Double-layer data driving model predictive control method of coal slurry concentration control system in coal gasification process

Abstract

The invention provides a double-layer data-driven predictive control method for coal slurry concentration control in a coal gasification process, and the method employs a double-layer control architecture: an upper layer dynamically updates model parameters based on historical operation data, and solves an optimization problem of coal / water-containing flow constraint and concentration fluctuation limitation through an open-loop simulation model; adjusting a flow set value in real time to actively inhibit concentration fluctuation caused by blockage of a feeding port, wherein an open-loop simulation model and a lower-layer control framework keep dynamic consistency; and the lower layer takes an incremental state space as a prediction model, implements a model prediction control strategy containing single-step input mapping, constructs a robust compensation mechanism by using a linear combination of sliding window historical data to reduce the influence of parameter uncertainty, and generates an optimal control sequence by solving a quadratic programming problem in real time. According to the method, the inhibition capability of the system on the time delay characteristic and the blockage disturbance is remarkably improved, and the problem that the coal slurry concentration fluctuates greatly due to blockage of the feeding port in the prior art is effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of coal gasification production process control, and particularly relates to a double-layer data-driven model predictive control method and system for a coal slurry concentration control system. BACKGROUND

[0002] Coal gasification is a key process for converting coal into syngas (mainly CO and H2), wherein coal slurry, as a carrier of gasification raw materials, has an important influence on process efficiency, energy consumption and equipment operation. First, coal needs to be crushed to a certain particle size to increase the specific surface area and promote the subsequent gasification reaction. Then, the coal powder is mixed with water to form a coal-water slurry. If the concentration of the coal-water slurry is too low, it will result in too much water, which will increase the evaporation energy consumption in the gasifier. If the concentration is too high, it will cause the viscosity to rise, resulting in pumping difficulties or pipe blockage. High-concentration coal slurry is transported by a high-pressure pump, atomized into fine droplets by a nozzle, and then enters the gasifier to undergo partial oxidation reaction at high temperature and high pressure to generate CO, H2 and CO2, etc. Finally, the final product is generated after gas cooling, washing and desulfurization treatment.

[0003] Coal slurry concentration is a main factor affecting the quality of syngas. In the process of coal slurry preparation, related research mainly focuses on two aspects: coal slurry concentration optimization and coal / water feeding device control.

[0004] There are many influencing factors for the coal slurry gasification device optimization coal blending technology, and it is difficult to achieve scientific and accurate optimization only by personal experience. The commonly used coal slurry concentration optimization model comprehensively considers the coal blending indicators, market price, inventory cost, cost of implementing coal blending operation, etc., and the optimization goal is to minimize the total coal cost. Particle swarm algorithm is used to solve the multi-objective coal blending optimization problem. At present, the coal slurry concentration optimization problem in the coal gasification process has achieved remarkable results. In order to avoid repeated research, the present application only considers the control of the coal feeding device / water feeding device.

[0005] The coal / water supply device control system is mainly composed of a coal bunker / water storage tank, a feeder / water pump motor, a baffle, a conveyor belt / transmission pipeline and a nuclear scale / flow meter. The coal / water flow can be changed by changing the rotating speed of the feeder / water pump motor. The feeding port of the coal supply device is often blocked due to uneven coal quality, and the feeding pipeline of the water supply device is often blocked due to scale or organic matter. The control system aims to ensure that the ratio of coal flow and water flow meets the coal slurry concentration requirement by adjusting the rotating speed of the feeder / water pump motor. At present, the factory determines the optimal coal / water flow based on production data in advance, and uses it as a fixed coal flow set value. The traditional control methods such as PID and fuzzy PID are used to control the feeder / water pump motor to make the ratio of coal flow and water flow meet the requirement. Under normal circumstances, this control scheme can maintain the coal slurry concentration at the set value. When the feeding port is seriously blocked, the on-site workers will use air cannons and chemical agents to promptly unblock the feeding port and the feeding pipeline. However, if the response is not timely, continuous blockage will cause the coal slurry concentration to deviate from the target value seriously, and excessive reliance on manual intervention also limits the automation level of the coal slurry concentration control system. Although MPC control has better control performance than PID in dealing with time-lag objects, the control performance will be reduced due to the existence of parameter uncertainty and model mismatch. SUMMARY

[0006] The application provides a double-layer data-driven model predictive control method for a coal slurry concentration control system to solve the problem that the ratio of coal flow and water flow (i.e. coal slurry concentration) deviates from the set value due to blockage of the feeding port.

