Blast furnace pipeline pressure analysis and early warning system and method based on improved neural network

Abstract

The invention relates to the field of blast furnace pipeline pressure analysis, and specifically to a blast furnace pipeline pressure analysis early warning system and method based on an improved neural network, comprising: a pipeline sensing module, a furnace chamber pressure module, a probability prediction module, an improved network module and an early warning adjustment module, wherein the pipeline sensing module is used to read and adjust the internal pressure of the pipeline, the furnace chamber pressure module is used to record blast furnace working parameters and generate a basic pressure function, the probability prediction module is used to train a neural network and output a pipeline pressure prediction model, the improved network module is used to correct and calibrate the model, and the early warning adjustment module is used to calculate the pipeline pressure, adjust a pressure relief threshold and send an early warning signal. The invention can improve the adaptability of the neural network, can better predict the pipeline pressure, help ensure the safety and stability of pipeline operation, improve the production efficiency of the blast furnace, reduce the risk of safety accidents, and ensure the safe and stable operation of the blast furnace.

Description

Technical Field

[0001] The invention relates to the field of blast furnace pipeline pressure analysis, and in particular to a blast furnace pipeline pressure analysis early warning system and method based on an improved neural network. Background Art

[0002] Blast furnace pipelines usually refer to the pipeline system used to transport raw materials and fuels in the ironmaking process, including feed pipelines for transporting raw materials such as coke and iron ore, and pressure pipelines for transporting fuel and air. The air pressure in the pressure pipeline is affected by many factors such as the feeding state, the working state of the blast furnace, the temperature and humidity of the furnace chamber, and the pipeline design. In the blast furnace production process, the pipeline pressure is actually in a constant swing.

[0003] Since blast furnace pipelines transport high-temperature and high-pressure gases and liquids, pressure exceeding the pipeline limit can easily cause safety risks. Although most pipelines have redundant designs, long-term high-pressure shocks can also cause pipeline fatigue. Therefore, real-time detection and early warning of pipeline pressure are required, and pressure prediction and pressure reduction preparations need to be made in advance.

[0004] Most of the current pipeline prediction algorithms are fitted through neural network models, but ordinary neural network models learn blast furnace data slowly and are more prone to deviations such as overfitting and underfitting, and cannot reflect the pipeline pressure changes well. In addition, the cost function of the existing neural network is usually an empirical function obtained from other blast furnaces, lacks adaptive adjustment to the actual state of the blast furnace, and the model accuracy is poor. Summary of the invention

[0005] The object of the present invention is to provide a blast furnace pipeline pressure analysis and early warning system and method based on an improved neural network to solve the problems raised in the above-mentioned background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a blast furnace pipeline pressure analysis and early warning system based on an improved neural network, comprising: a pipeline sensing module, a furnace chamber pressure module, a probability prediction module, an improved network module and an early warning adjustment module;

[0007] The pipeline sensing module is composed of a gravity pressure relief valve arranged in the blast furnace pipeline and a sensor inside the valve. The pressure relief valve is used to adjust the internal pressure of the pipeline. The sensor is used to detect the height of the counterweight block in the gravity pressure relief valve, thereby reading the internal pressure of the pipeline and being able to change the pressure relief threshold of the pressure relief valve according to instructions.

[0008] The furnace chamber pressure module is used to record the blast furnace operating parameters according to the sensor in the feed pipeline, sample the pipeline pressure and the operating parameters at the same time, store the sampling results in the database, use the sampled data in the database to perform exponential trend detection, generate a basic pressure function, and generate a probability distribution function of the pipeline pressure under the current operating parameters according to the measured pipeline pressure data;

[0009] The probability prediction module is used to define a regression matrix, use the regression matrix as a transformation matrix, the current base pressure as a mean value, and the actual pipeline pressure as input data, construct a cross entropy function, calculate the minimum value of the cross entropy function, use the regression matrix corresponding to the minimum value as a cost matrix, use a convolutional neural network to learn the sampled data in the database, and output a pipeline pressure prediction model after training;

[0010] The improved network module is used to generate a model loss function with the deviation of the model in the measured data as the dependent variable and the data sequence number as the independent variable, calculate the partial derivatives of the deviation of the pipeline pressure prediction model in each dimension, determine the size of the regularization factor by the partial derivative sequence, perform regularization processing on the model loss function, eliminate the overfitting items in the model, and output the corrected pipeline pressure fitting model;

[0011] The early warning adjustment module is used to input the blast furnace working data into the pipeline pressure fitting model at fixed intervals to obtain the probability distribution function of the pressure in the pipeline, multiply the interval length of the probability distribution exceeding the preset confidence interval by the function value, and obtain the impact cumulative amount. When the impact cumulative amount exceeds a first threshold, the pressure relief threshold of the pressure relief valve is adjusted, and when it exceeds a second threshold, a pressure early warning signal is issued.

