintering automation level.

[0058] 2. Based on the WOA-BP neural network, the present invention utilizes the global optimization ability of the whale algorithm to overcome the disadvantage that the traditional BP neural network is prone to falling into local solutions, optimizes the BP neural network basic model with the WOA algorithm, and improves the stability and convergence speed of the network model.

[0059] 3. By accurately predicting the optimal target moisture of sintering mixture under the current ore blending structure online, the present invention improves the current situation of traditional manual setting based on experience and enhances the intelligent level of sintering production. Description of the Drawings

[0060] Figure 1 It is the flowchart of the embodiment of the present invention.

[0061] Figure 2 It is the schematic diagram of sintering production of the embodiment of the present invention.

[0062] Figure 3 It is the flowchart for processing the historical data of sintering production in an embodiment of the present invention.

[0063] Figure 4 It is the schematic diagram of the original data of the steel plant in an embodiment of the present invention.

[0064] Figure 5 It is the flowchart for optimizing the WOA - BP neural network model in an embodiment of the present invention.

[0065] Figure 6 It is the schematic diagram of the structure of the BP neural network in an embodiment of the present invention.

[0066] Figure 7 It is the fitness curve diagram for predicting the optimal target moisture of the sintering mixture based on the WOA - BP neural network in an embodiment of the present invention.

[0067] Figure 8 It is the comparison diagram between the predicted output and the expected output of the optimal target moisture of the sintering mixture by the BP neural network and the WOA - BP neural network in an embodiment of the present invention.

[0068] Figure 9 It is the error comparison diagram of the optimal target moisture of the sintering mixture by the BP neural network and the WOA - BP neural network in an embodiment of the present invention. Specific embodiments

[0069] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0070] Refer to Figure 1 , a method for predicting the optimal target moisture of a sintering mixture in an embodiment of the present invention includes the following steps:

[0071] S1: Obtain the ore blending structure information, the moisture information of the mixture, and the sintering air permeability index information of the mixture before sintering at the sintering site through various measuring instruments;

[0072] The ore blending structure information of sintering includes the chemical components of each raw material and the feeding quality measured by the belt scale under the bunker;

[0073] The moisture of the mixture is the measured value of the microwave moisture meter at the discharge port of the secondary mixing machine;

[0074] The sintering air permeability information is the negative pressure value, the air volume, the suction area, and the bed thickness at each air box.

[0075] S2: Process the data information before sintering according to the quality of sintering. The data processing method is as Figure 3 shown, and specifically includes:

[0076] S21: Since the production time of a section of raw material in each process of the entire production line is different, the recorded data needs to be time-aligned. The time-alignment method is: taking the last bellows negative pressure measurement point of the sintering process as the reference, delay the discharge value of each raw material and the moisture measurement value of the mixture to this time node, so as to ensure that the measured variables in different processes are the characteristics of the same section of material.

[0077] Considering the complex working conditions on site, such as Figure 4 As shown, the data described in S1 is subjected to time series alignment processing; abnormal data groups including shutdown phase, equipment abnormal phase, and data jitter amplitude is too large are eliminated to obtain the historical data group under normal production conditions;

[0078] S22: Construct a sintering permeability evaluation index based on the negative pressure value of each bellows and the permeability information of the sintering machine material layer, further refine the data before sintering by the good or bad sintering conditions, and use this index as the evaluation basis to screen out the ore distribution structure and mixed material moisture data sample group when the sintering permeability is good, including a data set with moisture suitable for granulation. This data set is used as the optimal sample and data segmentation processing is performed.

[0079] S221: The method for screening data based on sintering permeability index is:

[0080] The comprehensive sintering permeability index is calculated according to the following formula.

[0081]

[0082] In the formula, m is the number of wind boxes of the sintering machine, a i is the weighted coefficient corresponding to the i-th bellows, and its size is determined according to the degree to which the sintering quality is affected by the sintering stage at each bellows position, and satisfies Q i is the air volume passing through the material layer at the i-th wind box, F i is the exhaust area at the i-th bellows, h i is the material layer thickness at the i-th wind box, (ΔP) i is the pressure difference between the upper and lower layers of the material at the i-th bellows, and n is the gas characteristic constant.

[0083] Based on this index, the top 40% of historical feed quantity and mixture moisture data samples with good sintering permeability were screened out.

[0084] S222: The data segmentation method is: the optimal data samples after screening are divided into a training set, a validation set, and a test set in a ratio of 70%:15%:15%.