[0007] One aspect of the application provides a double-layer data-driven model predictive control method for a coal slurry concentration control system, which comprises the following steps:

[0008] Real-time coal / water flow data and feeder / water pump motor rotating speed data of the coal slurry concentration control system are obtained, and the coal / water flow data includes the delay caused by the transportation of the conveyor belt / transmission pipeline;

[0009] An augmented state space model of the coal slurry concentration control system is constructed with the feeder / water pump motor rotating speed as the input and the coal / water flow as the output value, the time-lag relationship between the coal / water flow dynamic characteristics and the feeder / water pump motor rotating speed is described, and the model parameters include static gain, inertia time coefficient and fixed time lag;

[0010] The historical coal / water flow output value and the corresponding feeder / water pump motor rotating speed are recorded, and the system parameters are identified and updated. The parameter identification can use the least square method with forgetting factor, Moore-Penrose inverse method, etc.

[0011] A sliding window length is specified, and the history data of augmented state variables and system control variables are stored. The current time augmented state variable, the next time state variable and the control variable are expressed as the sum of the linear combination of the history data in the sliding window and the residual term. The combination coefficients and the residual term are optimization variables.

[0012] A double-layer control framework is designed. The upper optimization module updates the model parameters based on the history data, solves the optimization problem containing the maximum coal / water flow constraints and coal slurry concentration fluctuation constraints, and dynamically adjusts the coal / water flow set value. The lower control module adopts the input mapping model predictive control (IASS-IMMPC) based on the incremental form augmented state space, takes the control sequence, the combination coefficient and the residual term as the optimization variables, and generates the optimal feeder / water pump motor speed sequence by solving the quadratic programming problem.

[0013] The first speed optimization variable is applied to the system, the augmented state variable is updated, the sliding window is shifted, and the above process is continued.

[0014] In some optional embodiments, the augmented state variable χ i (t) contains the current output coal / water flow change value, the history output coal / water flow change value and the history control increment.

[0015] In some optional embodiments, the parameter identification adopts the Moore-Penrose generalized inverse, and the data used is the coal / water flow history data of a specified length and the corresponding feeder / water pump motor speed.

[0016] In some optional embodiments, the coal / water feeding device i with input mapping MPC is implemented by the following steps: defining the history state The input increment vector and the corresponding residual term δχ i (t), δu i (t) construct the linear combination relationship and wherein the combination coefficient and the residual term are to-be-optimized variables.

[0017] In some optional embodiments, the objective function of the quadratic programming problem is wherein the Hessian matrix M i is a positive definite matrix, and the optimization variable Γ i includes the residual term δu i (t), the combination coefficient ξ i and the speed sequence U i (t+1) after the next time.

[0018] In some optional embodiments, the upper coal / water flow set value optimization strategy includes:

[0019] The Moore-Penrose generalized inverse is used to update the system parameters based on historical data to determine the maximum flow output value allowed by the system;

[0020] The current coal / water flow trial value is set as the open-loop simulation structure setting value, wherein the simulation structure is consistent with the lower IASS-IMMPC, and is used to calculate the coal / water flow prediction value under the current trial value;

[0021] In the traversal solution, the trial value update step or update range is set until the coal slurry concentration prediction value of the coal slurry concentration control system (i.e., the ratio of the coal flow output value of the coal feeder to the water flow output value of the water feeder) is within the set fluctuation interval and the difference from the optimal setting value is minimized. If the prediction values do not meet the constraints, the current flow is set as the minimum value between the last time flow setting value and the current maximum flow output.

