Wind-light-diesel-V2G micro-grid active frequency random model prediction control method

Abstract

The invention discloses a wind-light-diesel-V2G micro-grid active frequency random model prediction control method, and belongs to the technical field of micro-grid frequency control. The invention aims to provide a wind-light-diesel-V2G micro-grid active frequency random model prediction control method for solving a micro-grid frequency control problem by combining an active frequency random model prediction secondary frequency control method of an unknown input decoupling observer. The method comprises the following steps: establishing a charging / discharging power model considering the charging demand of an EV user, establishing a micro-grid model containing a wind-light-diesel charging pile, establishing an unknown input observer of an improved Luenberger observer, and predicting and controlling secondary frequency optimization control based on a scene tree random model. According to the invention, the control effect under the uncertain wind-light load condition is improved. The micro-grid frequency control effect of the micro-grid under the uncertain conditions of wind, light and charging piles is improved, and large-scale replacement of distributed wind and light new energy and further popularization of electric vehicles are facilitated.

Description

Technical Field

[0001] The invention belongs to the technical field of microgrid frequency control. Background Art

[0002] As a relatively small-scale independent power system, microgrid is mainly composed of distributed energy resources such as solar energy and wind energy, as well as energy storage equipment, and has the ability to operate independently in a specific area. It can be connected to the main power grid, and can also operate independently when necessary to ensure the stability of power supply. The main advantages of microgrids include achieving energy independence by using renewable energy, improving the reliability and flexibility of power supply. Applicable to a variety of environments, such as rural areas, islands and industrial parks, microgrids can provide clean, sustainable and safe energy solutions. With the widespread application of new energy, the development of microgrids is also accelerating. At the same time, the popularity of electric vehicles and the advancement of energy storage technology have made V2G (Vehicle to Grid) technology receive more attention. This technology enables electric vehicles to achieve two-way energy exchange with the power grid. This means that electric vehicles are not only consumers of electricity, but can also feed back the electricity stored in the battery in the car to the power grid. This two-way energy flow makes electric vehicles an energy supplier.

[0003] To ensure the frequency stability of microgrids, it is necessary to design an effective frequency controller. However, due to the small inertia of the power supply in the microgrid, the large frequency fluctuation and the limited frequency regulation capability, this has become a continuous challenge facing the industry. Especially in wind, solar and electric vehicle microgrid systems, the power generation stability of new energy is not high, the randomness is large, and it is greatly affected by weather conditions, making it difficult to provide continuous and stable power output; at the same time, the charging demand of electric vehicles varies greatly, resulting in increased load fluctuations. This random power fluctuation of new energy and electric vehicles itself increases the instability of frequency. These random factors bring great difficulty to the design of frequency regulation control. In this highly random environment, the effect of traditional frequency control methods is usually not ideal.

[0004] Patent CN110021960A discloses a voltage and frequency control method for an islanded microgrid based on model predictive control. This control method is divided into three main parts: model prediction, rolling optimization, and feedback correction. In the model prediction link, it uses the filter voltage and current state equations based on the dq0 coordinates, as well as the prediction of the load current. The optimization objective function takes into account the frequency and voltage of the microgrid in the island mode. Through the rolling optimization process, the method finds the optimal control variables that can minimize the voltage and frequency deviations on the grid side within the prediction time step. The feedback correction link corrects the prediction model by collecting the grid status at the current moment, thereby avoiding control deviations caused by model mismatch and environmental interference.

[0005] Patent CN115065070A discloses a frequency control method for microgrids based on robust H∞ control. The microgrid environment contains electric vehicle charging stations and a high proportion of renewable energy. Maintaining frequency stability becomes challenging due to the uncertainty of renewable energy output and load fluctuations. This method addresses frequency deviations caused by uncertainty in the microgrid by designing a central controller based on H∞ control. By controlling the micro gas turbine to eliminate the difference between active output and load, the frequency of the microgrid is kept within an acceptable range. The design feature of the controller is a frequency design method based on a state space model, which effectively solves the frequency stability problem caused by model and disturbance uncertainty, and has theoretical and engineering value.