[0012] Further, the pipeline sensing module includes: a pressure relief valve unit and an internal pressure recording unit;

[0013] The pressure relief valve unit is used to release the internal pressure of the pipeline to the outside when the internal pressure of the pipeline is greater than the pressure relief threshold pressure, until the internal pressure of the pipeline is lower than the threshold pressure;

[0014] The internal pressure recording unit is used to read the internal pressure of the pipeline according to the height of the counterweight block on the pressure relief valve.

[0015] Further, the furnace chamber pressure module includes: a parameter sampling unit, a database unit and a basic pressure unit;

[0016] The parameter sampling unit is used to sample the working parameters of the blast furnace and the internal pressure of the pipeline at the same time, wherein the working parameters include the blast furnace charging speed, the temperature and humidity of the combustion chamber, and the wind speed of the blast chamber;

[0017] The database unit is used to store sampled data and construct the sampled data into a data set suitable for neural network training;

[0018] The basic pressure unit is used to perform exponential trend detection on the sampled data and output a basic pressure function, which reflects the relationship between each working parameter and the average value of the pipeline pressure.

[0019] Further, the probability prediction module includes: a cross regression unit and a cost input unit;

[0020] The cross regression unit is used to set an unknown regression matrix, fit the cross entropy function with the base pressure as the mean, and use the regression matrix output corresponding to the minimum value of the cross entropy function as the cost matrix;

[0021] The cost input unit is used to set the cost matrix as the objective function matrix of the convolutional neural network to learn the sampled data in the database and output the pipeline pressure prediction model.

[0022] Furthermore, the improved network module includes: a model loss unit, a regularization repair unit and a fitting output unit;

[0023] The model loss unit is used to calculate the loss function of the pipeline pressure prediction model and the measured data, and calculate the partial derivative of the loss function offset;

[0024] The regularization repair unit is used to multiply the partial derivative under the current base pressure by a preset coefficient to obtain the size of the regularization factor;

[0025] The fitting output unit is used to regularize the relevant items in the pipeline pressure prediction model and output the regularized pressure prediction model.

[0026] Further, the warning adjustment module includes: a pressure relief threshold unit and a cumulative warning unit;

[0027] The pressure relief threshold unit is used to predict the probability distribution function of the pipeline pressure in the future period by using the pressure prediction model at preset time intervals, and adjust the pressure relief threshold when the average expectation of the probability distribution function is higher than the threshold;

[0028] The cumulative warning unit is used to calculate the cumulative impact amount of the pipeline and issue a warning when the cumulative impact amount exceeds a threshold.

[0029] The blast furnace pipeline pressure analysis and early warning method based on improved neural network includes the following steps:

[0030] Step S1. A pressure relief valve is provided on the outer wall of the blast furnace pressure pipeline, and a sensor is provided in the feed pipeline. The pressure indication of the pressure relief valve and the readings of each sensor are read according to different sampling periods, and the read pipeline pressure and blast furnace operating parameters are stored in a database;

[0031] Step S2. Taking the mean value of the pipeline pressure within the sampling period as the output point and the blast furnace operating parameters as the trend variables, perform an exponential trend test on the mean value of the pipeline pressure. Substitute the obtained test parameters into the preset function model to output the basic pressure function of the pipeline pressure mean value and the blast furnace operating parameters;

[0032] Step S3. Perform a regression fitting on the pipeline pressure data. Replace the cost function matrix of the neural network with the regression matrix corresponding to the minimum cross-entropy function of the sampling data to obtain an improved neural network;

[0033] Step S4. The improved neural network learns the data in the database and outputs the first model. Calculate the loss amount of the first model under each operating parameter, multiply the maximum loss amount by the preset coefficient to obtain the regularization parameter, and perform regularization processing on the loss function of the first model with the regularization parameter to eliminate the overfitting term in the model and obtain the second model;

[0034] Step S5. Read the operating parameters of the current blast furnace, obtain the pipeline basic pressure from the basic pressure function, input the pipeline basic pressure into the second model, and output the probability distribution function and the expected function of the pipeline pressure. Issue an alarm when the probability that the pipeline pressure expected value is higher than the threshold exceeds the preset confidence interval.

