A steel bar cooling control system based on machine learning

Abstract

The present invention discloses a steel bar cooling control system based on machine learning, including a function test unit, a signal selector, a jitter filtering and sampling unit, a two-stage filtering and buffering sampling unit, a timing measurement unit, a control signal generator, a machine learning data collection unit, a model training unit, a trend compensation unit, a PI error compensation unit, and a compensation control unit; wherein, the function test unit is used to simulate the process signals of real steel bar rolling. The present invention realizes precise temperature control and rapid adjustment speed, enables the temperature of hot-rolled steel bars to drop sharply, ensures the tissue transformation temperature of hot-rolled steel bars, prevents the hot-rolled steel bars from being bent when placed in a cooling medium, thereby improving the qualified rate of steel bars in the production process of hot-rolled steel bars.

Description

Technical Field

[0001] The present invention relates to the technical field of hot-rolled steel bar production, and specifically to a steel bar cooling control system based on machine learning. Background Art

[0002] Hot-rolled steel bars are finished steel bars formed by hot rolling and naturally cooled, which are pressed from low-carbon steel and ordinary alloy steel at high temperature states. They are mainly used for the reinforcement of reinforced concrete and prestressed concrete structures, and are one of the steel products with the largest usage in civil engineering. The controlled rolling and controlled cooling process is an advanced technology to improve the strength and toughness of steel, and has been widely applied in steel rolling production. This technology fully exploits the potential of steel through process means, significantly improves and enhances the comprehensive mechanical properties of steel, so as to achieve the purposes of saving metal, developing new steel varieties, and saving energy and reducing consumption, bringing significant economic benefits to enterprises. Post-rolling cooling will affect the performance and shape of the rolled material, and sometimes also affect the output of the rolling mill.

[0003] However, there are some drawbacks in the cooling mechanism of traditional hot-rolled steel bar production devices during use, such as:

[0004] The existing hot-rolled steel bar cooling control system has poor temperature control effect, slow adjustment speed, and cannot control the temperature in time. During the rapid cooling stage, the hot-rolled steel bars are prone to bending when cooled in the cooling medium, which affects the production efficiency and quality of hot-rolled steel bars and the qualification rate. Summary of the Invention

[0005] The purpose of the present invention is to provide a steel bar cooling control system based on machine learning to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A steel bar cooling control system based on machine learning includes a function test unit, a signal selector, a jitter filtering sampling unit, a two-stage filtering buffer sampling unit, a timing measurement unit, a control signal generator, a machine learning data collection unit, a model training unit, a trend compensation unit, a PI error compensation unit, and a compensation control unit; wherein,

[0008] The function test unit is used to simulate the process signal of real steel bar rolling, and calculate and output the outlet steel temperature signal according to the test parameters set by the user through the input valve control data.

[0009] The signal selector is used to select the actual signal or the output signal simulated by the function test unit according to the parameters set by the user in cooperation with the function test unit.

[0010] The jitter filtering and sampling unit is used to perform anti-jitter filtering on the hot metal detection input signal according to the filtering parameters set by the user;

[0011] The two-stage filtering and buffering sampling unit is used to filter the input signals of the inlet steel temperature and the outlet steel temperature according to the filtering parameters set by the user;

[0012] The timing measurement unit is used to dynamically calculate the working parameters of the control signal generator according to the set delay parameters and measurement results of the user;

[0013] The control signal generator is used to generate enable signals for trend compensation, error compensation, cooling control, and feedback control in a fixed-time mode or a dynamic loading mode calculated according to speed, according to the working mode set by the user;

[0014] The machine learning data collection unit is used to collect parameters according to the settings of the user, sample and record the inlet steel temperature, the outlet steel temperature, and the valve control data;

[0015] The model training unit is used to train the model according to the model training parameters set by the user, and use the training data output by the machine learning data collection unit to solve the optimal control parameters;

[0016] The PI error compensation unit is used to generate error compensation data according to the proportional and integral parameters input by the user, in cooperation with the outlet steel temperature data;

[0017] The compensation control unit is used to control the coordinated work of trend compensation, error compensation, and machine learning.