[0085] S3: Figure 5As shown in the figure, an optimal sintering target moisture prediction model based on the WOA-BP neural network is constructed. The initial structure of the BP neural network is determined, and the WOA whale algorithm is used to optimize the BP neural network structure. The BP network is offline trained with the selected data samples to obtain the WOA-BP model that meets the boundary conditions (meeting the error accuracy or reaching the maximum number of iterations).

[0086] S301: Adopt the BP neural network structure as shown in Figure 6 the figure and initialize it.

[0087] S3011: Select the ore blending raw material components as the network input layer. In this example, the specific raw material components are 7 types, namely blended ore, solvent, fuel, dust, dolomite, quicklime, and cold return ore, that is, the input layer x i , i = (1, 2, 3, 4, 5, 6, 7); the node of the network output layer is the moisture after mixing.

[0088] S3012: Determine the number of neurons in the hidden layer, and calculate the number of neurons in the hidden layer of the BP neural network according to the following empirical formula.

[0089]

[0090] In the formula: m is the number of neurons in the hidden layer, n is the number of nodes in the input layer, l is the number of nodes in the output layer, and α is any number between 1 and 10. The number of neurons in the hidden layer calculated in this example is 10.

[0091] S3013: Select the sigmod() function as the transfer function between the input layer and the hidden layer, select the tansig() function as the transfer function between the hidden layer and the output layer, set the network learning times to 1000, the learning rate to 0.02, and the training minimum error to 0.00001.

[0092] The output of the j-th node in the hidden layer is:[[]]

[0093]

[0094] Among them, w ij is the connection weight between the i-th node in the input layer and the j-th node in the hidden layer, and b j is the threshold on the j-th node in the hidden layer.

[0095] The output of the output layer node is:[[]] In the formula, w jk is the connection weight between the j-th node in the hidden layer and the k-th node in the output layer, and b k is the threshold on the k-th node in the output layer. In this example, there is only one output node, that is, k = 1.

[0096] Normalize the optimal samples in S2 and use them as the training samples of the BP network, then the mean square error function between the network output and the actual samples can be obtained: where y k is the output value of the network, and y' k is the actual value of the data sample.

[0097] S3014: Select a random number between 0 and 1 as the initial weights and thresholds of the network neuron connections.

[0098] S302: Initialize the WOA whale algorithm, set the whale population size, initial minimum weight and maximum weight, maximum number of iterations, and convergence factor.

[0099] S303: Use the WOA whale algorithm to optimize the initial weights and thresholds of the BP neural network. The specific calculation and optimization process includes:

[0100] Convert the initial weights and initial thresholds of the neural network into the position vector X of the whale * ;

[0101] Let the initial iteration number of the population be 1, and generate a random number p between [0, 1];

[0102] If the distance coefficient |A| < 1 and p < 0.5, the whale updates its position by the shrinking encircling method to perform encircling predation, as shown in formulas (1) to (5);

[0103] A = 2a·r1 - a (1)

[0104] C = 2r2 (2)

[0105] a = 2 - 2t / T max (3)

[0106] where r1 and r2 are random numbers in the interval [0, 1]; T max represents the maximum value of the iteration number, and t represents the current iteration number; during the iteration process, a linearly decreases in the interval [2, 0]; A and C are coefficient vectors;

[0107]

[0108]

[0109] In the formula, is the optimal position vector of the t-th generation; X (t) is the position vector of the t-th generation.

[0110] If the distance coefficient |A| < 1 and p ≥ 0.5, the whale updates its position by the spiral method to perform encircling predation, as shown in formulas (6) and (7);

[0111]

[0112]

[0113] where: D' is the distance between the optimal position vector of the t-th generation and the position vector of the t-th generation, X (t+1) is the position vector of the (t + 1)-th generation, b is a constant; l is a random number between 0 and 1.

[0114] If the distance coefficient |A| ≥ 1, it is the food exploration stage, and the whale updates its own position by continuous search, as shown in formulas (8) and (9);

[0115] D = |CX rand - X (t) | (8)

[0116] X (t+1) = X rand - A·D (9)

[0117] where: X rand is the position vector of any individual in the current generation.

[0118] Taking the training error of the BP neural network as the fitness function of the whale algorithm to evaluate the individual position, repeating the above content, continuously optimizing the whale individual position until the performance index is met or the maximum number of iterations is reached, to obtain the optimal weights and thresholds, and the fitness curve is as Figure 7 shown. Finally, decode the optimal weights and thresholds into the BP neural network to obtain the structurally optimal WOA - BP neural network model.