[0006] Patent CN115333160A discloses a robust frequency control method for a microgrid that combines wind turbines and electric vehicles. This includes establishing a frequency modulation small signal model based on the load shedding control of the wind turbine, building a unified control model for electric vehicles using an energy aggregator, and establishing a dynamic frequency response model for the microgrid based on these two models. The design goal is to determine the performance weighting function and the control weighting function, and use the linear matrix inequality method to solve the H-∞ robust frequency controller. This method effectively utilizes the frequency modulation capabilities of wind turbines and electric vehicles, and uses H-∞ robust control to suppress the negative impact of uncertainty and external disturbances on frequency regulation, thereby reducing the frequency disturbance of the microgrid under power fluctuations.

[0007] In summary, as the proportion of wind power and photovoltaic power generation capacity in microgrids increases, their power is affected by factors such as weather and presents strong randomness. At the same time, the access of electric vehicles to microgrids also brings strong randomness. The introduction of these randomnesses brings new challenges to microgrid frequency control. Summary of the invention

[0008] The purpose of the present invention is to propose a wind-solar-diesel-V2G microgrid active frequency stochastic model predictive control method that combines an active frequency stochastic model predictive secondary frequency control method with an unknown input decoupling observer to solve the microgrid frequency control problem.

[0009] The steps of the present invention are: S1. Establish a charging / discharging power model that takes into account the charging needs of EV users and establish a microgrid model that includes wind, solar, and diesel charging piles. S11. The real-time energy and charging / discharging power boundaries and real-time status of a single EV Output power limit of single EV: Where t enter With t leave is the time point when EV enters and leaves the charging station, P c With Pd is the charge / discharge power of a single EV; the energy boundary of a single EV: The overall control margin of the charging station: S12. Photovoltaic unit output probability model Probability density function: Where: I t is the solar radiation during period t; μ t ,σ t are the mean and variance of the lognormal random variable in period t respectively; The output of photovoltaic power station in period t W s,t and solar radiation I t The correlation is expressed as: Where: I std is the solar radiation under standard conditions; r c is a specific solar radiation; W sr is the rated output power of the PV array; S13. Probabilistic model of distributed wind turbine output Three-parameter Burr distribution function of wind speed probability density function Where: v t is the wind speed during period t; a t ,c t ,k t are the scale parameter, first shape parameter and second shape parameter of the Burr distribution in period t respectively; Wind turbine output W in period t w,t With wind speed v t The correlation is expressed as Where: v wr v in v out are the rated wind speed, cut-in wind speed and cut-out wind speed of the fan respectively; W wr is the rated output power of the fan; the load fluctuation can be expressed by the normal distribution function shown in the following formula Where: p is the standard deviation of load active power; μ p is the mean value of active power; S14. Establishment of LFC with wind-solar-diesel-V2G microgrid model S141, Diesel Generator Set The transfer function of the speed regulator is: Where: K n is the speed regulator gain; T g is the speed regulator time constant; The model of the prime mover is represented by the first-order inertia link: Where: K Y is the turbine gain; T Y is the steam capacity time constant; Power relationship between prime mover and generator: Where: ΔP m is the mechanical power deviation of the prime mover output; ΔP e is the electromagnetic power deviation transmitted by the generator to the grid; Δω is the rotor angular velocity deviation of the generator; M is the inertia constant of the unit; By Laplace transforming equation (15), we can obtain the power relationship in the frequency domain: ΔP m (s)-ΔP e (s)=MsΔω (16) Electromagnetic power deviation ΔP e Deviation from load power ΔP Load The relationship between ΔP e (s) = ΔP Load (s)+DΔω (17) Where: ΔP Load is the external load disturbance; D is the load damping coefficient; Combining (16) and (17): S142, wind power and photovoltaic unit model Control to output transfer function: Only consider the relationship between fan input and active power output S2. Establish unknown input observer of improved Lumberg observer The state equation of the microgrid model is written in the following state equation form: Among them A ob is the state matrix of the system, B obis the control matrix of the system, excluding unknown disturbance terms, C ob is the output matrix of the system, E ob is the matrix of unknown input perturbations, x k is the state vector, y k is the output vector, u k is a known input vector, d k is the equivalent load disturbance, both are discretized matrices; The state estimation error vector is defined as in is the state estimate; Observer designed based on dynamic system z k =F*z k-1 +TBu k-1 +K*y k (twenty three) Where: z k is the state of the observer, F, T, K, H are the designed matrices; Extensions get e k Become a Me k Function The following equation is derived 0=(HC-I)E (27) T=I-HC (28) F=A-HCA-K 1 C (29) K 2 =FH (30) S3, Secondary frequency optimization control based on scenario tree random model predictive control Typical scenarios for uncertain power ΔP={ΔP 1 ,ΔP 2 ..........ΔP s} (31) Assume that the uncertain power of new energy is obtained based on the actual measured data at the current time k as ΔP(K) = ΔP g , the uncertain power at time k+1 ΔP(K+1)=ΔP h As a random process, the prediction model is established. Then the element ΔP in the state transition probability matrix T(k) of this process is ghThe meaning of (k) is: the uncertain power at time k is ΔP g Under the condition of, the uncertain power at time (k+1) is ΔP h The probability of The microgrid LFC model is organized into a state space form Where: x is the state variable, u is the control variable, d is the load disturbance, and y is the output variable; A c , B c , C c , D c They are the state constant matrix, control constant matrix, disturbance constant matrix and output constant matrix of the continuous state equation respectively; The objective function is selected as quadratic J = min{∑(ζΔz T RΔz)-∑(ζΔu T GΔu)} (33) In the formula: In the formula: ζ is the probability of each scene in the scene tree, R and G are the weighted matrices of the objective function; Convert the constraints of each distributed power source into inequality form: CuΔU(k)≥b(k+1|k) (34) Define new variables, namely: The optimization problem then becomes minρ T ρ, by ρ T ρ=(Az-b) T The extreme value condition of (Az-b) can be derived to obtain the extreme value solution: Z * =(A T A) -1 A T b (36) The optimal solution of the control sequence at time k is: ΔU * (k) = K mpc (0-Y p (k+1|k)) (36) Take the first element in the optimal control sequence as output.

