A knowledge and data fusion based industrial process predictive control method and system

Abstract

The application discloses a kind of knowledge and data fusion's industrial process prediction control method and system, in first stage, i.e. " learning person " stage, according to the experience knowledge of operator, rule is condensed, control learning imitates the control behavior of operator by fuzzy rule, but control cycle changes from non-periodic low frequency to periodic high frequency;After experiencing the knowledge-driven control of first stage, sufficient high-frequency high-quality input-output data will be generated, which can fully reflect the influence of the change of manipulated variable on the controlled variable, so in the second stage, i.e. " beyond person " stage, based on the data set, the system model is identified to fit the current operating state, and the model predictive control algorithm is used to calculate the decision control variable, while the predictive model is updated adaptively to adapt to the changing operating conditions.The application combines the advantages of knowledge-driven control and data-driven control method, and realizes stable control with initial reliability and later adaptive ability.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of industrial process control, and particularly relates to a knowledge and data fusion industrial process predictive control method and system. BACKGROUND

[0002] Due to the lack of accurate mechanism models for some complex industrial processes such as roasting processes, the currently widely used intelligent control methods are knowledge-driven and data-driven methods. In the research of knowledge-driven control algorithms, Feng et al. proposed a trend-based event-triggered fuzzy control strategy, which combines qualitative trend analysis and event-triggered mechanism fuzzy control design, for temperature stability control of large roasting furnaces. In the research field of data-driven control methods, Asadi et al. proposed a data-driven adaptive controller for wide-area multi-input multi-output (MIMO) nonlinear systems based on input-output saturation, and applied it to the load frequency control problem of interconnected multi-region power grids with speed governor saturation, load and parameter uncertainty.

[0003] Although the knowledge-driven and data-driven methods have achieved certain application effects in the industrial field, there are still some problems that restrict the further improvement of control performance. The knowledge-driven control method can only perform rough control based on fuzzy rules, and the experience knowledge used has timeliness, which is difficult to guarantee long-term stable operation. On the other hand, the control effect of the data-driven method depends on the data quality, and poor quality data will affect the extraction of the industrial process running characteristics, thereby affecting the decision-making of the control algorithm. SUMMARY

[0004] The present application provides a knowledge-driven model predictive control industrial furnace control method and system, which makes the operating state of the roasting furnace reach a "dynamic balance" and realizes the roasting temperature stability control in a narrow operating interval.

[0005] To achieve the above technical purposes, the present application adopts the following technical solutions:

[0006] A knowledge and data fusion industrial process predictive control method, comprising:

[0007] Step 1, determining the main control quantity, the secondary control quantity and the output quantity corresponding to each control quantity of the industrial process, taking each control quantity and the corresponding output quantity as the output variable and the input variable of the fuzzy rule respectively, and constructing the corresponding fuzzy rule table; wherein the input variable threshold setting and the control quantity output value in each fuzzy rule table are obtained by statistically distributing the historical data thereof;

[0008] Building a data-driven model predictive control algorithm according to the historical control quantity and output quantity data;

[0009] Step 2, when the industrial process control system is put into operation, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, that is, the historical output data of each control variable is obtained, and the corresponding control variable is obtained based on the fuzzy rule table, and each control variable obtained is applied to the industrial process;

[0010] Step 3, when it is identified that the industrial process is in the intelligent control state for a preset time period in the past, the data-driven model predictive control algorithm is used to control the industrial process;

[0011] Wherein, the control state of the industrial process includes manual control and intelligent control, and the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and the data-driven model predictive control algorithm is used to control the industrial process, both of which belong to the intelligent control state;

[0012] Step 4, when it is identified that the industrial process is in the manual control state for a control period in the past, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and the process is switched to step 3.

[0013] Further, the input variable threshold value in each fuzzy rule table and the control variable output value are obtained according to the historical data and by using the kernel density estimation method to obtain the probability density function, and then the confidence interval is solved by setting the confidence on the probability density function, that is, the input variable threshold value and the control variable output value are obtained.

