Dynamic optimization method of CHP unit in variable load process based on SQP-IPM alternating solution
Through the SQP-IPM alternating solution method and Legendre polynomial discrete format, combined with the convergence depth control strategy, the precise modeling and rapid simulation problems of dynamic process control of variable load of CHP units are solved, and rapid convergence and efficient optimization are achieved.
Patent Information
- Application Number
- CN202210870755.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-22
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2042-07-22
AI Technical Summary
It is difficult to accurately model and quickly simulate and calculate dynamic process control of variable loads of CHP units. The existing optimization algorithms have problems such as excessive iteration, long calculation time, and degradation of controller performance during the solution process.
Using the method based on SQP-IPM alternating solution, the convergence depth control strategy is designed to improve the real-time and fast convergence of the method through the Legendre polynomial discrete format.
The rapid convergence and optimization calculation of the dynamic process of variable load of CHP units is realized, which significantly reduces the optimization calculation time and improves the calculation efficiency and convergence of the algorithm.
Smart Images

Figure CN115268381B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of CHP unit load-variable dynamic process control, and in particular to a CHP unit load-variable process dynamic optimization method based on SQP-IPM alternating solution. Background Art
[0002] The dynamic regulation characteristics of the modified combined heat and power (CHP) units are complex, showing the characteristics of multi-variable, multi-process coupling, multi-parameter superposition and strong nonlinearity. Differential Algebraic Equations (DAEs) are often used to describe their dynamic characteristics in the operating range, which often makes it difficult to accurately model and quickly simulate the variable load response capability of the unit. The dynamic process control optimization problem with DAEs model constraints of CHP units is often solved by the joint method. Although this method has advantages in dealing with complex constraint problems, its practice of discretizing state variables and control variables at the same time often doubles the scale of the nonlinear programming (NLP) proposition after discretization, resulting in increased difficulty in optimization calculation and increased solution time.
[0003] Existing research on NLP problem solving is mainly in the following aspects: For large-scale optimization problems with a large number of equality constraints and limited degrees of freedom, a simplified space method can be used to reduce the dimension of the sequential quadratic programming (SQP) subproblem through a space decomposition strategy to improve the algorithm's solution efficiency. However, the space decomposition process will increase the computational cost. For problems with inequality constraints, the interior point method (IPM) is often used. IPM converts inequality constraints into obstacle terms and adds them to the objective function. The original problem solution is converted into a series of equality-constrained obstacle subproblems to avoid the determination of the active set in the solution process. Although the dependence on the initial point is not strong, when solving large-scale complex process optimization problems where the initial point is far from the optimal point, it cannot take into account both convergence and convergence speed.
[0004] At the same time, the convergence criterion has a great influence on the calculation time, and calculation delay will lead to the degradation of controller performance.
[0005] Optimization algorithms often use traditional termination criteria to perform full or partial convergence control. Traditional criteria can only give rigid conclusions of "convergence" or "non-convergence". When considering the degree of change of variables, the degree of reduction (or increase) of the objective function, etc., the optimization calculation index often enters a stage where the calculation cost is far greater than the improvement of the convergence effect after reaching a certain convergence level. When solving dynamic optimization problems, nonlinear optimization algorithms based on traditional criteria often consume a lot of computing time due to excessive iterations, making it impossible for the control effect to act on the controlled system in a timely manner. Summary of the invention
[0006] To solve the above technical problems, the present invention provides a dynamic optimization method for the variable load process of a CHP unit based on the SQP-IPM alternating solution, which can effectively solve the optimal control problem of the variable load dynamic process of the CHP unit, and the continuous control process can converge quickly.
[0007] The technical solution adopted by the present invention is:
[0008] The dynamic optimization method of CHP unit variable load process based on SQP-IPM alternating solution includes the following steps:
[0009] Step 1: Formulate the variable load optimization control problem of the CHP unit of the power plant into a proposition;
[0010] Step 2: Solve the dynamic optimization problem in step 1, and use the collocation method based on Legendre polynomial discretization format to discretize the state variables and control variables at the same time, forming a nonlinear programming NLP problem;
[0011] Step 3: Design an SQP-IPM-based alternating solution method to solve NLP problems, and introduce a convergence depth control strategy into the alternating solution method to improve the real-time performance of the method.
[0012] In step 1, the form of the dynamic optimization proposition is:
[0013]
[0014] In formula (1): min J(x(t 0 ),t 0 ,x(t f ),t f ) represents the objective function of minimizing the dynamic optimization proposition with respect to the state variables; t 0 ,t f are the starting and ending time points of the unit optimization control respectively; t represents the time variable in the control time domain;
[0015] x(t), y(t), u(t) represent the differential state variable, algebraic output variable, and control variable of the CHP unit at time t, respectively; F(x(t), y(t), u(t), t) represents the relationship function model between the state variable, output variable, and control variable;
[0016] g E , g IE are equality and inequality path constraints; h E 、h IE For the time t f The terminal value constraint of
[0017] x(t 0 ), y(t 0 )、x(t f ), y(t f ) represent the initial value of the state variable, the initial value of the output variable, the final value of the state variable, and the final value of the output variable respectively;
[0018] The optimal control performance objective function is described as:
[0019]
[0020] In formula (2): t 0 and t f are the starting and ending time points of the unit optimization control respectively; u is the vector composed of the control variables of the CHP unit; y is the vector composed of the algebraic output variables of the CHP unit, y * is the set value of the output variable obtained by steady-state optimization; R and S are positive definite weight matrices, which can determine the initial value by normalizing the variables; J is the objective function, u(t) is the control variable value at time t, u(t-1) is the control variable value at time t-1; y(t) is the output variable value at time t.
[0021] In the variable load optimization control problem of the cogeneration unit, u is the control variable, y is the output variable, and x is the state variable, where:
[0022]
[0023] In formula (3), x is the state variable, corresponding to ψ d is the drum pressure, P H is the generating power of the unit, ψ z Extraction steam pressure for steam turbine heating, V f is the boiler combustion rate;
[0024] u is the control variable, corresponding to V T is the opening of the steam inlet regulating valve of the high-pressure cylinder of the steam turbine, V B is the fuel quantity of the unit, V H Adjust the butterfly valve opening for heating extraction steam;
[0025] y is the output variable, corresponding to ψ t is the pressure before the turbine, P H is the generating power of the unit, m H is the extraction steam flow for heating.
[0026] In the step 1,
[0027] ①. The mechanism model of CHP unit is as follows:
[0028] V m (t) = V B (tt B ) (4);
[0029]
[0030] Formula (4) to (5) represent the dynamic relationship of coal pulverizing system, V B (t) is the coal mass flow rate of the unit, V m (t) is the actual coal mass flow rate entering the coal mill in the pulverizing system, t B is the delay time constant of the milling process, V f (t) is the boiler combustion rate, T f is the milling inertia time constant, t represents the control process time domain value, V B (tt B ) means tt B The unit coal mass flow rate at the moment;
[0031]
[0032] ψ d (t)-ψ t (t) = K 2 (K 1 V f (t)) 1.5 (7);
[0033] Equations (6) to (7) represent the dynamic relationship of the drum boiler system, V T (t) is the opening of the turbine regulating valve, ψ d (t) is the drum pressure, ψ t (t) is the pressure before the turbine, C d is the boiler heat storage coefficient, δ is a constant coefficient, K 1 is the fuel gain static parameter, K 2 is the superheater resistance coefficient, K 3 is the static parameter of turbine gain;
[0034]
[0035] ψ 1 (t)=0.01ψ t (t)V T (t) (9);
[0036]
[0037] Equations (8) to (10) represent the dynamic relationship of the steam turbine system, V H (t) is the opening of the extraction steam regulating butterfly valve, P H (t) is the generating power of the unit, ψ z (t) is the exhaust pressure of the intermediate pressure cylinder, ψ 1 (t) is the first stage pressure of the steam turbine, m r (t) is the mass flow rate of circulating water, ε r (t) is the circulating water return temperature, T t is the turbine inertia time constant, C z is the heat storage coefficient of the heating network heater, K 4 K is the proportion of high and medium pressure cylinders in the turbine work, 5 is the static parameter of low pressure cylinder gain, K 6 is the effective specific heat capacity of the circulating water in the heat network;
[0038] m H (t) = K 7 K 6 m r (t)(96ψ z (t)-ε r (t)+103) (11);
[0039] Formula (11) represents the dynamic relationship of the heating system, m H (t) is the heating extraction steam flow rate; K 7 It is the steam flow coefficient converted from the effective heat of extraction steam for heating.