[0010] The present invention mainly proposes an unknown decoupled input observer based on the Lumberg observer to realize the decoupled observation of the state quantity and the unknown input disturbance to improve the observation accuracy; proposes a scene tree random model predictive controller, through rolling optimization and feedback correction, and uses random model predictive control to obtain the optimal frequency control quantity under various constraints, thereby improving the control effect under the uncertain conditions of wind and solar load. The proposed technology improves the frequency control effect of microgrids under the uncertain conditions of wind, solar and charging piles, and helps the large-scale replacement of distributed wind and solar new energy and the further popularization of electric vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is the boundary and real-time status diagram of the real-time energy and charging and discharging power of a single EV in the power station; Figure 2 This is the microgrid framework diagram; Figure 3 It is a load frequency control model diagram of a single-region power system; Figure 4 This is the microgrid frequency control model diagram; Figure 5 Schematic diagram of the unknown input observer; Figure 6 It is the control structure diagram of the microgrid LFC model; Figure 7 Optimize the flow chart for the microgrid LFC model; Figure 8 This is the comparison chart of the load disturbance and estimated load disturbance of the microgrid in Case 1; Fig. 9 This is a comparison chart of SMPC and PI control effects in case 1; Fig.10 This is a comparison chart of various disturbances and estimated load disturbances in the microgrid in Case 2; Fig.11 This is a comparison chart of the control effects of SMPC and PI in Case 2. DETAILED DESCRIPTION

[0012] The purpose of this invention is to propose a microgrid active random model predictive control method, which takes into account the presence of small generators such as diesel engines, distributed power sources such as photovoltaic wind power, and electric vehicles with V2G functions in the microgrid. The traditional proportional-integral control method will reduce the control effect due to the influence of randomness, and the output utilization rate of new energy units is not high and cannot meet the system frequency control performance requirements.