[0014] Further, the data-driven model predictive control algorithm is built according to the historical control variable and output variable data, specifically:

[0015] First, the historical control variable and output variable data are obtained and preprocessed: standardized processing and average filtering processing;

[0016] Then, according to the preprocessed data and by using the subspace identification method, the prediction model is obtained:

[0017]

[0018] Wherein, And represent the past and future system states, , ; And represent the control variable and the output variable in the past system state, And represent the control variable and the output variable in the future system state, , , , , And are the dimensions of the control and output, respectively, is a hyperparameter, N is the size of the data set and the hyperparameter k, i.e. ; and is the coefficient matrix of the subspace predictor, , ;

[0019] Further, the current number of real-time control changes is determined whether to meet the threshold change_th, if it meets, the control and output data of the latest control period are obtained to update the prediction model, otherwise the current prediction model is kept unchanged;

[0020] Finally, the model predictive controller is used to obtain the control quantity and applied to the industrial process.

[0021] Further, the model predictive controller is used to obtain the control quantity, specifically by solving the following optimization problem to obtain the control quantity:

[0022]

[0023]

[0024]

[0025]

[0026] wherein, is the prediction horizon, is the control horizon, is the time point within the prediction horizon and the control horizon, ; , , and is semi-positive definite, is positive definite, P and R are hyperparameters, representing the output weight and control weight, respectively, is the set control target, is the control change at time point , is the average value of the control change within the control horizon; and are the upper and lower limits of the control change; and are the upper and lower limits of the cumulative control change, represents the number of historical control changes considered, multiplied by the control period T is considered as the expected time lag of the industrial process control.

[0027] Further, the main control variable refers to a core control variable of the industrial process, and the secondary control variable refers to a control variable used to satisfy the constraint condition of the industrial process.

[0028] Further, if the controlled object of the industrial process is an industrial roasting furnace, the main control variable is temperature, and the input variable of the fuzzy rule includes a temperature deviation value, a first-order change trend of the temperature, a second-order change trend of the temperature, and a first-order change trend of the SO2 gas concentration.

[0029] The main control variable is temperature, and the input variable of the fuzzy rule includes a temperature deviation value, a first-order change trend of the temperature, a second-order change trend of the temperature, and a first-order change trend of the SO2 gas concentration. A core two-dimensional rule table TAB is constructed by using the temperature deviation value and the first-order change trend of the temperature. According to the comparison between the temperature deviation and the first-order change trend of the temperature and each threshold value, the position of the current state in the TAB is obtained, and a basic temperature control variable is obtained. According to the comparison between the second-order change trend of the temperature and the first-order change trend of the SO2 gas concentration and the threshold values thereof, the basic temperature control variable is corrected, and a temperature control variable output by the main rule table is obtained.

[0030] The secondary control variable includes the SO2 concentration of the tail gas, and the input variable of the fuzzy rule includes the SO2 concentration of the tail gas, the first-order change trend of the SO2 concentration of the tail gas, and the second-order change trend of the SO2 concentration of the tail gas. A tail gas rule table TAB2 is constructed by using the SO2 concentration of the tail gas and the first-order change trend of the SO2 concentration of the tail gas. According to the comparison between the tail gas deviation threshold value and the tail gas first-order trend threshold value and each threshold value, the position of the current state in the TAB is obtained, and a basic SO2 concentration control variable of the tail gas is obtained. According to the comparison between the second-order trend of the tail gas and the threshold values thereof, the basic SO2 concentration control variable of the tail gas is corrected, and a SO2 concentration control variable of the tail gas output by the secondary rule table is obtained.

[0031] Further, step 1 also establishes an auxiliary rule, including an event triggering rule, an increase / decrease overrun rule, and / or a material fluctuation correction rule. When the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, the auxiliary rule is also used to control the industrial process.

[0032] The event triggering rule is that if the first-order change trend of the temperature of the industrial process in the past several time periods is identified to be changed by more than a threshold value SJ_th, the control period is ignored, and the knowledge-driven fuzzy rule control algorithm is immediately used to control the industrial process.

[0033] The increase / decrease overrun rule is that the cumulative change of the control variable within a preset time length is limited.

[0034] The material fluctuation correction rule is that if the change of the feed amount is greater than a threshold value wuliao_th, a correction belt feed speed is applied to suppress the abnormal fluctuation of the material.