[0018] As a further solution of the present invention: The trend compensation unit includes a second-order exponential smoothing predictor, a data point preset valve control curve generation unit, a preset steel temperature gradient valve control curve generation unit, a model real-time calculation unit, a data selector, and a synchronization signal generator; wherein,

[0019] The second-order exponential smoothing predictor is used to collect and record the inlet steel temperature data according to the smoothing parameters input by the user, and predict the steel temperature data in advance by using the second-order exponential smoothing prediction algorithm;

[0020] The data point preset valve control curve generation unit is used to output trend compensation data according to the curve information data table set by the user;

[0021] The preset steel temperature gradient valve control curve generation unit is used to input the temperature changing according to the gradient into the model to calculate the trend compensation data with the starting temperature of the steel head and the gradient set by the user;

[0022] The model real-time calculation unit is used to dynamically calculate trend compensation data according to the change of the inlet steel temperature data and the model control parameters set by the user or obtained by machine learning.

[0023] As a further solution of the present invention: The function test unit includes a hot metal detection signal output, an inlet steel temperature signal output, and an outlet steel temperature signal output.

[0024] As a further solution of the present invention: The input end of the signal selector is connected to the output end of the function test unit, the output end of the signal selector is respectively connected to the input ends of the jitter filtering sampling unit, the two-stage filtering buffer sampling unit, and the timing measurement unit, the output end of the jitter filtering sampling unit is connected to the input end of the control signal generator, the output end of the timing measurement unit is connected to the input end of the control signal generator, the output end of the two-stage filtering buffer sampling unit is connected to the input end of the compensation control unit, the output end of the control signal generator is connected to the input end of the supplementary control unit, the cooling enable signal output end of the control signal generator is connected to the input end of the function test unit, and a reverse enable signal output end is also provided on the control signal generator, the output end of the compensation control unit is connected to the input end of the function test unit, a reverse valve opening output end is also provided on the compensation control unit, and the compensation control unit is respectively connected to the model training unit, the machine learning data collection unit, the PI error compensation unit, and the trend compensation unit.

[0025] As a further solution of the present invention: The output end of the compensation control unit is respectively connected to the input ends of the second-order exponential smoothing predictor, the data point preset valve control curve generation unit, and the preset steel temperature gradient valve control curve generation unit, the output end of the second-order exponential smoothing predictor is connected to the input end of the model real-time calculation unit, the output ends of the data point preset valve control curve generation unit, the preset steel temperature gradient valve control curve generation unit, and the model real-time calculation unit are connected to the input end of the data selector, the output end of the data selector is connected to the input end of the synchronization signal generator, and the output ends of the data selector and the synchronization signal generator are connected to the input end of the compensation control unit.

[0026] As a further solution of the present invention: The first-stage filtering of the two-stage filtering buffer sampling unit is sliding window filtering, and the second-stage filtering is buffer peak filtering.

[0027] As a further solution of the present invention: The timing measurement unit is used to measure the time from the effective of the hot metal detection signal to the effective of the inlet steel temperature in real time, and measure the time from the effective of the hot metal detection signal to the effective of the outlet steel temperature in real time.

[0028] As a further aspect of the present invention: The control signal generator includes enable signals for generating trend compensation, error compensation, cooling control, and backlash control.

[0029] As a further aspect of the present invention: The PI error compensation unit includes a synchronous compensation mode and an asynchronous compensation mode.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] The present invention achieves precise temperature control and rapid adjustment speed, enabling the temperature of hot-rolled steel bars to drop sharply, ensuring the tissue transformation temperature of hot-rolled steel bars, preventing bending when the hot-rolled steel bars are placed in the cooling medium, and thus improving the qualification rate of steel bars during the production process of hot-rolled steel bars. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 FIG. is a schematic structural diagram of a steel bar cooling control system based on machine learning.

[0033] Figure 2 FIG. is a program flowchart of a function test unit in a steel bar cooling control system based on machine learning.

[0034] Figure 3 FIG. is a program flowchart of a jitter filtering sampling unit in a steel bar cooling control system based on machine learning.

[0035] Figure 4 FIG. is a program flowchart of a two-stage filtering buffer sampling unit in a steel bar cooling control system based on machine learning.

[0036] Figure 5 FIG. is a program flowchart of a timing measurement unit in a steel bar cooling control system based on machine learning.

[0037] Figure 6 FIG. is a program flowchart of a control signal generator in a steel bar cooling control system based on machine learning.

[0038] Figure 7 FIG. is a program flowchart of a model training unit in a steel bar cooling control system based on machine learning.

[0039] Figure 8 FIG. is a program flowchart of a PI error compensation unit in a steel bar cooling control system based on machine learning.