[0035] Further, Step S1 includes:

[0036] Step S11. Install pressure relief valves in the blast furnace pressure pipeline. The pressure relief valves include: a gravity pressure relief valve and an electronic pressure relief valve. Install sensors in the feeding pipeline. The feeding pipeline includes: a fuel pipeline, a furnace chamber pipeline, and a blast chamber pipeline;

[0037] Step S12. Preset the pressure sampling period T1 and the working sampling period T2, where T1 < T2 and T2 is an integer multiple of T1. Read the readings of the pressure relief valves every T1 time duration to obtain the pipeline pressure data, and read the readings of each sensor every T2 time duration to obtain the operating parameters of the blast furnace. The operating parameters include: fuel feeding speed, combustion chamber temperature, and blast chamber air velocity;

[0038] Step S13. Use the working data obtained from a single sampling as the table header of the data table, and store the pipeline pressure data collected within the sampling period of the working data as the data within the table into the database to form a sampling data table.

[0039] Further, Step S2 includes:

[0040] Step S21. For the sampling data table, take the mean value of the data within the table as the output point and the header data as the trend variables for an exponential trend test. The exponential trend test is as follows:

[0041] List the following trend test equation:

[0042] ,

[0043] Wherein, D1, D2, ...Dm represent the means of the data in the 1st, 2nd, ...mth sampling data tables, respectively; a1, a2, ...an represent the inspection parameters corresponding to the 1st, 2nd, ...nth working parameters of the blast furnace, respectively; m is the number of sampling data tables, n is the number of working parameters, and m ≥ n; x(m, n) represents the nth working parameter recorded in the mth sampling data table;

[0044] Solve the trend test equation to obtain the values ​​of a1, a2, ...an;

[0045] Step S22. According to the preset function model, output the basic pressure function F(x1,x2,…,xn) of the pipeline pressure mean and the blast furnace working parameters, where F(x1,x2,…,xn)=a1 x1 +a2 x2 +…+an xn , x1, x2, …xn represent the 1st, 2nd, …nth working parameters respectively.

[0046] Further, step S3 includes:

[0047] Step S31. Calculate the cross entropy loss function of the pipeline pressure data relative to the blast furnace base pressure according to the loss function construction method based on the pipeline pressure data in the sampled data table:

[0048] ,

[0049] Among them, J(W,D) represents the cross entropy loss function, G(i) represents the i-th pipeline pressure data value in the sampled data table, lg is the logarithm with base 10, W is the cost function matrix, D is the mean of the data in the sampled data table, and e is the base of the natural logarithm;

[0050] Step S32. Calculate the cost function matrix W corresponding to the minimum value of the cross entropy loss function, replace the existing cost function matrix in the neural network model with W, and generate a new neural network model, which is recorded as an improved neural network.

[0051] Further, step S4 includes:

[0052] Step S41. Using the improved neural network to learn the pipeline pressure data in the database, fitting the probability distribution model of the pipeline pressure value relative to the mean value as the first model output;

[0053] Step S42. Input the working parameters in the sampling data table into the basic pressure function to obtain the basic pressure in the sampling data table, input the basic pressure as the mean into the first model, output the probability distribution function P1(x) of the pipeline pressure, and calculate the loss of the probability distribution function:

[0054]

[0055] Where E is the loss amount of the probability distribution function;

[0056] Step S43. Execute step S42 for all sampled data tables in the database to obtain the minimum value E0 of E, multiply it by the preset magnification coefficient q to obtain the regularization parameter r, and regularize the first model according to the regularization parameter to eliminate the overfitting items in the model:

[0057]

[0058] Among them, J(θ) represents the coefficient function of the first model, θ i Represents the current coefficient of the i-th fitting term in the first model;

[0059] Step S44: Correct the coefficients of each fitting item in the first model according to the coefficient function to obtain the second model.

[0060] Further, step S5 includes:

[0061] Step S51. Obtain the working parameters of the blast furnace charging pipeline, obtain the basic pressure of the pressure pipeline according to the basic pressure function, input the basic pressure into the second model, obtain the probability distribution function of the pipeline pressure, multiply it by the basic pressure, and obtain the expected function of the pipeline pressure;

[0062] Step S52. Preset a pipeline pressure threshold, calculate the probability that the expected function value of the pipeline pressure is above the pressure threshold, and when the obtained probability exceeds the preset confidence interval, send an early warning signal of abnormal blast furnace pipeline pressure to the management center.

[0063] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0064] The present invention sets a gravity pressure relief valve in the blast furnace pipeline to record the internal pressure of the pipeline, samples the pipeline pressure and blast furnace operating parameters at the same time, performs exponential trend detection on the sampled data, and generates a basic pressure function according to the detection results, thereby reflecting the correlation between the blast furnace operating parameters and the pipeline pressure, which can help to reasonably control the blast furnace pressure, improve the blast furnace production efficiency and raw material utilization rate, and avoid excessive pressure regulation.