[0013] The steps of the present invention are: Step 1: Establish a charging / discharging power model that takes into account the charging needs of EV users and establish a microgrid model that includes wind, solar, and diesel charging piles; Based on the model of a single EV, the energy boundary and power boundary of charging and discharging are obtained, and then the overall output power boundary of the charging station is obtained. At the same time, the charging demand of electric vehicles and the life of electric vehicle batteries are considered to obtain the power model of the EV power station. According to statistical laws, a random mathematical model of wind and solar load power is established, and a microgrid model in the form of a transfer function is built.

[0014] Step 2: Establish the unknown decoupled input observer of the improved Lumberg observer; Traditional observers use all input values, including unknown input disturbances, when performing state estimation. Therefore, there is a large deviation in the observed value, which leads to errors in the final estimated value. Based on the traditional Lumberg observer, when designing the feedback matrix, the system state variables and unknown disturbance variables contained in the error value function are set to zero to obtain the design matrix of the observer. At this time, the error value function does not contain system state variables and disturbance variables, realizing the decoupling of observed values ​​and disturbance variables and improving observation accuracy.

[0015] Step 3: Establish a random model predictive control quadratic frequency optimization control algorithm based on the scenario tree; There are random fluctuations in the power of wind and solar electric vehicles. Conventional PI control cannot cope with high randomness and the control effect is poor. The present invention generates a scenario tree model of its uncertain power in real time, thereby obtaining a series of typical scenarios and corresponding probabilities; the above models and scenarios are applied to the random model pre-controller. According to the control target and considering multiple types of control constraints, the active power instructions of each station are obtained and issued in real time, and finally the secondary frequency regulation target is achieved.

[0016] Step 1: Establish a charging / discharging power model that takes into account the charging needs of EV users and establish a microgrid model containing wind, solar and diesel charging piles. (1) Charging Station According to the real-time energy and charge / discharge power boundary and real-time status of a single EV in the charging station, Figure 1 As shown. The output power boundary of a single EV can be obtained, as shown in equations 1 and 2. Based on this, the energy boundary of a single EV can be obtained, as shown in equations 3 and 4. Furthermore, based on the energy and power boundary of a single EV, the real-time status of the EV in the charging station is combined with the entry time to obtain the overall control margin of the charging station, as shown in equations 5 and 6: In formula (1) and formula (2), t enter With t leave is the time point when EV enters and leaves the charging station, P c With P d is the charge / discharge power of a single EV;

[0017] In summary, based on single EV i With EV j As an example, EV i When entering the station, the 0 <E min Status (E 0 is the initial energy of the electric vehicle when it enters the charging station, E min To maintain the energy limit of electric vehicle battery life, EV j When entering the station, the 0 >E min state. From this, the energy and power boundaries of the EV cluster participating in the microgrid frequency regulation can be obtained. The overall output power boundary of the charging station can also be obtained. Based on this, when the frequency fluctuates, the charging power of the EVs in the EV charging station will also change when participating in the frequency regulation, which will inevitably affect the charging process.

[0018] (2) Photovoltaic unit output probability model As an important factor affecting the uncertainty of photovoltaic output, solar radiation usually follows a log-normal distribution. Its probability density function is described as follows: Where: I t is the solar radiation during period t; μ t ,σ t are the mean and variance of the lognormal random variable in period t, respectively.

[0019] The output of photovoltaic power station in period t W s,t and solar radiation I t The correlation can generally be expressed as: Where: I std is the solar radiation under standard conditions; r c is a specific solar radiation; W sr is the rated output power of the PV array.

[0020] (3) Output probability model of distributed wind turbines The uncertainty of wind turbine output is largely attributed to the random fluctuation of wind speed in nature. The probability density function of wind speed is usually expressed by the two-parameter Weibull distribution function, which can fit the medium- and long-term wind speed distribution well, but is slightly insufficient for studying the short-term wind speed probability distribution. Therefore, this paper uses the three-parameter Burr distribution function, which has been gradually used to describe the wind speed probability density function in recent years, to fit the wind speed distribution. The model is slightly complex but the fitting effect is good, which is specifically expressed as Where: v t is the wind speed during period t; a t ,c t ,k t are the scale parameter, first shape parameter and second shape parameter of the Burr distribution in period t respectively.