[0022] In some optional embodiments, the set coal slurry concentration fluctuation interval decays exponentially with respect to time, and by adjusting the interval size and decay rate, the feasibility of the upper optimization problem can be ensured, and the optimal setting value is the coal / water flow factory experience value.

[0023] One aspect of the embodiments of the present application provides a double-layer data-driven model predictive control system for a coal slurry concentration control system, which is used to implement the double-layer data-driven model predictive control method for the coal slurry concentration control system as described above, and the system comprises:

[0024] A data acquisition module that acquires and stores coal / water flow and corresponding feeder speed data in real time;

[0025] A model construction module that generates an augmented state space model and updates parameters online using the Moore-Penrose generalized inverse;

[0026] An upper optimization module that dynamically adjusts the coal / water flow setting value;

[0027] A lower control module that combines single-step input mapping, incremental form augmented state quantity, and MPC to further reduce the influence of system parameter uncertainty and ensure consistency with the upper open-loop simulation results as much as possible;

[0028] An execution module that sends the optimized setting value and control instruction to the feeder. The optimized control quantity calculated by the lower control module is the speed setting value of the feeder / water pump motor, which can be quickly tracked to the setting value by the incremental PID control of the feeder / water pump motor speed.

[0029] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure.

[0030] The coal slurry concentration control system of the present application has the following beneficial effects:

[0031] (1) The single-step input mapping MPC control strategy based on the incremental form of the augmented state space is designed, which accelerates the system adjustment time and improves the multi-step prediction accuracy under model uncertainty;

[0032] (2) A double-layer framework is introduced, which can maintain relatively stable coal slurry concentration in the presence of continuous slow blockage and sudden major blockage. BRIEF DESCRIPTION OF DRAWINGS

[0033] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments with reference to the attached drawings.

[0034] Figure 1 is a flow chart of the double-layer data-driven model predictive control method of the coal slurry concentration control system of the coal gasification process according to an embodiment of the present application;

[0035] Figure 2 is a schematic diagram of the coal slurry concentration control system according to an embodiment of the present application, and a complete system can be composed of one or more coal feeding devices and one water feeding device. As shown in the figure, the coal feeding device is composed of a coal bunker, a baffle, a disc feeder, a variable speed motor, a transmission belt and a nuclear scale, the variable speed motor is a control mechanism, and the nuclear scale is a sensing mechanism; the water feeding device is composed of a water storage tank, a water pump motor, a transmission pipeline and a flowmeter, the water pump motor is a control mechanism, and the flowmeter is a sensing mechanism;

[0036] Figure 3 is a schematic diagram of the double-layer structure according to an embodiment of the present application, taking two coal feeding devices and one water feeding device as an example;

[0037] Figure 4 is a flow chart of the overall algorithm according to an embodiment of the present application;

[0038] Figure 5 and Figure 6 is a simulation result diagram according to an embodiment of the present application, wherein Figure 5 is a schematic diagram of the flow rate ratio of each device when the feeding port / feeding pipeline is suddenly blocked to a large extent, Figure 6 is a schematic diagram of the flow rate ratio of each device when the feeding port / feeding pipeline is continuously blocked. DETAILED DESCRIPTION

[0039] The coal gasification process coal slurry concentration control system double-layer data-driven model predictive control method provided by the application can realize accurate tracking of the set value of the coal / water flow by constructing an incremental form of an augmented state space model, combining a single-step input mapping method with MPC, and in the presence of system parameter uncertainty. Meanwhile, the method designs a double-layer structure, combines updated system parameters to solve the upper-layer optimization problem to recalculate the coal / water flow set value, and can maintain the stability of the coal slurry concentration in the presence of blockage. The simulation results show the effectiveness of the application.

[0040] As shown in Figure 1 The specific implementation steps of the coal gasification process coal slurry concentration control system double-layer data-driven model predictive control method provided by the application are as follows:

[0041] P100, obtaining real-time coal / water flow data and feeder / water pump motor speed data of the coal slurry concentration control system refers to collecting the coal / water flow (unit: t / h) and feeder / water pump motor speed (unit: RPM) data in time sequence in the coal blending process, wherein the flow ratio on each device directly reflects the coal slurry concentration. In actual application, the corresponding data can be obtained through a nuclear scale, a flow meter sensing device and a photoelectric encoder.