[0040] ②. CHP unit control model:
[0041] Design a proportional-integral control algorithm to coordinately adjust V B (t), V T (t), V H (t) direction and magnitude, so that the output variable ψ t (t), m H (t), P H (t) changes, and its mathematical model is as follows:
[0042]
[0043] Among them, K PT , K IT Control VT (t) Proportional coefficient and integral coefficient of the variable; K PB , K IB , respectively control V B (t) Proportional coefficient and integral coefficient of the variable; K PH , K IH , respectively control V H (t) Proportional coefficient and integral coefficient of variables; SP ψ , SP P , SP m are the set values of main steam, unit power generation and heating extraction steam flow respectively; ψ t (t) is the current value of the main steam pressure control process, P H (t) is the current value of the unit power control process, m H (t) is the current value of the heating extraction steam flow control process.
[0044] Considering the safe and stable operation of the CHP unit, the fluctuation range of the process value of the output variable and the error range of the steady-state value during the control process cannot exceed the allowable limit and should meet the following constraints:
[0045]
[0046]
[0047] Among them, △ ψ ,△ P ,△ m are the fluctuation ranges of main steam pressure, electric power and extraction steam flow respectively; δ ψ , δ P , δ m are the error ranges of main steam pressure, electric power and extraction steam flow respectively; within the optimization control period, ψ t (t f ), P H (t f )、m H (t f ) are the end time t f Final values of main steam pressure, electrical power, and extraction steam flow.
[0048] ③. Boundary constraints:
[0049] CHP unit in [t 0 ,t f ] The variable load regulation is completed within the period, and the initial and final states of each control variable and output variable are as follows:
[0050]
[0051]
[0052] Where, t 0 and t f are the starting and ending time points of unit optimization control respectively;
[0053] are the starting time t 0 Output: the pressure before the steam turbine, the generating capacity of the unit, and the value of the extraction steam flow for heating;
[0054] The termination time t f Output: the pressure before the steam turbine, the generating capacity of the unit, and the value of the extraction steam flow for heating;
[0055] are the starting time t 0 Control the values of the opening of the high-pressure cylinder steam inlet regulating valve, the unit fuel quantity, and the opening of the extraction steam regulating butterfly valve;
[0056] The termination time t f The controlled quantities include the opening of the high-pressure cylinder steam inlet regulating valve, the fuel quantity of the unit, and the opening of the steam extraction regulating butterfly valve.
[0057] ④. Path constraints:
[0058] In the process of variable load regulation, the control variable V T (t), V B (t), V H (t) The following constraints need to be met:
[0059]
[0060] In the formula, and are the minimum and maximum values of the turbine valve opening change respectively; and are the minimum and maximum values of the unit fuel quantity change, respectively; and are the minimum and maximum values of the opening change of the extraction steam regulating butterfly valve respectively. The step 2 comprises the following steps:
[0061] 1) Time domain transformation:
[0062] The optimal control time domain t∈[t 0 ,t f ] is divided into p segments of finite elements, that is, 0 = t 0 <t 1 <…t q-1 <t q <…t p =tf , it can be seen that the step size of each time domain is:
[0063] △t=t q -t q-1 ,q∈[1,p] (18);
[0064] Where: △t is the step size of each time domain, q is the number of finite elements, p is the total number of finite element segments, t q is the time variable of the qth finite element, t q-1 is the time variable of the q-1th finite element;
[0065] In the Gaussian pseudospectral method, the configuration point interval is selected as τ∈[-1,1], so the time variable on the qth segment is converted into the variable τ through time domain transformation, and the transformation is as follows:
[0066]
[0067] Where: τ is the variable converted by time domain transformation on the qth segment time variable;
[0068] Then the description of the optimal control problem in each time zone is as follows:
[0069]
[0070] In formula (20): J (q) is the objective function in the qth time zone, τ 0 is the initial configuration point of the time domain transformation of the time variable on the qth segment, τ f The terminal configuration points of the time-variable time-domain transformation on the qth segment;
[0071] x (q) (τ) is the differential state variable of the τth configuration point in the qth time period, x (q) (τ 0 ) is the τth time segment in the qth time segment 0 The differential state variables of the collocation points, x (q) (τ f ) is the τth time segment in the qth time segment f Differential state variables of configuration points;
[0072] y (q) (τ) is the algebraic output variable of the τth configuration point in the qth time period, y (q) (τ 0 ) is the τth time segment in the qth time segment 0 The algebraic output variable of the collocation point, y (q) (τ f ) is the τth time segment in the qth time segment f Algebraic output variables of collocation points;
[0073] u (q) (τ) is the control variable of the τth configuration point in the qth time segment, F is the DAEs dynamic model, and △t is the step size of each time domain.
[0074] 2) Variable discretization:
[0075] The Gaussian pseudospectral method discretizes the control variables and state output variables at the Legendre-Gauss (LG) configuration points at the same time, and then approximates the control variables and state variables by constructing Lagrange interpolation functions. 0 ,τ f ], the N-order Legendre polynomial is:
[0076]
[0077] Where: P N (τ) is the Nth order Legendre polynomial, N is the order, Find the Nth order derivative of the function with respect to τ;
[0078] With the zero point and τ of formula (21) 0 Composed of N+1 configuration points, at each t (q) In the time domain, the differential state variables and control variables are approximately expressed by polynomials as follows:
[0079]
[0080] Where: x (q) (τ) is the differential state variable of the τth configuration point in the qth time period, is the interpolation function of the state variable, X (q) (τ i ) is the value of the state variable at the i-th configuration point of the q-th finite element;
[0081] u (q) (τ) is the control variable of the τth configuration point in the qth time period, is the interpolation function of the control variable, U (q) (τ i ) is the value of the control variable at the i-th configuration point of the q-th finite element.
[0082] In formula (22), the Lagrange interpolation basis function is:
[0083]
[0084] Where: N is the interpolation order; is the interpolation function of the state variables, is the interpolation function of the control variable, The independent variable of the interpolation function at the j-th collocation point, τ i is the interpolation function independent variable on the i-th configuration point; τ is the interpolation function independent variable of the time variable on the q-th segment transformed by time domain transformation;
[0085] By taking the derivative of the state variable from equation (22), the interpolation function can be approximated to the differential form of the state variable, and we can obtain k The state variable derivative of the point can be approximately expressed as:
[0086]
[0087] in: is the derivative of the differential state variable at the kth collocation point of the pth segment finite element, is the interpolation function of the state variable at the kth collocation point of the pth segment finite element, X (q) (τ i ) is the state variable at the i-th configuration point of the q-th finite element;
[0088] Combining equation (20) and equation (24), we can get the differential algebraic equality constraint:
[0089]
[0090] Where: △t is the step size of each time domain; F is the DAEs dynamic model, X(τ k ) is the τth k The state variables at the configuration points, M(τ k )τ k Other variables at the configuration point, τ k is the kth collocation point of the time-domain transformation of the time variable on the qth segment;
[0091] In the process of variable load control of the unit, the continuity of the change trajectory of each state output must be ensured, so the state variable must also change continuously, that is, the state variable at the terminal moment of each segment is used as the state variable at the initial moment of the next segment, and the constraint equation at the connection is:
[0092] X (q-1) (τ f )=X (q) (τ 0 ) (26);
[0093] Where: X (q-1) (τ f ) is the q-1th finite element τ f The state variables on the configuration points, X (q) (τ 0 ) is the qth finite element τ 0 State variables on configuration points;
[0094] Since the LG configuration point does not contain the terminal state value, the terminal state variable is calculated according to the Legendre-Gauss quadrature formula from equation (26):
[0095]
[0096] In the formula, A k and f(τ k ) are Gaussian integral weight coefficient and quadrature function respectively.