[0040] Figure 9 FIG. is a program flowchart of a second-order exponential smoothing predictor in a steel bar cooling control system based on machine learning.

[0041] Figure 10It is a program flow chart of a data point preset valve control curve generation unit in a steel bar cooling control system based on machine learning.

[0042] Figure 11 It is a program flow chart of a preset steel temperature gradient valve control curve generation unit in a steel bar cooling control system based on machine learning.

[0043] Figure 12 It is a program flow chart of a model real-time calculation unit in a steel bar cooling control system based on machine learning.

[0044] Figure 13 It is a program flow chart of a trend compensation unit in a steel bar cooling control system based on machine learning.

[0045] Figure 14 It is a program flow chart of a trend compensation state machine in a steel bar cooling control system based on machine learning.

[0046] Figure 15 It is a program flow chart of a compensation control unit in a steel bar cooling control system based on machine learning.

[0047] Figure 16 It is a program flow chart of an error compensation state machine in a steel bar cooling control system based on machine learning. Detailed implementation manners

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Please refer to Figures 1 - 16 , in the embodiments of the present invention, a steel bar cooling control system based on machine learning includes:

[0050] A function test unit. The function test unit can simulate the process signals of real steel bar rolling at the equipment work site, including hot metal detection signal output, inlet steel temperature signal output and outlet steel temperature signal output, and calculate and output the outlet steel temperature signal according to the test parameters set by the user through the valve control data input externally. Before passing steel, it helps the user to conduct on-line testing of the hardware equipment on site. The program interface description of the function test unit is shown in Table 1. The program adopts the state machine programming idea and non-blocking operation mechanism. The program flow chart is as Figure 2 shown;

[0051] Table 1 Interface description of the function test unit

[0052]

[0053] The jitter filtering sampling unit filters the input digital signal of the hot metal detection according to the filtering parameters set by the user to prevent jitter and avoid interfering with the control signal generator. Both the sampling period and the number of sampling points can be set according to on-site requirements. The program interface description of the jitter filtering sampling unit is shown in Table 2. The program adopts the state machine programming concept and the non-blocking operation mechanism. The program flow chart is as Figure 3 shown;

[0054] Table 2 Interface Description of the Jitter Filtering Sampling Unit

[0055] Name Data Type Function stat Bool Signal Input sysTm LInt Current System Time Input prdMs Int Sampling Interval potNum Int Number of Consecutive Valid Points for Judgment paramsErrFlg Bool Parameter Error Flag Output realStat Bool Current Signal Status Output ftHdV Int Judgment Signal Value Output ftHdStat Bool Judgment Signal Status Output

[0056] The two-stage filtering buffer sampling unit filters the input signals of the inlet steel temperature and the outlet steel temperature according to the filtering parameters set by the user. The first-stage filtering is sliding window filtering, and the second-stage filtering is buffer peak filtering. The sampling period, the number of sampling points, and the working mode can all be set according to on-site requirements. Each stage of the two-stage filtering buffer sampling unit is provided with a separate output interface. The program interface description of the two-stage filtering buffer sampling unit is shown in Table 3. The program adopts the state machine programming concept and the non-blocking operation mechanism. The program flow chart is as Figure 4 shown;

[0057] Table 3 Interface Description of the Two-stage Filtering Buffer Sampling Unit

[0058]

[0059] The timing measurement unit measures in real time the effective passing time of the steel bar from the installation position of the hot metal detection to the installation position of the inlet temperature measurement and from the installation position of the hot metal detection to the installation position of the outlet temperature measurement during the steel bar production process. According to the set distance and time parameters of the user and in combination with the measurement results, it dynamically calculates the enabling time of the control signal of the control signal generator, including the control time of the trend compensation enabling signal, the control time of the counterattack enabling signal, the control time of the cooling enabling signal, the control time of the error compensation enabling signal, the machine learning offset time, the adjustment period, and other time signals. The program interface description of the timing measurement unit is shown in Table 4. The program adopts the state machine programming concept and the non-blocking operation mechanism. The program flow chart is as Figure 5 shown;

[0060] Table 4 Interface Description of the Timing Measurement Unit

[0061] Name Data Type Function MCStat Bool Hot Metal Detection Signal Input tdPV Real Inlet Steel Temperature Input errPV Real Outlet Steel Temperature Input sysTm LInt Current System Time Input ADThr Real Steel Temperature Threshold ivldChkTmThr Int Invalid Duration Threshold tmOvThr DInt Measurement Timeout measTmMax DInt Maximum Measurement Time