[0042] P200, taking the feeder / water pump motor speed increment as the input and the coal / water flow as the output, taking the current coal / water flow change and the historical control change as the state quantity, and constructing an incremental form of an augmented state space model. In the application, the change quantity instead of the absolute quantity is selected to constitute the state quantity in order to reduce the influence of parameter uncertainty and further improve the accuracy of the predicted output; the augmented state space model is selected in order to implicitly include the time delay factor of the system into the state quantity, facilitating model derivation and controller design.

[0043] P300, the least square method containing a forgetting factor or the Moore-Penrose generalized inverse method is selected to complete system parameter identification. The control system parameters targeted by the application are simple, and it is recommended to preferentially use the Moore-Penrose generalized inverse.

[0044] P400, the length of the sliding window is specified, and the history data of the augmented state variables and system control variables are stored. According to the linear system characteristics, the current time and the next time augmented state variables, the next time control variables are expressed as the linear combination of the sliding window history data and the residual term, and the influence range of the uncertain parameters is reduced from the original state variables to the residual of the state variables, wherein the combination coefficients and the residual term are solved as optimization variables in the online optimization. The present application adopts a single-step input mapping method, and only the next-step predicted output is expressed in the form of the above linear combination. Although the compensation ability is gradually weakened in multi-step prediction, the prediction model combined with the incremental form augmented state space still has high prediction accuracy. Considering the calculation complexity, the present application does not adopt a multi-step input mapping method.

[0045] P500, a double-layer control problem is designed, a coal slurry concentration fluctuation constraint is set, and the current maximum allowable coal / water flow output is solved according to the updated model parameters. The upper layer solves the optimization problem based on the maximum flow constraint and the coal slurry concentration fluctuation constraint to dynamically adjust the coal flow set value, and the lower layer solves the single-step input mapping MPC optimization problem based on the optimized set value to obtain the optimized speed sequence. In the present application, the cost function of the upper layer optimization is defined as where Y * is the optimal coal slurry flow, which can be set as the empirical value of the plant coal slurry flow, Y r is the coal slurry flow set value, which is an optimization variable, and the coal / water flow set value Y i,r of each device can be determined through the flow ratio of each device (i.e., the coal slurry concentration, wherein the ratio of the coal flow of each coal feeding device reflects the proportion of different coals in the coal slurry); min r max where Y min is the minimum coal slurry flow requirement determined according to the plant experience, Y max is the sum of the maximum output flow of each device; 2) represents the predicted flow output of the i-th device at the future k-th step under the set value Y i,r , and the flow output prediction model is the aforementioned augmented state space model. The prediction control variable is obtained by solving the IASS-IMMPC; 3) where p max is defined as p max (k) = θ k p0+(1-θ k )p ∞ , θ ∈ (0, 1), p0, p ∞ > 0. η il ​​The flow ratio of device i and device l is taken as an example, and the flow ratio of two coal feeders and one water feeder is set to 3:2:1 when the coal slurry concentration is set to 50% and the proportions of the two coals are 1 / 3 and 1 / 6. In practical applications, the value of p0 should be large enough to ensure that the initial flow ratio meets the constraints.

[0046] P600, the first control quantity of the optimization control sequence is applied to the system, the augmented state quantity is updated, the sliding window is translated, and the window data information is updated. The optimization value output by the lower MPC includes the control increment residual term, the augmented state residual term, the linear combination coefficient, and the optimization control increment sequence after the next time. The control increment at the next time is represented as the historical data of the control quantity in the sliding window multiplied by the linear combination coefficient plus the control increment residual term.