[0065] The present invention can define a regression matrix, use the regression matrix as a transformation matrix to construct a cross entropy prediction function, take the regression matrix corresponding to the minimum value of the function as a cost function matrix, train a convolutional neural network, and improve the adaptability of the neural network by using the historical pressure changes of the blast furnace, so as to better predict the pipeline pressure and improve the operation efficiency and safety of the pipeline.

[0066] The present invention can regularize the model loss function, eliminate the overfitting items in the model, adjust the pressure relief threshold of the pressure relief valve according to the probability distribution and the preset confidence interval, and accumulate the pipeline impact time. When the impact accumulates to the threshold, a warning signal is fed back, which helps to ensure the safety and stability of the pipeline operation, improve the blast furnace production efficiency, reduce the risk of safety accidents, and ensure the safe and stable operation of the blast furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0068] Figure 1 It is a structural schematic diagram of a blast furnace pipeline pressure analysis and early warning system based on an improved neural network of the present invention;

[0069] Figure 2 It is a schematic diagram of the steps of the blast furnace pipeline pressure analysis and early warning method based on the improved neural network of the present invention. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] See also Figure 1 , the present invention provides a technical solution: a blast furnace pipeline pressure analysis and early warning system based on an improved neural network, comprising: a pipeline sensing module, a furnace chamber pressure module, a probability prediction module, an improved network module and an early warning adjustment module;

[0072] The pipeline sensing module is composed of a gravity pressure relief valve arranged in the blast furnace pipeline and a sensor inside the valve. The pressure relief valve is used to adjust the internal pressure of the pipeline. The sensor is used to detect the height of the counterweight block in the gravity pressure relief valve, thereby reading the internal pressure of the pipeline and being able to change the pressure relief threshold of the pressure relief valve according to instructions.

[0073] The pipeline sensing module includes: a pressure relief valve unit and an internal pressure recording unit;

[0074] The pressure relief valve unit is used to release the internal pressure of the pipeline to the outside when the internal pressure of the pipeline is greater than the pressure relief threshold pressure, until the internal pressure of the pipeline is lower than the threshold pressure;

[0075] The internal pressure recording unit is used to read the internal pressure of the pipeline according to the height of the counterweight block on the pressure relief valve.

[0076] The furnace chamber pressure module is used to record the blast furnace operating parameters according to the sensor in the feed pipeline, sample the pipeline pressure and the operating parameters at the same time, store the sampling results in the database, use the sampled data in the database to perform exponential trend detection, generate a basic pressure function, and generate a probability distribution function of the pipeline pressure under the current operating parameters according to the measured pipeline pressure data;

[0077] The furnace chamber pressure module includes: a parameter sampling unit, a database unit and a basic pressure unit;

[0078] The parameter sampling unit is used to sample the working parameters of the blast furnace and the internal pressure of the pipeline at the same time, wherein the working parameters include the blast furnace charging speed, the temperature and humidity of the combustion chamber, and the wind speed of the blast chamber;

[0079] The database unit is used to store sampled data and construct the sampled data into a data set suitable for neural network training;

[0080] The basic pressure unit is used to perform exponential trend detection on the sampled data and output a basic pressure function, which reflects the relationship between each working parameter and the average value of the pipeline pressure.

[0081] The probability prediction module is used to define a regression matrix, use the regression matrix as a transformation matrix, the current base pressure as a mean value, and the actual pipeline pressure as input data, construct a cross entropy function, calculate the minimum value of the cross entropy function, use the regression matrix corresponding to the minimum value as a cost matrix, use a convolutional neural network to learn the sampled data in the database, and output a pipeline pressure prediction model after training;

[0082] The probability prediction module includes: a cross regression unit and a cost input unit;

[0083] The cross regression unit is used to set an unknown regression matrix, fit the cross entropy function with the base pressure as the mean, and use the regression matrix output corresponding to the minimum value of the cross entropy function as the cost matrix;

[0084] The cost input unit is used to set the cost matrix as the objective function matrix of the convolutional neural network to learn the sampled data in the database and output the pipeline pressure prediction model.

[0085] The improved network module is used to generate a model loss function with the deviation of the model in the measured data as the dependent variable and the data sequence number as the independent variable, calculate the partial derivatives of the deviation of the pipeline pressure prediction model in each dimension, determine the size of the regularization factor by the partial derivative sequence, perform regularization processing on the model loss function, eliminate the overfitting items in the model, and output the corrected pipeline pressure fitting model;

[0086] The improved network module includes: a model loss unit, a regularization repair unit and a fitting output unit;

[0087] The model loss unit is used to calculate the loss function of the pipeline pressure prediction model and the measured data, and calculate the partial derivative of the loss function offset;

[0088] The regularization repair unit is used to multiply the partial derivative under the current base pressure by a preset coefficient to obtain the size of the regularization factor;

[0089] The fitting output unit is used to regularize the relevant items in the pipeline pressure prediction model and output the regularized pressure prediction model.