[0097] Through the above discretization process, the optimal control problem of the CHP unit's variable load dynamic process can be transformed into a nonlinear programming problem for solution.
[0098] In step 3, the convergence depth control strategy is as follows:
[0099] The nonlinear programming problem shown in the original proposition (20) can be simplified as:
[0100]
[0101] in, is a scalar objective function, x∈R n is an n-dimensional real vector, c∈R m is an m-dimensional equation.
[0102] The solution to problem (28) must satisfy the following first-order optimality conditions:
[0103]
[0104] in, is the gradient vector of the objective function, is the Jacobian matrix of the constraint equation. λ is the Lagrange multiplier corresponding to the constraint equation, ν is the dual variable corresponding to x, and X = diag(x).
[0105] Assume that the iterative sequence generated by the optimization algorithm is {x k}, define the current iteration point x k Feasibility Error Objective function prediction improvement
[0106]
[0107]
[0108] in, is the i-th component of the state vector. c(x k ) is the constraint condition of the system, x k is the current iteration point, x k-1is the previous iteration point, is the gradient vector of the current iteration point, d k Search direction for the current iteration point.
[0109] S is introduced as a deformed Sigmoid function to soften the traditional rigid convergence criterion, which is defined as:
[0110] S(δ k ,ε 0 )=[tanh(ξ * (logδ k / logε 0 ))] / [tanh(ξ)] (32)
[0111] Among them, the function S(δ k ,ε 0 ) will be the interval [ε 0 ,1 / ε 0 ] are smoothly connected, ξ is used to measure the smoothness of the function S in this interval, and ξ is taken as 1.5.
[0112] The convergence depth
[0113]
[0114] Among them, ε 0 is the given tolerance, S is the deformed Sigmoid function, is the feasibility error, Predict the improvement for the objective function.
[0115] In step 3, the SQP-IPM alternating planning algorithm:
[0116] The CDC-based SQP-IPM alternating planning algorithm uses an IPM algorithm with strong global convergence to advance the initial point to the optimal point, uses the convergence depth to measure the distance between the current iteration point and the optimal point, controls the advancement degree, and then switches to an SQP algorithm with good local convergence to make the continuous control process of the iteration point converge quickly to the optimal point. The present invention introduces a progress degree index for switching between SQP-IPM alternating planning algorithms;
[0117] Feasibility change And the objective function change for:
[0118]
[0119]
[0120] in, are the feasibility errors of the current iteration point and the previous iteration point, respectively, f(xk )、f(x k-1 ) are x k 、x k-1 Iteration point objective function value.
[0121] The above formula can be used to calculate the degree of progress Defined as:
[0122]
[0123] in, is the feasibility change, is the change of the objective function, ε 0 is the given tolerance.
[0124] When the progress reaches 1.0, the algorithm has reached ε in terms of feasibility change and objective function improvement. 0 , which means that the algorithm has made almost no progress in improving the objective function and feasibility.
[0125] Based on the convergence depth and progress introduced above, the CDC-based SQP-IPM alternating planning algorithm framework is designed. The specific steps are as follows:
[0126] Step (1): Let k = 0 and give the initial point x 0 , specify the convergence depth threshold θ c , the progress threshold θ p , convergence depth advancement step d θ , the maximum convergence depth value θ max And set the IPM flag to 0;
[0127] Step (2): Select IPM and set its flag bit to 1 for solving the optimization problem;
[0128] Step (3): Iterate one step, k = k + 1. If the termination criterion of IPM is met, the solution is successful and the entire algorithm terminates. Otherwise, go to step (4).
[0129] Step (4): Calculate the convergence depth and check Is it satisfied? If so, the current iteration point x k Assign to x 0 , and clear all the standard bits of SQP, and go to step (6), otherwise go to step (5);
[0130] Step (5): If the IPM termination criterion fails to be met, go to step (3), otherwise the entire algorithm terminates if the solution fails;
[0131] Step (6) Select SQP and set its flag bit to 1, with x 0Solve the optimization problem for the initial point;
[0132] Step (7) iterates one step, k = k + 1. If the termination criterion of SQP is met, the solution is successful and the entire algorithm terminates. Otherwise, go to step (8).
[0133] Step (8) calculates the degree of progress. If If neither the termination criteria of the SQP nor the termination criteria of the SQP are met, go to step (7), otherwise go to step (9);
[0134] Step (9) Set the current iteration point x k Assign to x 0 , go to step (10);
[0135] Step (10) Let θ c =θ c +dθ, if θ c ≥θ max If the solution fails, the entire algorithm terminates; otherwise, the IPM flag is cleared and the process goes to step (2).
[0136] Through the above steps, dynamic optimization of the CHP unit load variation process is achieved.
[0137] Step 4: A simulation is performed using the variable load control demand of a 300MW extraction steam CHP unit as an example to verify the feasibility of the algorithm and the rapid convergence of the continuous control process.
[0138] Matlab is used to organize the model output data, and the change trajectory of the CHP unit's continuously variable load control variables and the output variable curve are plotted, including the change trajectory of the coal mass flow rate, the change trajectory of the turbine high-pressure regulating valve opening, the change trajectory of the extraction butterfly valve opening, the electric output curve, the main steam pressure output curve, and the heating steam flow output curve.
[0139] The present invention provides a dynamic optimization method for a CHP unit load variation process based on SQP-IPM alternating solution, and the technical effects are as follows:
[0140] 1) The convergence depth control criterion proposed by the method of the present invention can measure the dynamic changes of the iterative solution process of the CHP unit load variation. In the unit optimization solution, the over-convergence solution process can be quickly iterated to obtain a better solution; in the case of solution failure, the solution process can also be terminated in advance to reduce the consumption of calculation cost. This method can significantly reduce the optimization calculation time of the CHP unit load variation process.
[0141] 2) The method of the present invention selects IPM and SQP according to global and local convergence to solve the dynamic optimization problem of CHP unit load variation, and uses the convergence depth to control the mutual cooperation and alternating solution of the algorithms. The advantages of different algorithms are fully utilized, and the method has obvious improvements in computational efficiency and convergence when dealing with large-scale complex optimization problems of CHP unit load variation process optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0142] Figure 1 This is the rigid convergence criterion curve and the flexible convergence criterion curve.
[0143] Figure 2 Flowchart of the fast-converging SQP-IPM alternating solution for the dynamic optimization problem of a continuous control process.