[0047] In combination Figure 3 and Figure 4 The algorithm flow of a double-layer data-driven model predictive control method for a coal gasification process coal slurry concentration control system is shown, which includes the following steps:

[0048] S1: input the time series data in the coal gasification water-coal slurry production process to form a sliding window where χ i (t-k) represents the augmented system state data of the i-th device at the k-th production time in the past, Δy i (t-k) represents the system output increment at the k-th production time in the past, Δu i (t-k-j) represents the control increment at the k+j-th production time in the past, n L represents the inherent time delay of the system. Set the positive definite symmetric weight matrices Q, R, R1, and the coefficient r0>0.

[0049] S2: construct the augmented state space model χ i (t+1) = A i χ i (t) + B i Δu i (t), set the prediction horizon N c and the control horizon N u Solve IASS-IMMPC to obtain the set value of the feeder / water pump motor speed. The matrices are defined as:

[0050]

[0051] S3: according to the set value solved in S2, adjust the PWM duty cycle to control the motor speed through the incremental PID control algorithm to realize set value tracking.

[0052] S4: record h LLength-filtered system output data [y i (t),…,y i (t-h L +1)] and input data [Δu i (t-n L ),…,Δu i (t-h L -n L -1)] are updated system parameters by Moore-Penrose generalized inverse:

[0053]

[0054] S5: Solve the current allowed maximum coal / water flow based on the updated parameters Where u i,max is the upper limit of the motor speed of the i-th device. If y i,max changes, then under the factory minimum coal slurry flow Y min , the optimal coal slurry flow Y * , the traversal step size δy and the coal slurry concentration fluctuation constraint , the following optimization problem is solved by traversal method:

[0055]

[0056] s.t.Y min ≤Y r ≤Y max

[0057]

[0058] Where Y r of the k-th traversal is Y min +k×δy, and u i is obtained by solving the IASS-IMMPC problem in step S2 under the current Y setting. If no traversal value satisfies the above conditions, then set the coal / water flow to min{current maximum flow output, last time flow setting value}; otherwise, the optimized Y r is taken as the setting value of the lower IASS-IMMPC.

[0059] S6: Update the current time state and input to the sliding window, and move the sliding window forward by 1 step. Repeat steps S2 to S5. If the current setting value is less than the factory minimum coal slurry flow, stop solving and shut down for maintenance.

[0060] The IASS-IMMPC problem solved in step S2 is:

[0061]

[0062] s.t.ui,min ≤u i (t+k|t)≤u i,max ,k=0,1,...,N u -1

[0063] u i (t+k+1|t)=u i (t+k|t)+△u i (t+k+1|t),k=0,1,...,N u -1

[0064] △u i (t|t)=△u i,h (t-1)ξ i +δu i (t)

[0065]

[0066] in, U i =[Δu i (t+1|t),…,Δu i (t+N c -1|t)] T ,u i,min and u i,max are the input lower and upper limits of the i-th device, The upper bound of the input increment for the i-th device. U i ,ξ i and δu i (t) is the variable to be optimized. The above problem can be solved by quadratic programming.

[0067] Figure 5 、 Figure 6 This is a simulation example of the proposed method. Figure 5 In case of sudden congestion, Figure 6 It can be seen that the proposed method has the ability to adjust in both cases and can maintain the flow ratio of each device relatively stable, thus ensuring the relative stability of the water-coal slurry concentration.

[0068] The above content merely expresses the specific implementation methods of this application and does not limit the scope of protection of this application. For ordinary technicians in the technical field to which this application belongs, they can make several simple deductions or substitutions without departing from the concept of this application, which should be considered to fall within the scope of protection of this application.