[0119] S4: Use the offline trained WOA - BP model to perform online real - time prediction on the optimal target moisture content under the current ore blending structure, and at the same time continuously perform small - scale training optimization on the network parameters to adapt to the changes in the production environment.

[0120] The method for small - scale training optimization of network parameters is: retain the relevant data of daily production operations and process them according to the methods of steps S2 to S3.

[0121] In this example, data processing (delaying for alignment, removing bad values, data evaluation, data segmentation) is performed on 5000 groups of historical data of a steel plant, and 2350 groups of data considered reliable and effective are selected. Among the 2350 - group dataset, 1975 groups of data are selected as the training set and validation set of the network, and another 375 groups of data are selected as the test set to analyze and verify the prediction effects of WOA - BP and BP neural networks. Through Figure 8 、 Figure 9The experimental results show that the proposed prediction model provides a new method for determining the target moisture content of sinter mix and meets the requirements of on-site processes.

[0122] The above embodiments are only used to illustrate the design concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made according to the principles and design concepts disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A method for predicting the optimal target moisture content of sintering mixture, characterized in that: The following steps are involved: S1: Obtaining the ore structure information of the mixture before sintering, the moisture information of the mixture, and the sintering permeability index information of the mixture after sintering at the sintering site; S2: Processing the data information before sintering according to the sintering conditions, performing time series alignment on the data obtained in step S1 and eliminating abnormal data, constructing a sintering permeability evaluation index, and selecting the ore structure information and the mixed material moisture data sample group when the sintering permeability is good as the optimal sample; S3: Determine the initial structure of the BP neural network, use the WOA whale algorithm to optimize the BP neural network structure, and perform offline training on the optimized BP neural network through the screened data samples to obtain the optimal sintering target moisture prediction model based on the WOA-BP neural network that meets the boundary conditions; the specific steps are: S31: Determine the topological structure of the BP neural network and initialize it, including determining the network input layer parameters, output layer parameters, number of hidden layer neurons, activation function and learning accuracy; The optimal sample data after screening is normalized and used as the training sample of BP network to obtain the training error of BP neural network; the initial weights and initial thresholds between each neuron are assigned; S32: Using the WOA whale algorithm to optimize the initial weights and initial thresholds of the BP neural network, the specific steps are: S321: Initialize the WOA whale algorithm, set the whale population size, initial minimum weight and maximum weight, maximum number of iterations and convergence factor, convert the initial weight and initial threshold of the BP neural network into the whale position vector, generate a random operator, and select the encirclement mechanism or spiral method to update the individual position of the whale; S322: Using the training error of the BP neural network as the fitness function of the whale algorithm to evaluate the individual position of the whale, repeating step S321 to optimize the individual position of the whale until the evaluation requirements are met or the maximum number of iterations is reached to obtain the optimal weight and optimal threshold; S323: Decoding and outputting the optimal weight and the optimal threshold to the BP neural network to obtain a WOA-BP neural network model with the optimal structure; S33: Use the optimal data sample to train the BP neural network model optimized by the WOA whale algorithm, and continuously perform forward propagation of input data and reverse propagation of output result error until the output result error meets the calculation accuracy or reaches the maximum number of iterations; S4: The optimal sintering target moisture prediction model trained offline is used to predict the optimal target moisture under the current ore blending structure in real time online. At the same time, the network parameters are continuously trained and optimized on a small scale to adapt to changes in the production environment.

2. The optimal target moisture prediction method for sintering mixture according to claim 1, wherein: In the step S1, The ore blending structure information includes the chemical composition of each raw material and the material quality measured by the belt scale under the silo; The moisture information of the mixture includes the measurement value of the microwave moisture meter at the outlet of the secondary mixer; The sintering permeability index information includes the negative pressure value, air volume, exhaust area and material layer thickness at each bellows.

3. The optimal target moisture prediction method for sintering mixture according to claim 1, characterized in that: In the step S2, the specific steps are: S21: Perform time alignment processing on the data obtained in step S1, take the last wind box negative pressure measurement point in the sintering process as the reference, delay the feeding value of each raw material and the moisture measurement value of the mixed material to the time node, ensure that the measured variables in different processes are the characteristics of the same section of materials; eliminate the abnormal data groups including the shutdown stage, equipment abnormal stage, and data jitter with excessive amplitude, and obtain the historical data group under normal production conditions; S22: Construct a sintering permeability index to evaluate the sintering condition based on the negative pressure value of each bellows and the permeability information of the sintering machine material layer, and screen out the ore structure information and mixed material moisture data sample group when the sintering permeability is good, including a data set with moisture suitable for granulation. Use this data set as the optimal sample and perform data segmentation processing.