[0144] Figure 3 (1) is a graph showing the change trajectory of coal mass flow rate during the continuous load change process of the CHP unit;
[0145] Figure 3 (2) is a diagram showing the change trajectory of the high-pressure regulating valve opening of the steam turbine during the continuous load change process of the CHP unit;
[0146] Figure 3 (3) is a diagram showing the variation trajectory of the opening of the extraction butterfly valve during the continuous load variation process of the CHP unit;
[0147] Figure 3 (4) is a graph showing the electrical output curve of the CHP unit during the continuous variable load process;
[0148] Figure 3 (5) is a main steam pressure output curve diagram of the CHP unit during the continuous variable load process;
[0149] Figure 3(6) is a graph showing the heating steam flow output curve during the continuous variable load process of the CHP unit. DETAILED DESCRIPTION
[0150] The dynamic optimization method of CHP unit variable load process based on SQP-IPM alternating solution includes the following steps:
[0151] Step 1: Give a specific description of the dynamic optimization problem with equality and inequality constraints involving differential algebraic equations, and determine the dynamic optimization proposition of the continuous control process:
[0152] The variable load optimization control problem of the combined heat and power unit in a power plant is expressed as a proposition. The general form of the dynamic optimization proposition is:
[0153]
[0154] In formula (1), F is the dynamic model of DAEs of CHP unit, t 0 and t f are the starting and ending time points of unit optimization control respectively; g E , g IEare equality and inequality path constraints; h E 、h IE For the time t f The terminal value constraint; x(t), y(t), u(t) represent the differential state variable, algebraic output variable, and control variable of the CHP unit at time t respectively;
[0155] 1) Objective function description of the optimization proposition:
[0156] In order to ensure that the CHP unit control process can accurately, quickly and smoothly achieve variable load regulation, the mathematical form of the control variable and the output variable is selected as the optimal control performance objective function, which is described as
[0157]
[0158] In formula (2): t 0 and t f are the starting and ending time points of the unit optimization control, respectively; u is the vector composed of the control variables of the CHP unit, y is the vector composed of the algebraic output variables of the CHP unit, and y * is the set value of the output variable obtained by steady-state optimization, R and S are positive definite weight matrices, and the initial values can be determined by normalizing the variables; J is the objective function, u(t) is the control variable value at time t, u(t-1) is the control variable value at time t-1, and y(t) is the output variable value at time t.
[0159] In the variable load optimization control problem of cogeneration unit, u is the control variable, y is the output variable, and x is the state variable, where:
[0160]
[0161] In formula (3), x is the state variable, corresponding to ψ d is the drum pressure, P H is the generating power of the unit, ψ z Extraction steam pressure for steam turbine heating, V f is the boiler combustion rate; u is the control variable, corresponding to V T is the opening of the steam inlet regulating valve of the high-pressure cylinder of the steam turbine, V B is the fuel quantity of the unit, V H The butterfly valve opening is adjusted for heating extraction steam; y is the output variable, corresponding to ψ t is the pressure before the turbine, P H is the generating power of the unit, m H is the extraction steam flow for heating.
[0162] 2) Description of equality and inequality constraints of optimization propositions:
[0163] ① CHP unit mechanism model:
[0164] By analyzing the characteristics of the thermal components of the heating unit and the process of steam working in the turbine, a simplified nonlinear mechanism model description of the CHP unit can be obtained:
[0165] V m (t) = V B (tt B ) (4)
[0166]
[0167]
[0168] ψ d (t)-ψ t (t) = K 2 (K 1 V f (t)) 1.5 (7)
[0169]
[0170] ψ 1 (t)=0.01ψ t (t)V T (t) (9)
[0171]
[0172] m H (t) = K 7 K 6 m r (t)(96ψ z (t)-ε r (t)+103) (11)
[0173] Formulas (4)-(5) characterize the dynamic relationship of the coal pulverizing system, V B (t) is the coal mass flow rate of the unit, V m (t) is the actual coal mass flow rate entering the coal mill in the pulverizing system, t B is the delay time constant of the milling process, V f (t) is the boiler combustion rate, T f is the milling inertia time constant, t represents the control process time domain value, V B (tt B ) means tt B The unit coal mass flow rate at the moment;
[0174] Equations (6)-(7) represent the dynamic relationship of the drum boiler system, V T (t) is the opening of the turbine regulating valve, ψ d (t) is the drum pressure, ψt (t) is the pressure before the turbine, C d is the boiler heat storage coefficient, δ is a constant coefficient, K 1 is the fuel gain static parameter, K 2 is the superheater resistance coefficient, K 3 is the static parameter of turbine gain;
[0175] Equations (8)-(10) represent the dynamic relationship of the steam turbine system, V H (t) is the opening of the extraction steam regulating butterfly valve, P H (t) is the generating power of the unit, ψ z (t) is the exhaust pressure of the intermediate pressure cylinder, ψ 1 (t) is the first stage pressure of the steam turbine, m r (t) is the mass flow rate of circulating water, ε r (t) is the circulating water return temperature, T t is the turbine inertia time constant, C z is the heat storage coefficient of the heating network heater, K 4 K is the proportion of high and medium pressure cylinders in the turbine work, 5 is the static parameter of low pressure cylinder gain, K 6 is the effective specific heat capacity of the circulating water in the heat network;
[0176] Formula (11) represents the dynamic relationship of the heating system, m H (t) is the heating extraction steam flow rate; K 7 It is the steam flow coefficient converted from the effective heat of extraction steam for heating.
[0177] ②CHP unit control model:
[0178] According to the mechanism model of CHP unit, the control algorithm should realize three key functions: accurate tracking of electric power, rapid recovery of thermal power and safety of system pressure. A proportional-integral control algorithm is designed to coordinate the regulation of V B (t), V T (t), V H (t) direction and magnitude, so that the output variable ψ t (t), m H (t), P H (t) changes. Its mathematical model:
[0179]
[0180] Among them, K PT , K IT Control V T (t) Proportional coefficient and integral coefficient of the variable; K PB , K IB , respectively control V B(t) Proportional coefficient and integral coefficient of the variable; K PH , K IH , respectively control V H (t) Proportional coefficient and integral coefficient of variables; SP ψ , SP P , SP m are the set values of main steam, unit power generation and heating extraction steam flow respectively; ψ t (t) is the current value of the main steam pressure control process, P H (t) is the current value of the unit power control process, m H (t) is the current value of the heating extraction steam flow control process.
[0181] Considering the safe and stable operation of the CHP unit, the fluctuation range of the process value of the output variable and the error range of the steady-state value during the control process cannot exceed the allowable limit and should meet the following constraints:
[0182]
[0183]
[0184] Among them, △ ψ ,△ P ,△ m are the fluctuation ranges of main steam pressure, electric power and extraction steam flow respectively; δ ψ , δ P , δ m are the error ranges of main steam pressure, electric power and extraction steam flow respectively; within the optimization control period, ψ t (t f ), P H (t f )、m H (t f ) are the end time t f Final values of main steam pressure, electrical power, and extraction steam flow.
[0185] ③Boundary constraints:
[0186] CHP unit in [t 0 ,t f ] The variable load regulation is completed within the period, and the initial and final states of each control variable and output variable are as follows:
[0187]
[0188]
[0189] Where, t 0 and t fare the starting and ending time points of unit optimization control respectively;
[0190] are the starting time t 0 Output: the pressure before the steam turbine, the generating capacity of the unit, and the value of the extraction steam flow for heating;
[0191] The termination time t f Output: the pressure before the steam turbine, the generating capacity of the unit, and the value of the extraction steam flow for heating;
[0192] are the starting time t 0 Control the values of the opening of the high-pressure cylinder steam inlet regulating valve, the unit fuel quantity, and the opening of the extraction steam regulating butterfly valve;
[0193] The termination time t f The controlled quantities include the opening of the high-pressure cylinder steam inlet regulating valve, the fuel quantity of the unit, and the opening of the steam extraction regulating butterfly valve.
[0194] ④Path constraints:
[0195] In the process of variable load regulation, the control variable V T (t), V B (t), V H (t) The following constraints need to be met:
[0196]
[0197] In the formula, and are the minimum and maximum values of the turbine valve opening change, and are the minimum and maximum values of the unit fuel quantity change, and They are respectively the minimum and maximum values of the opening change of the extraction steam regulating butterfly valve.