Claims

1. A two-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process, characterized in that: The method comprises the following steps: Acquire real-time coal / water flow data and feeder / water pump motor speed data of the coal slurry concentration control system, wherein the flow data includes a delay measurement value caused by conveyor belt / transmission pipeline transportation; An augmented state-space model of the coal blending system is constructed using the feeder / water pump motor speed as input and the coal / water flow rate as output. This model describes the time-lag relationship between the dynamic characteristics of the coal / water flow rate and the feeder / water pump motor speed. The model parameters include static gain, inertia time coefficient, and fixed time lag. Based on the historical coal / water flow output values ​​and the corresponding feeder / water pump motor speeds, the system parameters are updated using the least squares method with forgetting factor or the Moore-Penrose inverse method; Set the sliding window to store the historical data of augmented state and control variables, and express the augmented state at the current moment, the state at the next moment, and the control variable as the sum of the linear combination of the historical data in the sliding window and the residual term. The combination coefficient and the residual term are the optimization variables. A two-layer control framework was designed: an upper-layer optimization module updates model parameters based on historical data, solves an optimization problem involving maximum coal / water flow constraints and coal slurry concentration fluctuation constraints, and dynamically adjusts the coal / water flow setpoints. The lower-layer control module employs input-mapping model predictive control (IASS-IMMPC) based on incremental augmented state space, using the control sequence, combination coefficients, and residual terms as optimization variables. It generates the optimal feeder / water pump motor speed sequence by solving a quadratic programming problem. Apply the first speed optimization variable to the system, update the augmented state variable and translate the sliding window, and then execute the above process repeatedly.

2. The double-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process according to claim 1 is characterized in that: The augmented state quantity χ i (t) includes the current output coal / water flow change value, the historical output coal / water flow change value and the historical control increment.

3. The double-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process according to claim 1 is characterized in that: The parameter identification is performed using Moore-Penrose generalized inverse, and the data length is user-defined historical coal / water flow and corresponding feeder / water pump motor speed data.

4. The double-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process according to claim 1 is characterized in that: The MPC of coal feeding device / water feeding device i with input mapping is implemented by the following steps: define the historical state in the sliding window Input increment vector And the corresponding residual term δχ i (t),δu i (t) Constructing a linear combination relationship The combination coefficient and residual term are the variables to be optimized.

5. The double-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process according to claim 4 is characterized in that: The objective function of the quadratic programming problem is Where the Hessian matrix M i is a positive definite matrix, and the optimization variable Γ i Including the residual term δχ i (t),δu i (t), combination coefficient ξ i and the speed sequence U after the next moment i (t+1).

6. The double-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process according to claim 1, characterized in that: The optimization strategy for the upper layer coal / water flow setpoint is: Based on historical data, the Moore-Penrose generalized inverse method is used to update system parameters to determine the maximum flow output of the current coal feeder / water feeder. The current flow rate trial value is used as the set value of the open-loop simulation structure, where the simulation structure is consistent with the lower layer IASS-IMMPC, and is used to calculate the coal flow rate prediction value under the current trial value; During the ergodic solution, the trial value update step size or update amplitude is set until the coal slurry concentration control system's predicted value (i.e., the ratio of the coal flow output value of the coal feeder to the water flow output value of the water feeder) falls within the set fluctuation range and minimizes the difference from the optimal set value. If none of the predicted values ​​meet the constraints, the current flow rate is set to the minimum between the previous flow set value and the current maximum flow output.

7. The double-layer data-driven model predictive control method for a coal slurry concentration control system in a coal gasification process according to claim 6, characterized in that: The set coal slurry concentration fluctuation range decays at an exponential rate with respect to time. The feasibility of the upper-level optimization problem can be ensured by adjusting the interval size and decay rate. The optimal setting value is the coal / water flow plant experience value.

8. A two-layer data-driven model predictive control system for a coal slurry concentration control system in a coal gasification process, used to implement the method according to any one of claims 1 to 7, characterized in that: The system comprises: Data acquisition module, which acquires and stores coal / water flow and corresponding feeder speed data in real time; Model building module, which generates the augmented state space model and uses Moore-Penrose generalized inverse to update parameters online; The upper optimization module dynamically adjusts the coal / water flow setpoints; The lower-level control module combines single-step input mapping, incremental augmented state quantity, and MPC to further reduce the impact of system parameter uncertainty and ensure consistency with the upper-level open-loop simulation results as much as possible; The execution module sends the optimized setting values ​​and control instructions to the feeder; Among them, the optimized control quantity calculated by the lower control module is the speed setting value of the feeder / water pump motor, and the speed of the feeder / water pump motor can be quickly tracked to the set value by incremental PID control.

Images