[0062] The control signal generator generates enabling signals for trend compensation, error compensation, cooling control, and counterattack control. It can select the fixed-time mode, the dynamic loading mode calculated according to speed, and the dynamic loading mode calculated according to the timing measurement results according to the working mode set by the user. The program interface description of the control signal generator is shown in Table 5. The program adopts the state machine programming idea and the non-blocking operation mechanism. The program flow chart is as Figure 6 shown;

[0063] Table 5 Interface description of the control signal generator

[0064] Name Data Type Function MCStat Bool Hot Metal Detection Signal Input sysTm LInt Current System Time Input tdVldDlyTm DInt Trend Compensation Valid Signal Delay Time CCVldDlyTm DInt Cooling Valid Signal Delay Time errVldDlyTm DInt Error Compensation Valid Signal Delay Time tdIvldDlyTm DInt Trend Compensation Invalid Signal Delay Time errIvldDlyTm DInt Error Compensation Invalid Signal Delay Time CCIvldDlyTm DInt Cooling Invalid Signal Delay Time CCRToCCTm Int Retaliation Action Relative to Cooling Action Delay Time

[0065] The model training unit trains the model according to the model training parameters set by the user and uses the training data output by the machine learning data collection unit to solve the optimal control parameters. The program interface description of the model training unit is shown in Table 6. The program adopts the state machine programming idea and the non-blocking operation mechanism. The program flow chart is as Figure 7 shown;

[0066] Table 6 Interface description of the model training unit

[0067]

[0068] The PI error compensation unit has a PI regulator as its internal core. Its main feature is the rapidity of compensation, reducing or eliminating the static error. According to the proportional and integral parameters input by the user and in combination with the outlet steel temperature data, it generates error compensation data. It can select the synchronous compensation mode and the asynchronous compensation mode. The program interface description of the PI error compensation unit is shown in Table 7. The program adopts the state machine programming idea and the non-blocking operation mechanism. The program flow chart is as Figure 8 shown;

[0069] Table 7 Interface description of the PI error compensation unit

[0070] Name Data Type Function rst Bool Error Compensation Unit Reset Input man Bool Manual Setting Mode manValue Real Manual Setting Value sysTm LInt Current System Time Input PV Real Outlet Steel Temperature Input tgtValue Real Outlet Steel Temperature Target Value sync Bool Synchronization Signal Input Sync Bool Synchronization Adjustment Mode cof0 Real Coefficient P cof1 Real Coefficient I prdMs Int Adjustment Period cpePVThr Real Outlet Steel Temperature Input Threshold actMaxStepVaule Real Compensation Maximum Step Size actRg Rg Compensation Range

[0071] The second-order exponential smoothing predictor collects and records the inlet steel temperature data according to the smoothing parameters input by the user, and uses the second-order exponential smoothing prediction algorithm to predict the steel temperature data in advance. The program interface description of the second-order exponential smoothing predictor is shown in Table 8. The program adopts the state machine programming idea and the non-blocking operation mechanism. The program flow chart is as Figure 9 shown;

[0072] Table 8 Interface description of the second-order exponential smoothing predictor

[0073] Name Data Type Function rst Bool Predictor Reset Input dataVaule Real Data Current Value Input fcPrdNum Int Number of Periods of Predicted Data mtd Int Prediction Mode DataMinNum Int Minimum Amount of Predicted Data fcCof Real Prediction Coefficient fcVldPrdNum Int Number of Valid Periods of Prediction linParams LinParams Maximum Derivative of Predicted Data

[0074] Data Point Preset Valve Control Curve Generation Unit. The Data Point Preset Valve Control Curve Generation Unit outputs trend compensation data according to the curve information data table set by the user. The program interface description of the Data Point Preset Valve Control Curve Generation Unit is shown in Table 9. The program adopts the state machine programming idea and non-blocking operation mechanism. The program flow chart is as shown in Figure 10 shown;

[0075] Table 9 Interface Description of Data Point Preset Valve Control Curve Generation Unit

[0076] Name Data Type Function rst Bool Reset Input sysTm LInt Current System Time Input Params PCvParams Output maximum first derivative rmTm DInt Curve remaining time PCvValue Real Curve value PCvBuff PCvBasicData Data table