[0035] A knowledge and data fusion industrial process predictive control system includes an intelligent control algorithm construction module and an intelligent control module.

[0036] The intelligent control algorithm construction module is configured to determine the main control quantity, the secondary control quantity and the output quantity corresponding to each control quantity of the industrial process, and construct a corresponding fuzzy rule table by taking each control quantity and the corresponding output quantity as the output variable and the input variable of the fuzzy rule respectively; wherein the input variable threshold value and the control quantity output value in each fuzzy rule table are obtained by statistically distributing the historical data thereof.

[0037] The intelligent control algorithm construction module is further configured to build a data-driven model predictive control algorithm according to the historical control quantity and output quantity data.

[0038] The intelligent control module is configured to:

[0039] When the industrial process control system is put into use, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, that is, the historical output quantity data of each control quantity is obtained, and the corresponding control quantity is obtained based on the fuzzy rule table, and each control quantity obtained is applied to the industrial process.

[0040] When it is identified that the industrial process is in the intelligent control state for a preset time period, the data-driven model predictive control algorithm is used to control the industrial process; wherein the control state of the industrial process includes manual control and intelligent control, and the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and the data-driven model predictive control algorithm is used to control the industrial process, both of which belong to the intelligent control state.

[0041] When it is identified that the industrial process is in the manual control state for a control cycle, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process.

[0042] Beneficial effects

[0043] The present application provides a knowledge and data fusion industrial process predictive control method and system, in the first stage, that is, the "learning person" stage, the control behavior of the operator is simulated by the knowledge refined fuzzy rule, and high quality data is provided for the next stage. In the second stage, that is, the "surpassing person" stage, the system model is identified based on the data to fit the current running state, and the adaptive model predictive control algorithm is used to calculate the decision variable. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is the method framework diagram of the embodiment of the present application;

[0045] Figure 2 is the distribution histogram in the embodiment of the present application;

[0046] Figure 3 is the probability density distribution curve in the embodiment of the present application;

[0047] Figure 4 is the main rule table described in the embodiment of the present application;

[0048] Figure 5 is the model adaptive updating strategy described in the embodiment of the present application;

[0049] Figure 6 is the switch flowchart described in the embodiment of the present application;

[0050] Figure 7 is the running condition of the method described in the embodiment of the present application before commissioning (from 19:00 on February 5, 2023 to 8:00 on February 6), wherein (a) and (b) are the running conditions of the feed belt speed and the temperature, respectively;

[0051] Figure 8 is the running condition of the method described in the embodiment of the present application after commissioning (from 19:00 on February 10, 2023 to 8:00 on February 11), wherein (a) and (b) are the running conditions of the feed belt speed and the temperature, respectively. DETAILED DESCRIPTION

[0052] The embodiment of the present application is described in detail below. The embodiment is based on the technical solution of the present application and gives a detailed implementation manner and specific operation process, which further explains and describes the technical solution of the present application.

[0053] The embodiment provides a knowledge and data fusion industrial process predictive control method, as shown in Figure 1 In the first stage, i.e., the "learning man" stage, the rules are condensed according to the experience knowledge of the operator, the control behavior of the operator is imitated through fuzzy rule control learning, but the control period is changed from non-periodic low frequency to periodic high frequency. After experiencing the knowledge-driven control in the first stage, a sufficient amount of high-frequency high-quality input-output data will be generated, which can fully reflect the influence of the change of the manipulated variable on the controlled variable. Therefore, in the second stage, i.e., the "superior man" stage, the system model is identified based on the data set to fit the current running state, and the model predictive control algorithm is used to calculate the decision variable, and the predictive model is updated adaptively to adapt to the changing running conditions. The present application combines the advantages of knowledge-driven control and data-driven control methods, and realizes stable control with initial reliability and later adaptive ability.

[0054] The embodiment takes an industrial roasting furnace as an example of an industrial process controlled object to explain and describe the knowledge and data fusion industrial process predictive control method of the present application, wherein the main control variable is temperature and the secondary control variable is tail gas SO2 concentration, and the method specifically comprises the following steps.