[0198] Step 2: Solve the dynamic optimization problem in step 1, using the configuration method based on the Legendre polynomial discrete format, and simultaneously discretize the state variables and control variables to form a large-scale NLP problem:
[0199] 1) Time domain transformation:
[0200] The optimal control time domain t∈[t 0 ,t f ] is divided into p segments of finite elements, that is, 0 = t 0 <t 1 <…t q-1 <t q <…t p=t f , it can be seen that the step size of each time domain is
[0201] △t=t q -t q-1 ,q∈[1,p] (18)
[0202] Where: △t is the step size of each time domain, q is the number of finite elements, p is the total number of finite element segments, t q is the time variable of the qth finite element, t q-1 is the time variable of the q-1th finite element;
[0203] In the Gaussian pseudospectral method, the configuration point interval is selected as τ∈[-1,1], so the time variable on the qth segment is converted into the variable τ through time domain transformation, and the transformation is as follows:
[0204]
[0205] Where: τ is the variable converted by time domain transformation on the qth segment time variable;
[0206] Then the description of the optimal control problem in each time zone is as follows:
[0207]
[0208] In formula (20): J (q) is the objective function in the qth time zone, τ 0 is the initial configuration point of the time domain transformation of the time variable on the qth segment, τ f The terminal configuration points of the time-variable time-domain transformation on the qth segment;
[0209] x (q) (τ) is the differential state variable of the τth configuration point in the qth time period, x (q) (τ 0 ) is the τth time segment in the qth time segment 0 The differential state variables of the collocation points, x (q) (τ f ) is the τth time segment in the qth time segment f Differential state variables of configuration points;
[0210] y (q) (τ) is the algebraic output variable of the τth configuration point in the qth time period, y (q) (τ 0 ) is the τth time segment in the qth time segment 0 The algebraic output variable of the collocation point, y (q) (τ f ) is the τth time segment in the qth time segment f Algebraic output variables of collocation points;
[0211] u (q) (τ) is the control variable of the τ-th configuration point in the q-th time segment, F is the DAEs dynamic model, and △t is the step size of each time domain;
[0212] 2) Variable discretization:
[0213] The Gaussian pseudospectral method discretizes the control variables and state output variables at the Legendre-Gauss (LG) configuration points at the same time, and then approximates the control variables and state variables by constructing Lagrange interpolation functions. 0 ,τ f ], the N-order Legendre polynomial is
[0214]
[0215] Where: P N (τ) is the Nth order Legendre polynomial, N is the order, Find the Nth order derivative of the function with respect to τ;
[0216] With the zero point and τ of formula (21) 0 Composed of N+1 configuration points, at each t (q) In the time domain, the differential state variables and control variables are approximately expressed by polynomials as
[0217]
[0218] Where: x (q) (τ) is the differential state variable of the τth configuration point in the qth time period, is the interpolation function of the state variable, X (q) (τ i ) is the value of the state variable at the i-th configuration point of the q-th finite element, u (q) (τ) is the control variable of the τth configuration point in the qth time period, is the interpolation function of the control variable, U (q) (τ i ) is the value of the control variable at the i-th configuration point of the q-th finite element.
[0219] In formula (22), the Lagrange interpolation basis function is
[0220]
[0221] Where: N is the interpolation order; is the interpolation function of the state variables, is the interpolation function of the control variable, The independent variable of the interpolation function at the j-th collocation point, τ iis the interpolation function independent variable on the i-th configuration point; τ is the interpolation function independent variable of the time variable on the q-th segment transformed by time domain transformation;
[0222] By taking the derivative of the state variable from equation (22), the interpolation function can be approximated to the differential form of the state variable, and we can obtain k The state variable derivative of the point can be approximately expressed as
[0223]
[0224] in: is the derivative of the differential state variable at the kth collocation point of the pth segment finite element, is the interpolation function of the state variable at the kth collocation point of the pth segment finite element, X (q) (τ i ) is the state variable at the i-th configuration point of the q-th finite element;
[0225] Combining equation (20) and equation (24), we can get the differential algebraic equality constraint:
[0226]
[0227] Among them: t is the step size of each time domain; F is the DAEs dynamic model, X(τ k ) is the τth k The state variables at the configuration points, M(τ k )τ k Other variables at the configuration point, τ k is the kth collocation point of the time-domain transformation of the time variable on the qth segment;
[0228] In the process of variable load control of the unit, the continuity of the change trajectory of each state output must be ensured, so the state variable must also change continuously, that is, the state variable at the terminal moment of each segment is used as the state variable at the initial moment of the next segment, and the constraint equation at the connection is:
[0229] X (q-1) (τ f )=X (q) (τ 0 ) (26)
[0230] Where: X (q-1) (τ f ) is the q-1th finite element τ f The state variables on the configuration points, X (q) (τ 0 ) is the qth finite element τ 0 State variables on configuration points;
[0231] Since the LG configuration point does not contain the terminal state value, the terminal state variable is calculated according to the Legendre-Gauss quadrature formula from equation (26):
[0232]
[0233] In the formula, A k and f(τ k ) are Gaussian integral weight coefficient and quadrature function respectively.
[0234] After the above discretization process, the optimal control problem of the CHP unit's variable load dynamic process can be transformed into a nonlinear programming problem (NLP) for solution.
[0235] Step 3: Design an SQP-IPM alternating solution method to solve the problem, and introduce a convergence depth control strategy into the alternating solution method to improve the real-time performance of the method;
[0236] 1) Convergence depth control strategy:
[0237] The nonlinear programming problem shown in the original proposition (20) can be simplified as
[0238]
[0239] in, is a scalar objective function, x∈R n is an n-dimensional real vector, c∈R m is an m-dimensional equation.
[0240] The solution to problem (28) must satisfy the following first-order optimality conditions:
[0241]
[0242] in, is the gradient vector of the objective function, is the Jacobian matrix of the constraint equation. λ is the Lagrange multiplier corresponding to the constraint equation, ν is the dual variable corresponding to x, and X = diag(x).
[0243] Assume that the iterative sequence generated by the optimization algorithm is {x k}, define the current iteration point x k Feasibility Error Objective function prediction improvement
[0244]
[0245]
[0246] in, is the i-th component of the state vector. c(x k ) is the constraint condition of the system, x k is the current iteration point, x k-1 is the previous iteration point, is the gradient vector of the current iteration point, d k Search direction for the current iteration point.
[0247] S is introduced as a deformed Sigmoid function to soften the traditional rigid convergence criterion, which is defined as:
[0248] S(δ k ,ε 0 )=[tanh(ξ * (logδ k / logε 0 ))] / [tanh(ξ)] (32)
[0249] Among them, the function S(δ k ,ε 0 ) will be the interval [ε 0 ,1 / ε 0 ] are smoothly connected, ξ measures the smoothness of the function S in this interval, and ξ is taken as 1.5.
[0250] The Sigmoid function is used to soften the traditional rigid convergence criterion, and different tolerances ε are given. 0 ,like Figure 1 As shown, curve 1 is 1×10 -10 , curve 2 is 1×10 -5 , curve 3 is 1×10 -3 , Curve 4 is the rigid criterion. The convergence depth is described as:
[0251]
[0252] Among them, ε 0 is the given tolerance, S is the deformed Sigmoid function, is the feasibility error, Predict the improvement for the objective function.
[0253] Compared with the first-order optimality condition, the convergence depth more intuitively and uniformly characterizes the quality of each iteration point generated by the optimization algorithm, including feasibility and optimality, and can quantitatively describe the degree of distance between it and the optimal point. When the convergence depth reaches 1.0, the feasibility error and the improvement in the objective function prediction reach ε 0 , at this point the iteration point is already a good approximation of the optimal point.
