A power generation control method of distributed parallel multi-layer model predictive control

Abstract

The application provides a power generation control method based on a distributed parallel multi-layer model predictive control, which comprises two modules of a distributed parallel system and a multi-layer input model predictive control. The distributed parallel system constructs a multi-region power generation control model based on model predictive control according to an actual power system, forms a distributed multi-agent virtual system, and corrects each other with the actual system. The regional control deviation, the frequency deviation, the epidemic prevention and health management information, the weather information, the four-hour predicted load value and the day-ahead scheduling constitute the multi-layer input of the model predictive control module. The distributed parallel multi-layer model predictive control method can solve the problems existing in the current power generation control method, effectively balance the active power of the power generation control system, reduce the frequency deviation, fully develop the potential of the artificial system, solve the problem of establishing an accurate mathematical model for a complex system, optimize the difference between the states evaluated by the actual system and the artificial system, and improve the fitting degree of the actual system and the virtual system.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of complex generator units of power systems, and relates to a distributed parallel multi-layer model predictive control method, which is suitable for power generation control of power systems. BACKGROUND

[0002] The existing power generation control method cannot well solve the nonlinearity, strong coupling and uncertainty in the power generation control process of complex power grids, and cannot establish an accurate and unified mathematical model for complex power grids.

[0003] In addition, the control object of the existing model predictive control system is a fixed mathematical model, which is not suitable for control adjustment when the system model parameters change.

[0004] Therefore, a distributed parallel multi-layer model predictive control power generation control method is proposed to control and adjust the system model with changing parameters, and fully play the role of the virtual part of the parallel system. SUMMARY

[0005] A distributed parallel multi-layer model predictive control power generation control method, which combines a distributed parallel system and a multi-layer input model predictive control, is used for power generation control, has the functions of predicting the state matrix, input matrix, output matrix, and disturbance matrix of the virtual control system, calculating the optimal control sequence of the model, and applying the sequence to the actual control system; plays an optimization role on the control object, can eliminate safety hazards, make the artificial system play a leading role, and can obtain the optimal control scheme through simulation; the power generation control method can analyze the behavior and reaction of each mirror virtual system through the calculation experiment of the artificial system, evaluate the effect of different mathematical models, and correct the state matrix, input matrix, output matrix, and disturbance matrix of the mirror virtual system and the actual system, judge whether the optimal control sequence applied to the virtual intelligent agent power generation system will appear abnormal, if abnormal, transmit the abnormal unit position information to the virtual system controller, so that the relevant personnel of the control intelligent agent system can quickly master the various situations of the system and take corresponding actions; in the use process, the steps of the i-th intelligent agent of the k-th parallel system are:

[0006] Step (1): Establish four parallel systems with the same parameters as the actual system;

[0007] Step (2): Establish a multi-region power generation control model based on model predictive control between the actual system and each parallel system;

[0008] Step (3): Obtain the frequency deviation Δf i,k of each region received by the i-th region of the k-th parallel system, the regional control deviation ACE i-j,k , and the four-hour predicted load value ΔPi,k,t＝4h , daily dispatch ΔP i,k,t＝24h , temperature and humidity WS i,k , rainfall JY i,k , wind direction FX i,k , wind speed FS i,k , rainfall JY i,k , solar radiation intensity QD i,k , risk area type LX i,k , crowd home management information XX i,k , case number SL i,k ;

[0009] Step (4): build a model predictive control model of the i-th region of the k-th parallel system;

[0010] The model predictive control model of the i-th region of the k-th parallel system is:

[0011]

[0012] In the formula, x i,k is a state variable, u i,k is an input variable, w i,k is a disturbance variable, A i,k is a state matrix of the system, B i,k is an input matrix of the system, D i,k is a disturbance matrix of the system; denotes differentiation; wherein x i,k is:

[0013] x i,k (t) = [Δf i,k , ACE i-j,k , ΔP i,k,t＝4h , ΔP i,k,t＝24h , WS i,k , JY i,k , FX i,k , FS i,k , JY i,k , QD i,k , LX i,k , XX i,k , SL i,k ] T (2)

[0014] Step (5): output the control amount of the model predictive control of the i-th region of the k-th parallel system;

[0015] y i,k (t) = C i,k x i,k (t) (3)

[0016] In the formula, Ci,k Let y be the output matrix of the system. i,k The power ΔP sent to the i-th generator of the k-th parallel system i,k ;

[0017] Step (6): Compare the frequency deviation Δf of the four parallel systems. i,k Where k = 1, 2, 3, 4, choose Δf i,k A corresponding to the minimum value of the parallel system i,k B i,k C i,k D i,k It is passed to the actual system as the control parameters of the actual system.

[0018] The present invention has the following advantages and effects compared with the prior art:

[0019] (1) Distributed parallel multilayer model predictive control combines parallel systems and model predictive control. It has an optimization effect on control objects with nonlinear, uncertain, time-varying and time-delay characteristics. Moreover, the mutual correction between the actual system and the virtual system on the state variable vector, input and output variable vector and disturbance vector can make up for the shortcomings of model predictive control that can only control fixed mathematical models.