[0055] I. Building an intelligent control algorithm

[0056] 1. Knowledge-driven fuzzy rule control algorithm

[0057] The main rule is a four-dimensional rule table, and the four dimensions are temperature deviation value, temperature first-order change trend, temperature second-order change trend, and SO2 gas concentration first-order change trend. The secondary rule is a three-dimensional rule table, and the three dimensions are tail gas SO2 concentration, tail gas SO2 concentration first-order change trend, and tail gas SO2 concentration second-order change trend. The auxiliary rules include event triggering rule, increase / decrease overrun rule, and material fluctuation correction rule.

[0058] (1) Main rule table: let temperature deviation thresholds be TD1_th, TD2_th, and TD3_th, let temperature first-order change trend thresholds be TR1_th, TR2_th, and TR3_th, let temperature second-order change trend threshold be TRR_th, let SO2 gas concentration first-order change trend threshold be SOR_th, let temperature change second-order trend correction amount be TRR_c, and let SO2 change first-order trend correction amount be SOR_c. Let the core two-dimensional rule table be TAB, then according to the comparison of temperature deviation and temperature first-order change trend with each threshold, the position of the current state in TAB is obtained, the basic control amount is obtained, and then according to the comparison of temperature second-order change trend and gas concentration first-order change trend with their thresholds, the control amount is further corrected, and the main rule fuzzy control amount is obtained.

[0059] (2) Secondary rule table: let tail gas deviation thresholds be weiqi_delta_th_1, weiqi_delta_th_2, weiqi_delta_th_3, and weiqi_delta_th_4, let tail gas first-order trend thresholds be weiqi_yijie_th_1, weiqi_yijie_th_2, and weiqi_yijie_th_3, let tail gas second-order trend thresholds be weiqi_erjie_th_1 and weiqi_erjie_th_2, let tail gas second-order trend correction amounts be weiqi_erjie_c_1 and weiqi_erjie_c_2. Let the tail gas rule table be TAB_SO2, then according to the comparison of tail gas deviation threshold and tail gas first-order trend threshold with each threshold, the position of the current state in TAB_SO2 is obtained, the basic secondary control amount is obtained, and then according to the comparison of tail gas second-order trend with its threshold, the corrected secondary rule fuzzy control amount is obtained.

[0060] (3) Auxiliary rules: for the event triggered rule, if the first order trend of temperature change in the past two time periods exceeds the threshold SJ_th, the rule is triggered, the control period is ignored, and control is immediately performed. For the increase / decrease overrun rule, considering the existence of time delay, the change of control quantity needs time to reflect on the temperature change. In order to prevent excessive increase / decrease from causing temperature fluctuations, the cumulative change of control quantity in a period of time is limited. For the material fluctuation correction rule, if the material blocking and material passing occurs, the feed rate will fluctuate sharply, which is different from normal control. If the feed rate change is greater than the threshold wuliao_th, a correction control quantity is applied to suppress the abnormal fluctuation of the material.

[0061] The time period in the event triggered rule of the embodiment is the same as the time period in the subsequent parameter setting based on the probability density distribution, and is the same as the control period in number (generally set to be the same, but can be set to be different according to the specific industrial process), which has no relationship on the time axis of the control process. For example, if the control period is set to be k minutes, the time period is generally equal to k minutes, which are two defined hyperparameters and will not change. In the event triggered rule, for example, the time period is set to be 2 minutes in the actual control process, and the current time is 16:40. Then the two time periods are 16:36-16:38 and 16:38-16:40 respectively, and the control quantity issuing time point can be 16:37. Therefore, the nodes before and after the time period have no relationship with the control quantity issuing time node. When the threshold is exceeded, even if the current time does not reach the control period (k minutes) from the last control quantity issuing time point, the control quantity is immediately issued.

[0062] (4) Parameter setting based on probability density distribution: The above-mentioned threshold values and the basic control quantity in the rule table are not subjective choices, but are obtained by analyzing the probability density distribution.