[0254] 2) SQP-IPM alternating planning algorithm:
[0255] The CDC-based SQP-IPM alternating planning algorithm uses an IPM algorithm with strong global convergence to advance the initial point to the optimal point, uses the convergence depth to measure the distance between the current iteration point and the optimal point, controls the advancement degree, and then switches to an SQP algorithm with good local convergence to make the continuous control process of the iteration point converge quickly to the optimal point. The present invention introduces a progress degree index for switching between SQP-IPM alternating planning algorithms.
[0256] Feasibility change And the objective function change for:
[0257]
[0258]
[0259] in, are the feasibility errors of the current iteration point and the previous iteration point, respectively, f(x k )、f(x k-1 ) are x k 、x k-1 Iteration point objective function value.
[0260] The above formula can be used to calculate the degree of progress Defined as:
[0261]
[0262] in, is the feasibility change, is the change of the objective function, ε 0 is the given tolerance.
[0263] When the progress reaches 1.0, the algorithm has reached ε in terms of feasibility change and objective function improvement. 0 , which means that the algorithm has made almost no progress in improving the objective function and feasibility.
[0264] Based on the convergence depth and progress introduced above, the CDC-based SQP-IPM alternating planning algorithm framework is designed. First, IPM is selected to solve the NLP problem. In the iterative process, the convergence depth of each step is calculated using formula (33). When the convergence depth meets the specified threshold θ c When the iteration point information is saved, the current algorithm is terminated and switched to SQP, and the iteration point saved last time is used as the initial point to continue the iteration. Since different algorithms have different local convergence radii, the present invention first gives a threshold value θ c An initial value If SQP is unsuccessful, the iteration point obtained from IPM does not enter the local convergence radius of these algorithms, and the convergence depth threshold is increased by d θ ∈(0,1), return to IPM for iteration. The specific steps of the alternating planning algorithm are as follows:
[0265] Step (1): Let k = 0 and give the initial point x 0 , specify the convergence depth threshold θ c , the progress threshold θ p , convergence depth advancement step d θ , the maximum convergence depth value θ max And set the IPM flag to 0;
[0266] Step (2): Select IPM and set its flag bit to 1 for solving the optimization problem;
[0267] Step (3): Iterate one step, k = k + 1. If the termination criterion of IPM is met, the solution is successful and the entire algorithm terminates. Otherwise, go to step (4).
[0268] Step (4): Calculate the convergence depth and check Is it satisfied? If so, the current iteration point x k Assign to x 0 , and clear all the standard bits of SQP, and go to step (6), otherwise go to step (5);
[0269] Step (5): If the termination criterion of IPM (solution failure) is not met, go to step (3), otherwise the entire algorithm terminates if the solution fails;
[0270] Step (6): Select SQP and set its flag bit to 1, with x 0 Solve the optimization problem for the initial point;
[0271] Step (7): Iterate one step, k = k + 1. If the termination criterion of SQP is met, the solution is successful and the entire algorithm terminates. Otherwise, go to step (8).
[0272] Step (8): Calculate the degree of progress. If If neither the termination criteria of the SQP nor the termination criteria of the SQP are met, go to step (7), otherwise go to step (9);
[0273] Step (9): Set the current iteration point x k Assign to x 0 , go to step (10);
[0274] Step (10): Let θ c =θ c +dθ, if θ c ≥θ maxIf the solution fails, the entire algorithm terminates; otherwise, the IPM flag is cleared and the process goes to step (2).
[0275] The specific implementation block diagram of the present invention based on CDC's SQP-IPM alternating solution to the dynamic optimization problem is as follows: Figure 2 shown.
[0276] Step 4: Take the variable load control demand of a 300MW extraction CHP unit as an example to simulate and verify the feasibility of the algorithm and the rapid convergence of the continuous control process. Matlab is used to organize the model output data, and the variable change trajectory and output variable curve of the CHP unit continuous variable load control are plotted, including the change trajectory of coal mass flow, the change trajectory of the high-speed control valve opening of the turbine, the change trajectory of the opening of the extraction butterfly valve, the electric output curve, the main steam pressure output curve, and the heating steam flow output curve.
[0277] Example:
[0278] In order to verify the effectiveness of the method proposed in the present invention, a 300MW extraction steam CHP unit model was built on the Matlab platform for simulation. The relevant parameters of the unit are shown in Tables 1, 2 and 3. Table 4 is the example data of the continuous variable load demand of the application scenario.
[0279] Table 1 CHP unit operating point parameters
[0280]
[0281] Table 2 CHP unit operating data
[0282]
[0283]
[0284] Table 3 CHP unit control system parameters
[0285] <![CDATA[V T Control Parameters]]> <![CDATA[V B Control Parameters]]> <![CDATA[V H Control Parameters]]> <![CDATA[K PT =-0.9]]> <![CDATA[K PB =0.1]]> <![CDATA[K PH =0.01 <!-- 17 -->]]> <![CDATA[K IT =-1]]> <![CDATA[K IB =0.01]]> <![CDATA[K IH =10]]>
[0286] Table 4 Continuously variable load requirements of CHP units under heating conditions
[0287] Time Initial Value Terminal Value 300s-600s (250,16.67,390) (220,16.67,380) 600s-900s (220,16.67,380) (200,16.67,390) 900s-1200s (200,16.67,390) (230,16.67,395) 1200s-1500s (230,16.67,395) (210,16.67,385) 1500s-1800s (210,16.67,385) (240,16.67,395) 1800s-2100s (240,16.67,395) (250,16.67,385) 2100s-2400s (250,16.67,385) (220,16.67,380) 2400s-2700s (220,16.67,380) (200,16.67,390) 2700s-3000s (200,16.67,390) (230,16.67,395) 3000s-3300s (230,16.67,395) (210,16.67,385) 3300s-3600s (210,16.67,385) (240,16.67,395) 3600s-3900s (240,16.67,395) (260,16.67,385)
[0288] 1.Simulation calculation performance analysis:
[0289] The CDC-based IPM-SQP alternating solution method, CDC-based IPM and single IPM are used to calculate the optimal control problem of the CHP unit in the continuous variable load process, and their performance indicators are shown in Table 5. As shown in Table 5, under different operating conditions and variable load requirements, the optimal control terminal values of the three methods can reach the set value, and the feasibility error also meets the actual requirements, and each period can converge successfully, but the number of iterations and solution time of the discrete model are quite different.
[0290] Table 5 Comparison of computing performance of different algorithms
[0291] Performance Indicators CDC-IPM-SQP CDC-IPM IPM Time-consuming solution 10.17s 19.92s 38.68s Convergence Success in all periods Success in all periods Success in all periods Terminal Value The final value is reached in each period The final value is reached in each period The final value is reached in each period
[0292] When solving with the improved CDC method, the time interval between two calculations can be shortened, and the deviation between the initial value and the optimal solution can also be reduced, so the number of iterations can be reduced. This feature can effectively improve the computational efficiency of the algorithm and greatly shorten the operation time. During the calculation of the example, compared with the IPM solution alone, the CDC criterion used in the CDC-IPM solution makes the constraint violation slightly greater than the result of the traditional H criterion, and the objective function value is slightly less than the optimal value, but the calculation time is shortened by about 50%. The use of the CDC criterion improvement and IPM-SQP alternating solution method reduces the variable load calculation time of the 12 time periods of the example by about 70%, effectively improving the real-time performance of the scenario application.
[0293] 2. Analysis of simulation variable load effect:
[0294] In view of the demand for continuous variable load in the application scenario of integrated energy system, the control process of continuously changing the electric and thermal loads in 12 periods under the heating condition of CHP unit is selected for simulation, and each period is 300s. The method (CDC-IPM-SQP) proposed in this invention is used for continuous dynamic optimization and solution. The control quantity trajectory in this process is shown in Figure 3 (1), Figure (2), and Figure (3), and the output quantity curve is shown in Figure 3 (4), Figure 3 (5), and Figure 3 (6).