[0090] The early warning adjustment module is used to input the blast furnace working data into the pipeline pressure fitting model at fixed intervals to obtain the probability distribution function of the pressure in the pipeline, multiply the interval length of the probability distribution exceeding the preset confidence interval by the function value, and obtain the impact cumulative amount. When the impact cumulative amount exceeds a first threshold, the pressure relief threshold of the pressure relief valve is adjusted, and when it exceeds a second threshold, a pressure early warning signal is issued.

[0091] The early warning adjustment module includes: a pressure relief threshold unit and an accumulation early warning unit;

[0092] The pressure relief threshold unit is used to predict the probability distribution function of the pipeline pressure in the future period by using the pressure prediction model at preset time intervals, and adjust the pressure relief threshold when the average expectation of the probability distribution function is higher than the threshold;

[0093] The cumulative warning unit is used to calculate the cumulative impact amount of the pipeline and issue a warning when the cumulative impact amount exceeds a threshold.

[0094] like Figure 2 As shown, the blast furnace pipeline pressure analysis and early warning method based on the improved neural network includes the following steps:

[0095] Step S1. A pressure relief valve is provided on the outer wall of the blast furnace pressure pipeline, and a sensor is provided in the feed pipeline. The pressure indication of the pressure relief valve and the readings of each sensor are read according to different sampling periods, and the read pipeline pressure and blast furnace operating parameters are stored in a database;

[0096] Step S1 includes:

[0097] Step S11. A pressure relief valve is arranged in the blast furnace pressure pipeline, wherein the pressure relief valve includes a gravity pressure relief valve and an electronic pressure relief valve, and a sensor is arranged in the feeding pipeline, wherein the feeding pipeline includes a fuel pipeline, a furnace chamber pipeline and a blast chamber pipeline;

[0098] Step S12. Preset a pressure sampling period T1 and a working sampling period T2, where T1 < T2 and T2 is an integer multiple of T1. Read the reading of the pressure relief valve every T1 time duration to obtain pipeline pressure data, and read the readings of each sensor every T2 time duration to obtain the working parameters of the blast furnace. The working parameters include: fuel feeding speed, combustion chamber temperature, and blast chamber air velocity;

[0099] Step S13. Use the working data obtained from a single sampling as the table header of the data table, and use the pipeline pressure data collected within the sampling period of the working data as the data within the table and store it in the database to form a sampling data table.

[0100] Step S2. Take the mean value of the pipeline pressure within the sampling period as the output point, and the working parameters of the blast furnace as trend variables. Conduct an exponential trend test on the mean value of the pipeline pressure, substitute the obtained test parameters into a preset function model, and output the basic pressure function of the pipeline pressure mean value and the working parameters of the blast furnace;

[0101] Step S2 includes:

[0102] Step S21. For the sampling data table, take the mean value of the data within the table as the output point, and use the header data as trend variables to conduct an exponential trend test. The exponential trend test is as follows:

[0103] List the following trend test equations:

[0104]

[0105] where D1, D2,... Dm respectively represent the mean values of the data within the table in the 1st, 2nd,... mth sampling data tables, a1, a2,... an respectively represent the test parameters corresponding to the 1st, 2nd,... nth working parameters of the blast furnace, m is the number of sampling data tables, n is the number of working parameters, and m ≥ n. x(m,n) represents the nth working parameter recorded in the mth sampling data table;

[0106] Solve the trend test equations to obtain the values of a1, a2,... an;

[0107] Step S22. According to the preset function model, output the basic pressure function F(x1, x2,..., xn) of the pipeline pressure mean value and the working parameters of the blast furnace, where F(x1, x2,..., xn) = a1 x1 + a2 x2 +... + an xn , and x1, x2,... xn respectively represent the 1st, 2nd,... nth working parameters.

[0108] Step S3. Conduct a regression fitting on the pipeline pressure data, and replace the cost function matrix of the neural network with the regression matrix corresponding to the minimum cross-entropy function of the sampling data to obtain an improved neural network;

[0109] Step S3 includes:

[0110] Step S31. Calculate the cross entropy loss function of the pipeline pressure data relative to the blast furnace base pressure according to the loss function construction method based on the pipeline pressure data in the sampled data table:

[0111]

[0112] Among them, J(W,D) represents the cross entropy loss function, G(i) represents the i-th pipeline pressure data value in the sampled data table, lg is the logarithm with base 10, W is the cost function matrix, D is the mean of the data in the sampled data table, and e is the base of the natural logarithm;

[0113] Step S32. Calculate the cost function matrix W corresponding to the minimum value of the cross entropy loss function, replace the existing cost function matrix in the neural network model with W, and generate a new neural network model, which is recorded as an improved neural network.