[0063] The implementation steps of setting the parameters in the embodiment are illustrated by taking the temperature first order change trend threshold as an example. First, a sufficient amount (a sufficient amount of data for kernel density estimation, such as 3 months) of historical data is selected, the temperature first order change trend of each time period is obtained, and the distribution is counted, as shown in Figure 2 According to the kernel density estimation method, the probability density function is obtained:

[0064]

[0065] In the formula, is the total number of sample points, h is a very small number tending to zero, is the kernel function. The confidence levels of 30%, 60% and 80% are selected, the confidence intervals are inversely solved, and the temperature change first order trend thresholds TR1_th=0.4, TR2_th=1.2 and TR3_th=1.8 are obtained.

[0066] Based on the above operation, the values of each threshold and basic control amount can be obtained, so as to determine the main rule table, the secondary rule table and each auxiliary rule. The main rule table is shown in Table 1. Figure 3

[0067] In addition, the correction amount of the rule table in the embodiment is determined according to expert knowledge.

[0068] 2. Data-driven model predictive control algorithm

[0069] (1) Data preprocessing: standardization processing and average filtering processing are performed on the collected data:

[0070]

[0071]

[0072] In the formula, is a one-dimensional variable, and respectively represent the mean and standard deviation of the one-dimensional variable of all samples, is the total number of samples, is the size of the sliding window, respectively represent the data obtained by standardization processing and average filtering processing.

[0073] (2) Adaptive model identification: since the operating conditions of the roasting furnace continuously change, a single fixed system model cannot maintain the fitting of the system. Moreover, the effectiveness of the model predictive control algorithm largely depends on the accuracy of the prediction model. Therefore, it is necessary to improve the fitting degree of the prediction model to the current system state through adaptive updating.

[0074] Define the past and future system states:

[0075]

[0076] In the formula, , , , , , are the dimensions of the control amount and the output amount respectively, N is obtained according to the size of the data set and the hyperparameter k, that is, .

[0077] According to the subspace identification method, the prediction model of the system is obtained:

[0078]

[0079] where the subspace estimator coefficient matrix ,​ .

[0080] Then, the model adaptive updating strategy is introduced. The accuracy of model identification depends on the quality of data used for identification, and high-quality data can better reflect the characteristics of the system. The quality of data is reflected in whether the data changes frequently, because only the changing data can reflect the mapping relationship between input and output data. The model updating strategy of the embodiment of the present application is shown in Figure 4 , the historical data and real-time data are input into the switcher, the prediction model is obtained through the update determination, and is transmitted to the MPC controller.

[0081] The detailed process of the switcher is shown in Figure 5 , and the basis for update determination is whether the number of control changes meets the threshold change_th requirement.

[0082] (3) Model predictive control: the main body of the MPC controller is to solve the following optimization problem:

[0083]

[0084]

[0085]

[0086] In the formula, is the prediction time domain, is the control time domain, generally , , , and P≥0 is semi-positive definite, R>0 is positive definite, is the set control target, and are the upper and lower limits of the control change. Since the roasting process has a time delay, and the time delay cannot be measured due to the influence of the feed components, an increase and decrease limit constraint condition is added to eliminate the influence of the time delay:

[0087]

[0088] In the formula, and are the upper and lower limits of the cumulative control change, represents the number of historical control changes considered, multiplied by the control period T is considered as the expected time delay.

[0089] II. Control of industrial processes using intelligent control algorithm

[0090] Referring to the flowchart shown in Figure 6 , the following steps are included:

[0091] (1) When the industrial process control system is put into operation, a knowledge-driven fuzzy rule control algorithm is used to control the industrial process, that is, historical output data of each control variable is obtained, and corresponding control variables are obtained based on a fuzzy rule table, and each control variable obtained is applied to the industrial process;

[0092] (2) When it is identified that the industrial process has been in the intelligent control state for the past N h , a data-driven model predictive control algorithm is used to control the industrial process;

[0093] Among them, the control state of the industrial process includes manual control and intelligent control, and the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and the data-driven model predictive control algorithm is used to control the industrial process, both of which belong to the intelligent control state;

[0094] (3) When it is identified that the industrial process has been in the manual control state for the past one control period T, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and step (1) is turned to.