[0020] (2) The artificial system in the parallel system builds a mathematical model according to actual needs, obtains adaptive control through calculation experiments, and applies the simulation results after repeated iterations and deductions to the virtual system. After analyzing the behavior and reaction of the simulation results to the virtual system, the simulation results are then applied to the actual system. This can effectively balance the contradiction between the control effect and resource occupation of the distributed multi-agent power grid, avoid the direct application of inappropriate mathematical models to the actual system causing abnormal operation of the power generation control system, and effectively eliminate safety hazards.

[0021] (3) Multi-input model predictive control can change the existing parallel system to be applied to the control of the actual system in the form of offline, static and auxiliary, so that the artificial system can play a leading role in the control system.

[0022] (4) The virtual system can continuously optimize and adjust itself based on the data provided by the actual system to achieve consistency with the actual system. Compared with traditional digital model simulation, the parallel system adopts a combination of virtual and real methods, which can simulate the working conditions of complex intelligent systems to a greater extent, provide real-time status monitoring, and obtain the verified optimal control scheme through simulation verification based on the mathematical model.

[0023] (5) Compared with digital twin technology, the parallel system-based approach is a technology that can be implemented without hardware devices.

[0024] (6) The control method using the model prediction method is a low-dimensional matrix operation method, which has much fewer parameters than the traditional deep learning. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is the power generation control framework diagram of the method. DETAILED DESCRIPTION

[0026] The power generation control method of the distributed parallel multi-layer model prediction control proposed by the present application is described in detail as follows in combination with the drawings:

[0027] Figure 1 is the power generation control framework diagram of the method. First, the model prediction control module receives the regional control deviation, epidemic prevention and health management information, weather information, four-hour model prediction load value, day-ahead scheduling and frequency deviation feedback from the frequency response module from each region. Then, the model prediction control module performs prediction optimization on all inputs, and sends the predicted power ΔP i,k to the generator, and the power P i,k emitted by the generator and the power ΔP si,k of the actual load are summed as P i,k总 is sent to the frequency response module, and the frequency response module sends the frequency deviation Δf i,k back to the model prediction control module. Finally, four mirror virtual systems identical to the actual system are simultaneously run, and the optimal state matrix A i,k , input matrix B i,k , output matrix C i,k , and disturbance matrix D i,k are calculated and experimented, and are mutually corrected with the actual system.

[0028] The above only describes the preferred embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation using the content of the specification and drawings, or direct or indirect application in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A power generation control method of distributed parallel multi-layer model predictive control, characterized by, The distributed parallel system and the multi-layer input model predictive control are combined for power generation control, and the state matrix, the input matrix, the output matrix and the disturbance matrix of a virtual control system predictive model are used to calculate the optimal control sequence of the model, and the sequence is applied to the actual control system; The control object is optimized, the safety hidden danger is eliminated, the artificial system plays a leading role, and the optimal control scheme is obtained through simulation and simulation. The power generation control method can analyze the behavior and reaction of each mirror virtual system through the calculation experiment of the artificial system, evaluate the effect of different mathematical models, and judge whether the optimal control sequence applied to the virtual intelligent agent power generation system will appear abnormal through the mutual correction of the state matrix, the input matrix, the output matrix and the disturbance matrix of the mirror virtual system and the actual system. If abnormality occurs, the abnormal unit position information is transmitted to the virtual system controller, so that the relevant personnel of the control intelligent agent system can quickly master the various conditions of the system and take corresponding actions; In use, the steps of the i-th intelligent agent of the k-th parallel system are as follows: Step (1): four parallel systems with the same parameters as the actual system are established; Step (2): the actual system and each parallel system establish a multi-region power generation control model based on model predictive control; Step (3): Obtain the frequency deviation Δf of each area of the ith area of the kth parallel system i，k , Area control deviation ACE i-j，k , Four-hour predicted load value ΔP i，k，t＝4h , Day-ahead scheduling ΔP i，k，t＝24h , Temperature and humidity WS i，k , Rainfall JY i，k , Wind direction FX i，k , Wind speed FS i，k , Rainfall JY i，k , Solar radiation intensity QD i，k , Risk area type LX i，k , Population home management information XX i，k , Number of cases SL i，k ; Step (4): the model predictive control model of the i-th region of the k-th parallel system is constructed; The model predictive control model of the i-th region of the k-th parallel system is: where x i，k is a state variable, u i，k is an input variable, w i，k is a disturbance variable, A i，k is a state matrix of the system, B i，k is an input matrix of the system, D i，k is a disturbance matrix of the system; denotes differentiation; where x i，k is: x i，k (t) = [Δf i，k , ACE i-j，k , ΔP i，k，t＝4h , ΔP i，k，t＝24h , WS i，k , JY i，k , FX i，k , FS i，k , JY i，k , QD i，k , LX i，k , XX i，k , SL i，k ] T (2) Step (5): output the control amount of the model predictive control of the i-th region of the k-th parallel system; y i，k (t) = C i，k x i，k (t) (3) where C i，k is the output matrix of the system, y i，k is the power ΔP i，k sent to the i-th generator of the k-th parallel system Step (6): Compare the frequency deviation Δf of the four parallel systems. i，k Where k = 1, 2, 3, 4, choose Δf i，k A corresponding to the minimum value of the parallel system i，k B i，k C i，k D i，k It is passed to the actual system as the control parameters of the actual system.

Images