[0114] Step S4. Improve the data in the neural network learning database, output the first model, calculate the loss of the first model under various working parameters, multiply the maximum loss by the preset coefficient to obtain a regularization parameter, regularize the loss function of the first model with the regularization parameter, eliminate the overfitting items in the model, and obtain the second model;

[0115] Step S4 includes:

[0116] Step S41. Using the improved neural network to learn the pipeline pressure data in the database, fitting the probability distribution model of the pipeline pressure value relative to the mean value as the first model output;

[0117] Step S42. Input the working parameters in the sampling data table into the basic pressure function to obtain the basic pressure in the sampling data table, input the basic pressure as the mean into the first model, output the probability distribution function P1(x) of the pipeline pressure, and calculate the loss of the probability distribution function:

[0118]

[0119] Where E is the loss amount of the probability distribution function;

[0120] Step S43. Execute step S42 for all sampled data tables in the database to obtain the minimum value E0 of E, multiply it by the preset magnification coefficient q to obtain the regularization parameter r, and regularize the first model according to the regularization parameter to eliminate the overfitting items in the model:

[0121]

[0122] Among them, J(θ) represents the coefficient function of the first model, θi Represents the current coefficient of the i-th fitting term in the first model;

[0123] Step S44: Correct the coefficients of each fitting item in the first model according to the coefficient function to obtain the second model.

[0124] Step S5. Read the current working parameters of the blast furnace, obtain the pipeline basic pressure from the basic pressure function, input the pipeline basic pressure into the second model, output the probability distribution function and expectation function of the pipeline pressure, and issue an early warning when the probability that the pipeline pressure expectation value is higher than the threshold exceeds the preset confidence interval.

[0125] Step S5 includes:

[0126] Step S51. Obtain the working parameters of the blast furnace charging pipeline, obtain the basic pressure of the pressure pipeline according to the basic pressure function, input the basic pressure into the second model, obtain the probability distribution function of the pipeline pressure, multiply it by the basic pressure, and obtain the expected function of the pipeline pressure;

[0127] Step S52. Preset a pipeline pressure threshold, calculate the probability that the expected function value of the pipeline pressure is above the pressure threshold, and when the obtained probability exceeds the preset confidence interval, send an early warning signal of abnormal blast furnace pipeline pressure to the management center.

[0128] Embodiment: The working parameters are detected every 10 seconds. When the working parameters are [1.5, 20, 14], the pressure of the blast furnace pipeline is detected every 2 seconds. The sampling data is [20, 21, 18, 19, 18]. Through exponential trend fitting, the pressure under the current working parameters is 20 MPa, and the mean of the sampling data is 19 MPa. The improved neural network is used to fit all the sampling data to obtain the second model. The working parameters [1.5, 20, 14] are input into the second model, and the probability of the sampling data being above 20 is 0.1, and the confidence interval is 0.05. It is judged that the pipeline pressure is abnormal and an early warning signal is issued.

[0129] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0130] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A blast furnace pipeline pressure analysis and early warning method based on an improved neural network, characterized in that: The method includes the following steps: Step S1. A pressure relief valve is set on the outer wall of the blast furnace pressure pipeline, and sensors are set in the feed pipeline. The pressure readings of the pressure relief valve and the readings of each sensor are read according to different sampling periods, and the pipeline pressure and blast furnace working parameters read are stored in the database. Step S2. Taking the mean value of the pipeline pressure within the sampling period as the output point and the blast furnace working parameters as trend variables, an exponential trend test is performed on the mean value of the pipeline pressure. The obtained test parameters are substituted into a preset function model to output the basic pressure function of the pipeline pressure mean value and the blast furnace working parameters. Step S3. Perform a regression fitting on the pipeline pressure data, and replace the cost function matrix of the neural network with the regression matrix corresponding to the minimum cross-entropy function of the sampling data to obtain an improved neural network. Step S4. The improved neural network learns the data in the database, outputs the first model, calculates the loss amount of the first model under each working parameter, multiplies the maximum loss amount by a preset coefficient to obtain a regularization parameter, and performs regularization processing on the loss function of the first model with the regularization parameter to eliminate the overfitting term in the model and obtain the second model. Step S5. Read the working parameters of the current blast furnace, obtain the pipeline basic pressure from the basic pressure function, input the pipeline basic pressure into the second model, output the probability distribution function and expected function of the pipeline pressure, and give an alarm when the probability that the pipeline pressure expected value is higher than the threshold exceeds the preset confidence interval.