[0095] A large-scale zinc smelting plant in Yunnan has a 109m 2 fluidized roasting furnace, which has been in operation for more than ten years, and the aging of the equipment has made temperature stable control more difficult. Before and after the application of the present application to the zinc roasting industrial site, the site operation is shown in Figure 7 and Figure 8 .

[0096] From the comparison of the control method before and after the operation, the roasting furnace temperature fluctuation is reduced from to . The running personnel clears the ash of the waste heat boiler every day, with an average duration of 2 hours, and the wind and material are greatly reduced during the ash cleaning, at which time the system is switched to manual control. Except for the manual control time during daily ash cleaning, the average automatic operation rate of the control system is as high as 98%. During the operation of the system, the proportion of roasting temperature deviation within 5℃ is 86.54%, and the proportion of deviation within 8℃ is 98.56%.

[0097] Therefore, the knowledge and data fusion industrial process predictive control method of the present application combines the advantages of knowledge-driven control and data-driven control methods, and realizes stable control with initial reliability and later self-adaptive ability. At the same time, the control scheme is suitable for other industrial field minute-level control application scenarios, and can be tested and applied in the future.

[0098] The above embodiments are the preferred embodiments of the present application, and those skilled in the art can make various modifications or improvements on the basis of the above embodiments without departing from the general concept of the present application, and these modifications or improvements should also belong to the scope of protection of the present application.

Claims

1. A knowledge and data fusion based industrial process predictive control method, characterized by, The method comprises the following steps: Step 1, determining the main control variables, the secondary control variables and the output variables corresponding to the control variables of the industrial process, taking each control variable and the corresponding output variable as the output variable and the input variable of the fuzzy rule respectively, and constructing a corresponding fuzzy rule table; wherein the input variable threshold setting and the control variable output value in each fuzzy rule table are obtained by statistically distributing the historical data thereof; Step 2, when the industrial process control system is put into use, a knowledge-driven fuzzy rule control algorithm is used to control the industrial process, that is, the historical output variable data of each control variable are obtained, and the corresponding control variable is obtained based on the fuzzy rule table, and each control variable obtained is applied to the industrial process; Step 3, when it is identified that the industrial process has been in the intelligent control state for a preset time period, a data-driven model predictive control algorithm is used to control the industrial process; Wherein the control state of the industrial process includes manual control and intelligent control, and the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and the data-driven model predictive control algorithm is used to control the industrial process, both of which belong to the intelligent control state; Step 4, when it is identified that the industrial process has been in the manual control state for a control period, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, and the process is transferred to step 3. The input variable threshold setting and the control variable output value in each fuzzy rule table are obtained by statistically distributing the historical data thereof and using the kernel density estimation method to obtain the probability density function, and then the confidence interval is inversely solved by setting the confidence on the probability density function, that is, the input variable threshold and the control variable output value are obtained.

2. The knowledge and data fusion based industrial process predictive control method according to claim 1, characterized in that, The data-driven model predictive control algorithm is built according to the historical control variable and output variable data, specifically:

3. The knowledge and data fusion based industrial process predictive control method of claim 1, wherein, Firstly, the historical control variable and output variable data are obtained and preprocessed: standardized processing and average filtering processing; Then, the prediction model is obtained by using the subspace identification method according to the preprocessed data: Then, it is judged whether the current real-time control variable change frequency meets the threshold change_th requirement, if yes, the control variable and output variable data of the latest control period are obtained to update the prediction model, otherwise the current prediction model remains unchanged; ； where and represent past and future system states, respectively, , ; and represent control and output quantities in past system states, respectively, and represent control and output quantities in future system states, respectively, , , , , and are dimensions of control and output quantities, respectively, is a hyperparameter, and N is determined according to the size of the data set and the hyperparameter k, i.e. ; and are coefficient matrices of the subspace predictor, , ; Finally, the control variable is obtained by using the model predictive controller and applied to the industrial process. The control variable is obtained by solving the following optimization problem:

4. The knowledge and data fusion based industrial process predictive control method of claim 3, wherein, The main control variable refers to the core control variable of the industrial process, and the secondary control variable refers to the control variable used to meet the constraint condition of the industrial process. ； ； ； ； wherein is the prediction horizon, is the control horizon, is a time point within the prediction horizon and the control horizon, ; , , and is semi-positive definite, is positive definite, P and R are hyperparameters, representing output weight and control weight, respectively, is the set control amount target, is the control change amount at time point ; is the average of control change amounts within the control horizon; and are the upper and lower limits of the control change amount; and are the upper and lower limits of the cumulative control change amount, denotes the number of periods of historical control change amounts considered, is multiplied by the control period T and is considered as the desired time lag of the industrial process control.