4. The optimal target moisture prediction method for sintering mixture according to claim 3, characterized in that: In the step S22, the specific steps are: S221: Set m as the number of wind boxes of the sintering machine; is the weighting coefficient corresponding to the i-th wind box, and its magnitude is determined according to the degree of influence of the sintering stage at each wind box position on the sintering quality, and satisfies ; is the magnitude of the air volume passing through the material layer at the i-th wind box, is the air extraction area at the i-th wind box, is the thickness of the material layer at the i-th wind box, is the pressure difference between the upper and lower parts of the material layer at the i-th wind box, n is the gas characteristic constant; calculate the sintering permeability index according to the following formula: ； According to the sintering permeability index, select the top 40% of historical material feeding amount and mixture moisture data samples with good sintering permeability; S222: Split the filtered data samples into training set, validation set and test set in a ratio of 70%:15%:15%.

5. The optimal target moisture prediction method for sintering mixture according to claim 1, characterized in that: In the step S31, the specific steps are: S311: Select the ore blending raw material components including homogenized ore, solvent, fuel, dust, dolomite, quicklime, and cold return ore as the input layer x i , i = (1, 2, 3, 4, 5, 6, 7); Set the output layer node as the moisture after mixing; S312: Set m as the number of neurons in the hidden layer, n as the number of nodes in the input layer, l as the number of nodes in the output layer, and α is any number between 1 and 10; calculate the number of neurons in the hidden layer of the BP neural network according to the following formula: ； S313: Select the function sigmod() as the transfer function between the input layer and the hidden layer, select the function tansig() as the transfer function between the hidden layer and the output layer, set the network learning times to 1000, the learning rate to 0.02, and the lowest training error to 0.00001; Let be the connection weight between the i -th node of the input layer and the j -th node of the hidden layer. is the threshold value on the j -th node of the hidden layer. Then the output y j of the -th node of the hidden layer is: ； Let be the connection weight between the j -th node of the hidden layer and the k -th node of the output layer, and be the threshold on the k -th node of the output layer; then the output of the output layer node is: ； Let be the actual value of the data sample; normalize the optimal sample obtained in step S2 and use it as the training sample of the BP network, and the mean square error function between the network output and the actual sample is obtained as follows: ； S314: Select a random number between 0 and 1 as the initial weight and initial threshold of the network neuron connection.

6. The optimal target moisture prediction method for sintering mixture according to claim 5, characterized in that: In the step S321, the specific steps are: Convert the initial weights and initial thresholds of the BP neural network into the position vectors of whales , calculate the fitness of individual whales, and update the positions of humpback whales; let be the position vector of the t -th generation, be the optimal position vector of the t -th generation; let and be random numbers belonging to the interval [0, 1]; represents the maximum value of the number of iterations, t represents the current number of iterations; During the iterative process, a decreases linearly in the interval [2, 0]; A and C is the coefficient vector; Let the initial iteration number of the population be 1, and generate a random number between [0, 1] p ; If the distance coefficient , and , then the whale updates its position by shrinking and surrounding to perform encircling predation. Let D be the distance between the optimal position vector of the t -th generation and the position vector of the t -th generation when the whale adopts the shrinking and surrounding method, as shown in formulas (1) to (5); （1）， （2）， （3）， （4）， （5）； is the position vector of the t +1 generation, and b is a constant; l is a random number between 0 and 1; if the distance coefficient , and , then the whale updates its position in a spiral manner to perform encircling predation. Let be the distance between the optimal position vector of the t generation and the position vector of the t generation when the whale ascends in a spiral, as shown in formulas (6) and (7); （6）， （7）； Let be the position vector of any individual in the current generation; if the distance coefficient , then it is the food exploration stage, and the whale updates its own position by continuous search, as shown in formulas (8) and (9); （8）， （9）。 7. A method for predicting the optimal target moisture content of sintering mixture according to claim 1, characterized in that: In step S4, the specific steps of performing small-scale training optimization on network parameters are as follows: According to step S2 to step S3, the network parameters are processed and the relevant data of daily production operation is retained.

8. A computer storage medium, characterized in that: A computer program executable by a computer processor is stored therein, and the computer program executes a method for predicting the optimal target moisture content of a sintering mixture as described in any one of claims 1 to 7.

Images