[0077] Preset Steel Temperature Gradient Valve Control Curve Generation Unit. The Preset Steel Temperature Gradient Valve Control Curve Generation Unit inputs the temperature changing according to the gradient into the model to calculate the trend compensation data with the starting temperature of the steel head and the gradient set by the user. The program of the Preset Steel Temperature Gradient Valve Control Curve Generation Unit adopts the state machine programming idea and non-blocking operation mechanism. The program flow chart is as shown in Figure 11 shown;

[0078] Model Real-time Calculation Unit. According to the change of the inlet steel temperature data and the model control parameters set by the user or obtained by machine learning, it dynamically calculates the trend compensation data. The program interface description of the Model Real-time Calculation Unit is shown in Table 10. The program adopts the state machine programming idea and non-blocking operation mechanism. The program flow chart is as shown in Figure 12 shown;

[0079] Table 10 Interface Description of Model Real-time Calculation Unit

[0080] Name Data type Function PV Real Inlet steel temperature input PVOs Real Outlet steel temperature target value input cof0 Real Model coefficient 0 cof1 Real Model coefficient 1 cof2 Real Model coefficient 2 cpeVaule PCvBasicData Model calculated value

[0081] Trend Compensation Unit. The Trend Compensation Unit compensates the temperature change trend according to the steel bar steel head temperature data, the real-time inlet steel temperature data and the user-set information. The Trend Compensation Unit has three compensation modes, namely the data point preset valve control curve mode, the preset steel temperature gradient valve control mode and the model real-time calculation mode. The program interface description of the Trend Compensation Unit is shown in Table 11. The program adopts the state machine programming idea and non-blocking operation mechanism. The program flow chart is as shown in Figure 13 shown;

[0082] Table 11 Interface Description of Trend Compensation Unit

[0083]

[0084] The compensation control unit controls parts such as trend compensation, error compensation, and machine learning based on the control signals generated by the control signal generator, the inlet steel temperature data, and the outlet steel temperature data, calculates the compensation data to compensate the steel bar temperature, and realizes the precise and rapid control of the steel bar temperature. The program interface description of the compensation control unit is shown in Table 12. The program adopts the state machine programming idea and the non-blocking operation mechanism. The program flow chart of the trend compensation state machine is as shown in Figure 14 shown, and the program flow chart of the compensation control unit is as shown in Figure 15 , and the program flow chart of the error compensation state machine is as shown in Figure 16 shown;

[0085] Table 12 Interface description of the trend compensation unit

[0086] Name Data type Function sysTm LInt Current system time input CCEn Bool Cooling enable signal input tdEn Bool Trend compensation enable signal input errEn Bool Error compensation enable signal input tdPVDataUpdPrd Int Inlet steel temperature update period tdPVData DCADFtData Inlet steel temperature data errPVData DCADFtData Outlet steel temperature data dataOutput CpeData Compensation output data trnResApply Bool Online application of training results PCvBuff Bool Preset curve data table buffer trnBuff Int Machine learning data collection buffer tgtVaule Real Outlet steel temperature target value manParams ManParams Manual / automatic control parameters actRg Rg Compensation range trnParams TrnParams Machine learning parameters tdCpeParams TdCpeParams Trend compensation parameters errCpeParams ErrCpeParams Error compensation parameters PFParams PFParams Following parameters

[0087] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent replacements or changes, and should be covered within the protection scope of the present invention.

Claims

1. A steel bar cooling control system based on machine learning, characterized in that, It includes a functional test unit, a signal selector, a jitter filtering and sampling unit, a two-stage filtering and buffering sampling unit, a timing measurement unit, a control signal generator, a machine learning data collection unit, a model training unit, a trend compensation unit, a PI error compensation unit, and a compensation control unit. Among them, The functional test unit is used to simulate the process signals of real steel rolling, and calculate and output the outlet steel temperature signal based on the test parameters set by the user through the input valve control data. The signal selector is used to select the actual signal or the output signal simulated by the functional test unit according to the parameters set by the user and in cooperation with the functional test unit. The jitter filtering and sampling unit is used to perform anti-jitter filtering on the hot metal detection input signal according to the filtering parameters set by the user. The two-stage filtering and buffering sampling unit is used to filter the inlet steel temperature and outlet steel temperature input signals according to the filtering parameters set by the user. The timing measurement unit is used to dynamically calculate the working parameters of the control signal generator according to the set delay parameters and measurement results of the user. The control signal generator is used to generate the enable signals for trend compensation, error compensation, cooling control, and feedback control in a fixed-time mode or a dynamic loading mode calculated according to speed, which can be selected according to the working mode set by the user. The machine learning data collection unit is used to collect parameters according to the settings of the user, sample and record the inlet steel temperature, outlet steel temperature, and valve control data. The model training unit is used to train the model according to the model training parameters set by the user and use the training data output by the machine learning data collection unit to solve the optimal control parameters. The PI error compensation unit is used to generate error compensation data according to the proportional and integral parameters input by the user in cooperation with the outlet steel temperature data. The compensation control unit is used to control the coordinated work of trend compensation, error compensation, and machine learning.