5. The knowledge and data fusion based industrial process predictive control method of claim 1, wherein, If the controlled object of the industrial process is an industrial roasting furnace, 6. The knowledge and data fusion based industrial process predictive control method of claim 1, wherein, ​ The main control quantity is temperature, and input variables of the fuzzy rule include temperature deviation value, temperature first-order change trend, temperature second-order change trend and SO2 gas concentration first-order change trend; a core two-dimensional rule table TAB is constructed by using the temperature deviation value and the temperature first-order change trend, a position of a current state in the TAB is obtained according to comparison of the temperature deviation and the temperature first-order change trend with each threshold value, and a basic temperature control quantity is obtained; the basic temperature control quantity is corrected according to comparison of the temperature second-order change trend and the gas concentration first-order change trend with their threshold values, and a temperature control quantity output by the main rule table is obtained; The secondary control quantity includes tail gas SO2 concentration, and input variables of the fuzzy rule include tail gas SO2 concentration, tail gas SO2 concentration first-order change trend and tail gas SO2 concentration second-order change trend; A tail gas rule table TAB2 is constructed by using the tail gas SO2 concentration and the tail gas SO2 concentration first-order change trend; A position of a current state in the TAB is obtained according to comparison of the tail gas deviation threshold value and the tail gas first-order trend threshold value with each threshold value, and a basic tail gas SO2 concentration control quantity is obtained; The basic tail gas SO2 concentration control quantity is corrected according to comparison of the tail gas second-order trend with its threshold value, and a tail gas SO2 concentration control quantity output by the secondary rule table is obtained.

7. The knowledge and data fusion based industrial process predictive control method of claim 6, wherein, The step 1 also establishes auxiliary rules, including event trigger rules, increase / decrease overrun rules and / or material fluctuation correction rules; when the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, the industrial process is also controlled based on the auxiliary rules; The event trigger rule is that if it is identified that the temperature first-order change trend of the industrial process in a plurality of time periods in the past all exceeds a threshold value SJ_th, the control period is ignored, and the knowledge-driven fuzzy rule control algorithm is immediately used to control the industrial process; The increase / decrease overrun rule is that the cumulative change amount of the control quantity in a preset time length is limited; The material fluctuation correction rule is that if the change in the feed amount is greater than a threshold value wuliao_th, a correction belt feed speed is applied to suppress abnormal material fluctuation.

8. A knowledge and data fusion industrial process predictive control system, characterized by, The intelligent control algorithm construction module and the intelligent control module are included; The intelligent control algorithm construction module is used to determine main control quantities, secondary control quantities and output quantities corresponding to the control quantities of the industrial process, and construct corresponding fuzzy rule tables by taking the control quantities and the corresponding output quantities as output variables and input variables of the fuzzy rules respectively; wherein, input variable threshold values in each fuzzy rule table and control quantity output values are obtained by statistically distributing historical data; The intelligent control algorithm construction module is also used to build a data-driven model predictive control algorithm according to historical control quantity and output quantity data; The intelligent control module is used to: When the industrial process control system is put into use, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process, that is, historical output quantity data of the control quantities are obtained, corresponding control quantities are obtained based on the fuzzy rule table, and the obtained control quantities are applied to the industrial process. When it is identified that the industrial process is in the intelligent control state for a preset time period in the past, a data-driven model predictive control algorithm is used to control the industrial process; wherein the control state of the industrial process includes manual control and intelligent control, and the control of the industrial process by the knowledge-driven fuzzy rule control algorithm and the control of the industrial process by the data-driven model predictive control algorithm both belong to the intelligent control state; When it is identified that the industrial process is in the manual control state for a control period in the past, the knowledge-driven fuzzy rule control algorithm is used to control the industrial process.

Images