[0295] The results show that the control variable V T 、V B 、V H The trajectory enables the CHP unit to achieve coherent planning for the dynamic process optimization control of load variation, provides accurate quantitative indicators for repeated adjustment of flow and valve opening during unit operation, and can guide the continuous load variation regulation of actual CHP units. H , thermal power (related variables m H ) can accurately meet the variable load demand, and the unit main steam pressure ψ t , Heating steam flow m HThe curve continuously depicts the pressure and heat fluctuations caused by the variable load process, which is conducive to the analysis of safe and stable operation in the integrated energy system calculation scenario.
Claims
1. Dynamic optimization method for CHP unit variable load process based on SQP-IPM alternating solution, characterized by The following steps are involved: Step 1: Formulate the variable load optimization control problem of the CHP unit of the power plant into a proposition; Step 2: Solve the dynamic optimization problem in step 1, and use the collocation method based on the Legendre polynomial discretization format to discretize the state variables and control variables at the same time to form an NLP problem; Step 3: Design an SQP-IPM alternating solution method to solve NLP problems, and introduce a convergence depth control strategy in the alternating solution method; Through the above steps, dynamic optimization of the CHP unit load variation process is achieved; In step 1, the form of the dynamic optimization proposition is: In formula (1): min J(x(t0),t0,x(t f ),t f ) represents the objective function of minimizing the dynamic optimization proposition with respect to the state variables; t0, t f They are the starting and ending time points of unit optimization control respectively; t represents the time variable in the control time domain; x(t), y(t), u(t) represent the differential state variable, algebraic output variable, and control variable of the CHP unit at time t, respectively; F(x(t), y(t), u(t), t) represents the relationship function model between the state variable, output variable, and control variable; g E , g IE are equality and inequality path constraints; h E 、h IE For the time t f The terminal value constraint of x(t0), y(t0), x(t f ), y(t f ) represent the initial value of the state variable, the initial value of the output variable, the final value of the state variable, and the final value of the output variable respectively; The optimal control performance objective function is described as: In formula (2): t0 and t f are the starting and ending time points of the unit optimization control respectively; u is the vector composed of the control variables of the CHP unit; y is the vector composed of the algebraic output variables of the CHP unit, y * is the set value of the output variable obtained by steady-state optimization; R and S are positive definite weight matrices, which can determine the initial value by normalizing the variables; J is the objective function, u(t) is the control variable value at time t, u(t-1) is the control variable value at time t-1; y(t) is the output variable value at time t; In the variable load optimization control problem of the cogeneration unit, u is the control variable, y is the output variable, and x is the state variable, where: In formula (3), x is the state variable, corresponding to ψ d is the drum pressure, P H is the generating power of the unit, ψ z Extraction steam pressure for steam turbine heating, V f is the boiler combustion rate; u is the control variable, corresponding to V T is the opening of the steam inlet regulating valve of the high-pressure cylinder of the steam turbine, V B is the fuel quantity of the unit, V H Adjust the butterfly valve opening for heating extraction steam; y is the output variable, corresponding to ψ t is the pressure before the turbine, P H is the generating power of the unit, m H is the extraction steam flow for heating; In the step 1, ①. The mechanism model of CHP unit is as follows: Formula (4) to (5) represent the dynamic relationship of coal pulverizing system, V B (t) is the coal mass flow rate of the unit, V m (t) is the actual coal mass flow rate entering the coal mill in the pulverizing system, t B is the delay time constant of the milling process, V f (t) is the boiler combustion rate, T f is the milling inertia time constant, t represents the control process time domain value, V B (tt B ) means tt B The unit coal mass flow rate at the moment; ψ d (t)-ψ t (t)=K2(K1V f (t)) 1.5 (7); Equations (6) to (7) represent the dynamic relationship of the drum boiler system, V T (t) is the opening of the turbine regulating valve, ψ d (t) is the drum pressure, ψ t (t) is the pressure before the turbine, C d is the boiler heat storage coefficient, δ is a constant coefficient, K1 is the fuel gain static parameter, K2 is the superheater resistance coefficient, and K3 is the turbine gain static parameter; ψ1(t)=0.01ψ t (t)V T (t) (9); Equations (8) to (10) represent the dynamic relationship of the steam turbine system, V H (t) is the opening of the extraction steam regulating butterfly valve, P H (t) is the generating power of the unit, ψ z (t) is the exhaust pressure of the intermediate pressure cylinder, ψ1(t) is the first stage pressure of the steam turbine, m r (t) is the mass flow rate of circulating water, ε r (t) is the circulating water return temperature, T t is the inertia time constant of the steam turbine, C z is the heat storage coefficient of the heating network heater, K4 is the proportion of high and medium pressure cylinders in the work of the turbine, K5 is the static parameter of the low pressure cylinder gain, and K6 is the effective specific heat capacity of the circulating water in the heating network; m H (t)=K7K6m r (t)(96ψ z (t)-ε r (t)+103)(11); Formula (11) represents the dynamic relationship of the heating system, m H (t) is the heating extraction steam flow rate; K7 is the steam flow coefficient converted from the effective heat of the heating extraction steam; ②. CHP unit control model: Design a proportional-integral control algorithm to coordinately adjust V B (t), V T (t), V H (t) direction and magnitude, so that the output variable ψ t (t), m H (t), P H (t) changes, and its mathematical model is as follows: Among them, K PT , K IT Control V T (t) Proportional coefficient and integral coefficient of the variable; K PB , K IB , respectively control V B (t) Proportional coefficient and integral coefficient of the variable; K PH , K IH , respectively control V H (t) Proportional coefficient and integral coefficient of variables; SP ψ , SP P , SP m are the set values of main steam, unit power generation and heating extraction steam flow respectively; ψ t (t) is the current value of the main steam pressure control process, P H (t) is the current value of the unit power control process, m H (t) is the current value of the heating extraction steam flow control process; Considering the safe and stable operation of the CHP unit, the fluctuation range of the process value of the output variable and the error range of the steady-state value during the control process cannot exceed the allowable limit and should meet the following constraints: Among them, △ ψ ,△ P ,△ m are the fluctuation ranges of main steam pressure, electric power and extraction steam flow respectively; δ ψ , δ P , δ m are the error ranges of main steam pressure, electric power and extraction steam flow respectively; within the optimization control period, ψ t (t f ), P H (t f )、m H (t f ) are the end time t f Final values of main steam pressure, electric power, and extraction steam flow rate; ③. Boundary constraints: CHP unit in [t0,t f ] The variable load regulation is completed within the period, and the initial and final states of each control variable and output variable are as follows: Where t0 and t f are the starting and ending time points of unit optimization control respectively; are the values of the output turbine inlet pressure, unit power generation, and heating extraction steam flow at the starting time t0 respectively; The termination time t f Output: the pressure before the steam turbine, the generating capacity of the unit, and the value of the extraction steam flow for heating; They are respectively the values of the opening of the high-pressure cylinder steam inlet regulating valve, the fuel quantity of the unit, and the opening of the extraction steam regulating butterfly valve at the starting time t0; The termination time t f Control the values of the opening of the high-pressure cylinder steam inlet regulating valve, the unit fuel quantity, and the opening of the extraction steam regulating butterfly valve; ④. Path constraints: In the process of variable load regulation, the control variable V T (t), V B (t), V H (t) The following constraints need to be met: In the formula, and are the minimum and maximum values of the turbine valve opening change respectively; and are the minimum and maximum values of the unit fuel quantity change, respectively; and are the minimum and maximum values of the opening change of the extraction steam regulating butterfly valve respectively; the step 2 comprises the following steps: 1) Time domain transformation: The optimal control time domain t∈[t0,t f ] is divided into p segments of finite elements, that is, 0 = t0 <t1<…t q-1 <t q <…t p =t f , it can be seen that the step size of each time domain is: △t=t q -t q-1 ,q∈[1,p] (18); Where: △t is the step size of each time domain, q is the number of finite elements, p is the total number of finite element segments, t q is the time variable of the qth finite element, t q-1 is the time variable of the q-1th finite element; In the Gaussian pseudospectral method, the configuration point interval is selected as τ∈[-1,1], so the time variable on the qth segment is converted into the variable τ through time domain transformation, and the transformation is as follows: Where: τ is the variable converted by time domain transformation on the qth segment time variable; The optimal control problem in each time zone is described as follows: In formula (20): J (q) is the objective function in the qth time region, τ0 is the initial configuration point of the time domain transformation of the time variable in the qth time region, τ f The terminal configuration points of the time-variable time-domain transformation on the qth segment; x (q) (τ) is the differential state variable of the τth configuration point in the qth time period, x (q) (τ0) is the differential state variable of the τ0th configuration point in the qth time period, x (q) (τ f ) is the τth time segment in the qth time segment f Differential state variables of configuration points; y (q) (τ) is the algebraic output variable of the τth configuration point in the qth time period, y (q) (τ0) is the algebraic output variable of the τ0th configuration point in the qth time segment, y (q) (τ f ) is the τth time segment in the qth time segment f Algebraic output variables of collocation points; u (q) (τ) is the control variable of the τ-th configuration point in the q-th time segment, F is the DAEs dynamic model, and △t is the step size of each time domain; 2) Variable discretization: The Gaussian pseudospectral method discretizes the control variables and state output variables at the Legendre-Gauss (LG) configuration points at the same time, and then approximates the control variables and state variables by constructing Lagrange interpolation functions. f ], the N-order Legendre polynomial is: Where: P N (τ) is the Nth order Legendre polynomial, N is the order, Find the Nth order derivative of the function with respect to τ; The zero point of equation (21) and τ0 form N+1 configuration points. (q) In the time domain, the differential state variables and control variables are approximately expressed by polynomials as follows: Where: x (q) (τ) is the differential state variable of the τth configuration point in the qth time period, is the interpolation function of the state variable, X (q) (τ i ) is the value of the state variable at the i-th configuration point of the q-th finite element; u (q) (τ) is the control variable of the τth configuration point in the qth time period, is the interpolation function of the control variable, U (q) (τ i ) is the value of the control variable at the i-th configuration point of the q-th finite element; In formula (22), the Lagrange interpolation basis function is: Where: N is the interpolation order; is the interpolation function of the state variables, is the interpolation function of the control variable, The independent variable of the interpolation function at the j-th collocation point, τ i is the interpolation function independent variable on the i-th configuration point; τ is the interpolation function independent variable of the time variable on the q-th segment transformed by time domain transformation; By taking the derivative of the state variable from equation (22), the interpolation function can be differentiated to approximate the differential form of the state variable, and we can obtain k The state variable derivative of the point can be approximately expressed as: in: is the derivative of the differential state variable at the kth collocation point of the pth segment finite element, is the interpolation function of the state variable at the kth collocation point of the pth segment finite element, X (q) (τ i ) is the state variable at the i-th configuration point of the q-th finite element; Combining equation (20) and equation (24), we can get the differential algebraic equality constraint: Where: △t is the step size of each time domain; F is the DAEs dynamic model, X(τ k ) is the τth k The state variables at the configuration points, M(τ k )τ k Other variables at the configuration point, τ k is the kth collocation point of the time-domain transformation of the time variable on the qth segment; The state variables at the terminal moment of each segment are used as the state variables at the initial moment of the next segment, and the constraint equation at the connection is: X (q-1) (t f )=X (q) (τ0) (26); Where: X (q-1) (τ f ) is the q-1th finite element τ f The state variables on the configuration points, X (q) (τ0) is the state variable at the τ0th configuration point of the qth finite element; Since the LG configuration point does not contain the terminal state value, the terminal state variable is calculated according to the Legendre-Gauss quadrature formula from equation (26): In the formula, A k and f(τ k ) are Gaussian integral weight coefficient and quadrature function respectively; In step 3, the convergence depth control strategy is as follows: The nonlinear programming problem shown in the original proposition (20) can be simplified as: in, is a scalar objective function, x∈R n is an n-dimensional real vector, c∈R m is an m-dimensional equation; The solution to problem (28) must satisfy the following first-order optimality conditions: in, is the gradient vector of the objective function, is the Jacobian matrix of the constraint equation; λ is the Lagrange multiplier corresponding to the constraint equation, ν is the dual variable corresponding to x, X = diag(x); Assume that the iterative sequence generated by the optimization algorithm is {x k }, define the current iteration point x k Feasibility Error Objective function prediction improvement in, is the i-th component of the state vector; c(x k ) is the constraint condition of the system, x k is the current iteration point, x k-1 is the previous iteration point, is the gradient vector of the current iteration point, d k Search direction for the current iteration point; S is introduced as a deformed Sigmoid function to soften the traditional rigid convergence criterion, which is defined as: S(δ k ,ε0)=[tanh(ξ * (logδ k / logε0))] / [tanh(ξ)] (32) Among them, the function S(δ k ,ε0) smoothly connects the interval [ε0,1 / ε0], and ξ is used to measure the smoothness of the function S in this interval; The convergence depth Among them, ε0 is the given tolerance, S is the deformed Sigmoid function, is the feasibility error, Predict the improvement for the objective function.
2. The method for dynamic optimization of CHP unit load variation process based on SQP-IPM alternating solution according to claim 1 is characterized by: In step 3, the SQP-IPM alternating planning algorithm is as follows: The progress index is introduced to switch between SQP-IPM alternating planning algorithms; Feasibility change And the objective function change for: in, are the feasibility errors of the current iteration point and the previous iteration point, respectively, f(x k )、f(x k-1 ) are x k 、x k-1 Iteration point objective function value; The above formula can be used to calculate the degree of progress Defined as: in, is the feasibility change, is the change of the objective function, and ε0 is the given tolerance.
3. The method for dynamic optimization of CHP unit load variation process based on SQP-IPM alternating solution according to claim 2 is characterized by: Based on the introduced convergence depth and progress degree, the CDC-based SQP-IPM alternating planning algorithm framework is designed. The specific steps are as follows: Step (1): Let k = 0 and give the initial point x 0 , specify the convergence depth threshold θ c , the progress threshold θ p , convergence depth advancement step d θ , the maximum convergence depth value θ max And set the IPM flag to 0; Step (2): Select IPM and set its flag bit to 1 for solving the optimization problem; Step (3): Iterate one step, k = k + 1. If the termination criterion of IPM is met, the solution is successful and the entire algorithm terminates. Otherwise, go to step (4). Step (4): Calculate the convergence depth and check Is it satisfied? If so, the current iteration point x k Assign to x 0 , and clear all the standard bits of SQP, and go to step (6), otherwise go to step (5); Step (5): If the IPM termination criterion fails to be met, go to step (3), otherwise the entire algorithm terminates if the solution fails; Step (6) Select SQP and set its flag bit to 1, with x 0 Solve the optimization problem for the initial point; Step (7) iterates one step, k = k + 1. If the termination criterion of SQP is met, the solution is successful and the entire algorithm terminates. Otherwise, go to step (8). Step (8) calculates the degree of progress. If If neither the termination criteria of the SQP nor the termination criteria of the SQP are met, go to step (7), otherwise go to step (9); Step (9) Set the current iteration point x k Assign to x 0 , go to step (10); Step (10) Let θ c =θ c +dθ, if θ c ≥θ max If the solution fails, the entire algorithm terminates; otherwise, the IPM flag is cleared and the process goes to step (2).