2. The steel bar cooling control system based on machine learning according to claim 1, characterized in that, The trend compensation unit includes a second-order exponential smoothing predictor, a data point preset valve control curve generation unit, a preset steel temperature gradient valve control curve generation unit, a model real-time calculation unit, a data selector, and a synchronization signal generator. Among them, The second-order exponential smoothing predictor is used to collect and record the inlet steel temperature data according to the smoothing parameters input by the user, and predict the steel temperature data in advance by using the second-order exponential smoothing prediction algorithm. The data point preset valve control curve generation unit is used to output trend compensation data according to the curve information data table set by the user. The preset steel temperature gradient valve control curve generation unit is used to input the temperature changing according to the gradient into the model to calculate the trend compensation data with the starting temperature of the steel head and the gradient set by the user. The model real-time calculation unit is used to dynamically calculate the trend compensation data according to the change of the inlet steel temperature data and the model control parameters set by the user or obtained by machine learning.

3. The steel bar cooling control system based on machine learning according to claim 1, characterized in that The functional test unit includes a hot metal detection signal output, an inlet steel temperature signal output, and an outlet steel temperature signal output.

4. A steel bar cooling control system based on machine learning according to claim 1, characterized in that, The input end of the signal selector is connected to the output end of the function test unit. The output end of the signal selector is respectively connected to the input ends of the jitter filtering sampling unit, the two-stage filtering buffer sampling unit and the timing measurement unit. The output end of the jitter filtering sampling unit is connected to the input end of the control signal generator. The output end of the timing measurement unit is connected to the input end of the control signal generator. The output end of the two-stage filtering buffer sampling unit is connected to the input end of the compensation control unit. The output end of the control signal generator is connected to the input end of the supplementary control unit. The cooling enable signal output end of the control signal generator is connected to the input end of the function test unit. And a reverse enable signal output end is further provided on the control signal generator. The output end of the compensation control unit is connected to the input end of the function test unit. A reverse valve opening output end is further provided on the compensation control unit. The compensation control unit is respectively connected to the model training unit, the machine learning data collection unit, the PI error compensation unit and the trend compensation unit.

5. A steel bar cooling control system based on machine learning according to claim 2, characterized in that, The output end of the compensation control unit is respectively connected to the input ends of the second-order exponential smoothing predictor, the data point preset valve control curve generation unit and the preset steel temperature gradient valve control curve generation unit. The output end of the second-order exponential smoothing predictor is connected to the input end of the model real-time calculation unit. The output ends of the data point preset valve control curve generation unit, the preset steel temperature gradient valve control curve generation unit and the model real-time calculation unit are connected to the input end of the data selector. The output end of the data selector is connected to the input end of the synchronization signal generator. And the output ends of the data selector and the synchronization signal generator are connected to the input end of the compensation control unit.

6. The steel bar cooling control system based on machine learning according to claim 1, characterized in that, The first-stage filtering of the two-stage filtering buffer sampling unit is sliding window filtering, and the second-stage filtering is buffer peak filtering.

7. A steel bar cooling control system based on machine learning according to claim 1, characterized in that, The timing measurement unit is used to measure in real time the time from the effectiveness of the hot metal detection signal to the effectiveness of the inlet steel temperature, and measure in real time the time from the effectiveness of the hot metal detection signal to the effectiveness of the outlet steel temperature.

8. A steel bar cooling control system based on machine learning according to claim 1, characterized in that, The control signal generator includes enabling signals for generating trend compensation, error compensation, cooling control and reverse control.

9. A steel bar cooling control system based on machine learning according to claim 1, characterized in that, The PI error compensation unit includes a synchronous compensation mode and an asynchronous compensation mode.

Images