2. The blast furnace pipeline pressure analysis and early warning method based on improved neural network according to claim 1 is characterized in that: Step S1 includes: Step S11. A pressure relief valve is set in the blast furnace pressure pipeline. The pressure relief valve includes a gravity pressure relief valve and an electronic pressure relief valve. Sensors are set in the feeding pipeline. The feeding pipeline includes a fuel pipeline, a furnace chamber pipeline, and a blast chamber pipeline. Step S12. Preset a pressure sampling period T1 and a working sampling period T2, where T1 < T2 and T2 is an integer multiple of T1. Read the readings of the pressure relief valve every T1 time length to obtain pipeline pressure data, and read the readings of each sensor every T2 time length to obtain the blast furnace working parameters. The working parameters include fuel feeding speed, combustion chamber temperature, and blast chamber air velocity. Step S13. Using the working data obtained from a single sampling as the table header of the data table, and taking the pipeline pressure data collected within the sampling period of the working data as the data within the table and storing it in the database to form a sampling data table.

3. The blast furnace pipeline pressure analysis and early warning method based on improved neural network according to claim 2 is characterized in that: Step S2 includes: Step S21. For the sampling data table, take the mean value of the data within the table as the output point and the header data as trend variables for an exponential trend test. The exponential trend test is as follows: List the following trend test equation: ， where D1, D2,... Dm respectively represent the mean values of the data within the table in the 1st, 2nd,... mth sampling data tables, a1, a2,... an respectively represent the test parameters corresponding to the 1st, 2nd,... nth working parameters of the blast furnace, m is the number of sampling data tables, n is the number of working parameters, and m ≥ n, and x(m,n) represents the nth working parameter recorded in the mth sampling data table. Solve the trend test equation to obtain the values of a1, a2,... an. Step S22. According to the preset function model, output the basic pressure function F(x1,x2,…,xn) of the pipeline pressure mean and the blast furnace working parameters, where F(x1,x2,…,xn)=a1 x1 +a2 x2 +…+an xn , x1, x2, …xn represent the 1st, 2nd, …nth working parameters respectively.

4. The blast furnace pipeline pressure analysis and early warning method based on improved neural network according to claim 3 is characterized in that: Step S3 includes: Step S31. Calculate the cross entropy loss function of the pipeline pressure data relative to the blast furnace base pressure according to the loss function construction method based on the pipeline pressure data in the sampled data table: ， Among them, J(W,D) represents the cross entropy loss function, G(i) represents the i-th pipeline pressure data value in the sampled data table, lg is the logarithm with base 10, W is the cost function matrix, D is the mean of the data in the sampled data table, and e is the base of the natural logarithm; Step S32. Calculate the cost function matrix W corresponding to the minimum value of the cross entropy loss function, replace the existing cost function matrix in the neural network model with W, and generate a new neural network model, which is recorded as an improved neural network.

5. The blast furnace pipeline pressure analysis and early warning method based on improved neural network according to claim 4 is characterized in that: Step S4 includes: Step S41. Using the improved neural network to learn the pipeline pressure data in the database, fitting the probability distribution model of the pipeline pressure value relative to the mean value as the first model output; Step S42. Input the working parameters in the sampling data table into the basic pressure function to obtain the basic pressure in the sampling data table, input the basic pressure as the mean into the first model, output the probability distribution function P1(x) of the pipeline pressure, and calculate the loss of the probability distribution function: ， Where E is the loss amount of the probability distribution function; Step S43. Execute step S42 for all sampled data tables in the database to obtain the minimum value E0 of E, multiply it by the preset magnification coefficient q to obtain the regularization parameter r, and regularize the first model according to the regularization parameter to eliminate the overfitting items in the model: ， Among them, J(θ) represents the coefficient function of the first model, θ i Represents the current coefficient of the i-th fitting term in the first model; Step S44. Correct the coefficients of each fitting item in the first model according to the coefficient function to obtain a second model; Step S5 includes: Step S51. Obtain the working parameters of the blast furnace charging pipeline, obtain the basic pressure of the pressure pipeline according to the basic pressure function, input the basic pressure into the second model, obtain the probability distribution function of the pipeline pressure, multiply it by the basic pressure, and obtain the expected function of the pipeline pressure; Step S52. Preset a pipeline pressure threshold, calculate the probability that the expected function value of the pipeline pressure is above the pressure threshold, and when the obtained probability exceeds the preset confidence interval, send an early warning signal of abnormal blast furnace pipeline pressure to the management center.

6. The blast furnace pipeline pressure analysis and early warning system based on improved neural network is characterized by: The system includes the following modules: pipeline sensing module, furnace chamber pressure module, probability prediction module, improved network module and early warning adjustment module; The pipeline sensing module is composed of a gravity pressure relief valve arranged in the blast furnace pipeline and a sensor inside the valve. The pressure relief valve is used to adjust the internal pressure of the pipeline. The sensor is used to detect the height of the counterweight block in the gravity pressure relief valve, thereby reading the internal pressure of the pipeline and being able to change the pressure relief threshold of the pressure relief valve according to instructions. The furnace chamber pressure module is used to record the blast furnace operating parameters according to the sensor in the feed pipeline, sample the pipeline pressure and the operating parameters at the same time, store the sampling results in the database, use the sampled data in the database to perform exponential trend detection, generate a basic pressure function, and generate a probability distribution function of the pipeline pressure under the current operating parameters according to the measured pipeline pressure data; The probability prediction module is used to define a regression matrix, use the regression matrix as a transformation matrix, the current base pressure as a mean value, and the actual pipeline pressure as input data, construct a cross entropy function, calculate the minimum value of the cross entropy function, use the regression matrix corresponding to the minimum value as a cost matrix, use a convolutional neural network to learn the sampled data in the database, and output a pipeline pressure prediction model after training; The improved network module is used to generate a model loss function with the deviation of the model in the measured data as the dependent variable and the data sequence number as the independent variable, calculate the partial derivatives of the deviation of the pipeline pressure prediction model in each dimension, determine the size of the regularization factor by the partial derivative sequence, perform regularization processing on the model loss function, eliminate the overfitting items in the model, and output the corrected pipeline pressure fitting model; The early warning adjustment module is used to input the blast furnace working data into the pipeline pressure fitting model at fixed intervals to obtain the probability distribution function of the pressure in the pipeline, multiply the interval length of the probability distribution exceeding the preset confidence interval by the function value, and obtain the impact cumulative amount. When the impact cumulative amount exceeds a first threshold, the pressure relief threshold of the pressure relief valve is adjusted, and when it exceeds a second threshold, a pressure early warning signal is issued.

7. The blast furnace pipeline pressure analysis and early warning system based on improved neural network according to claim 6 is characterized by: The pipeline sensing module includes: a pressure relief valve unit and an internal pressure recording unit; The pressure relief valve unit is used to release the internal pressure of the pipeline to the outside when the internal pressure of the pipeline is greater than the pressure relief threshold pressure, until the internal pressure of the pipeline is lower than the threshold pressure; The internal pressure recording unit is used to read the internal pressure of the pipeline according to the height of the counterweight block on the pressure relief valve.

8. The blast furnace pipeline pressure analysis and early warning system based on improved neural network according to claim 7 is characterized by: The furnace chamber pressure module includes: a parameter sampling unit, a database unit and a basic pressure unit; The parameter sampling unit is used to sample the working parameters of the blast furnace and the internal pressure of the pipeline at the same time, wherein the working parameters include the blast furnace charging speed, the temperature and humidity of the combustion chamber, and the wind speed of the blast chamber; The database unit is used to store sampled data and construct the sampled data into a data set suitable for neural network training; The basic pressure unit is used to perform exponential trend detection on the sampled data and output a basic pressure function, which reflects the relationship between each working parameter and the average value of the pipeline pressure.

9. The blast furnace pipeline pressure analysis and early warning system based on improved neural network according to claim 8 is characterized in that: The probability prediction module includes: a cross regression unit and a cost input unit; The cross regression unit is used to set an unknown regression matrix, fit the cross entropy function with the base pressure as the mean, and use the regression matrix output corresponding to the minimum value of the cross entropy function as the cost matrix; The cost input unit is used to set the cost matrix as the objective function matrix of the convolutional neural network to learn the sampled data in the database and output the pipeline pressure prediction model; The improved network module includes: a model loss unit, a regularization repair unit and a fitting output unit; The model loss unit is used to calculate the loss function of the pipeline pressure prediction model and the measured data, and calculate the partial derivative of the loss function offset; The regularization repair unit is used to multiply the partial derivative under the current base pressure by a preset coefficient to obtain the size of the regularization factor; The fitting output unit is used to regularize the relevant items in the pipeline pressure prediction model and output the regularized pressure prediction model.

10. The blast furnace pipeline pressure analysis and early warning system based on improved neural network according to claim 9 is characterized in that: The early warning adjustment module includes: a pressure relief threshold unit and an accumulation early warning unit; The pressure relief threshold unit is used to predict the probability distribution function of the pipeline pressure in the future period by using the pressure prediction model at preset time intervals, and adjust the pressure relief threshold when the average expectation of the probability distribution function is higher than the threshold; The cumulative warning unit is used to calculate the cumulative impact amount of the pipeline and issue a warning when the cumulative impact amount exceeds a threshold.

Images