[0021] Wind turbine output W in period t w,t With wind speed v t The correlation can generally be expressed as Where: v wr v in v out are the rated wind speed, cut-in wind speed and cut-out wind speed of the fan respectively; W wr is the rated output power of the fan.

[0022] Load fluctuation can be expressed by the normal distribution function shown below Where: p is the standard deviation of load active power; μ p is the mean value of active power.

[0023] (2) Establishment of LFC including wind-solar-diesel-V2G microgrid model The microgrid built includes a diesel engine, a photovoltaic power station, a wind turbine, an electric vehicle charging station and loads; Figure 2 As shown, the microgrid operates in an islanded manner.

[0024] The electric vehicle charging station model is established according to the model in the previous section, and the remaining power supply models are as follows: (1) Diesel generator set As the most basic component in the power system, the speed regulator is indispensable in both primary and secondary frequency regulation. The transfer function expression of the speed regulator is: Where: K n is the speed regulator gain; T g is the speed regulator time constant.

[0025] In traditional power systems, steam turbines are generally selected as prime movers. We use the first-order inertia link to represent the mathematical model of the prime mover, namely: Where: K Y is the turbine gain; T Y is the steam capacity time constant.

[0026] The power relationship between the prime mover and the generator can be expressed as: Where: ΔP m is the mechanical power deviation of the prime mover output; ΔP e is the electromagnetic power deviation delivered by the generator to the grid; Δω is the rotor angular velocity deviation of the generator; M is the inertia constant of the unit.

[0027] By performing Laplace transform on equation (15), we can obtain the power relationship in the frequency domain, namely: ΔP m (s)-ΔP e (s)=MsΔω (16)

[0028] Electromagnetic power deviation ΔP e Deviation from load power ΔP Load There is also a certain relationship between them, as shown below: ΔP e (s) = ΔP Load (s)+DΔω (17) Where: ΔP Load is the external load disturbance; D is the load damping coefficient.

[0029] Combining (16) with (17), we can obtain:

[0030] In summary, by connecting the above parts, we can get a complete load frequency control model of a single-region power system, such as Figure 3 shown.

[0031] (2) Wind and photovoltaic unit model The photovoltaic power generation model uses the state averaging method to obtain the mathematical model of photovoltaics, that is, the transfer function from control to output:

[0032] The structure of the wind turbine is more complicated than that of the photovoltaic power plant, and it is more troublesome to derive the transfer function formula of the wind turbine. Therefore, we simplify the wind turbine transfer function into a first-order inertia link and only consider the relationship between the wind turbine input and the active power output, namely:

[0033] According to the equivalent model established above, the frequency control model of the microgrid can be constructed, such as Figure 4 shown.

[0034] Step 2: Establish unknown input observer of modified Lumberg observer The model predictive controller requires the current system state as the initial value for predictive control, but the system state quantity is often not directly measurable, so a state observer is needed for state measurement. There are a large number of uncertain disturbances in microgrids, such as wind and solar load disturbances, which pose challenges to system state observation. The Kalman filter commonly used in the model predictive controller is used for state observation. However, the Kalman filter uses all input values, including unknown input disturbances, when estimating the state, so its observed value has a large deviation, which leads to errors in the final estimated value.

[0035] In order to improve the observation accuracy under the uncertainty of wind and solar load and improve the impact of unknown input disturbance on system state observation, we adopt an unknown input observer that decouples the state estimation from the unknown input disturbance term in the original system. The schematic diagram is shown in Figure 5 shown.

[0036] The state equation of the microgrid model is written in the following state equation form: Among them A ob is the state matrix of the system, B ob is the control matrix of the system, excluding unknown disturbance terms, C ob is the output matrix of the system, E ob is the matrix of unknown input perturbations, x k is the state vector, y k is the output vector, u k is a known input vector, d k is the equivalent load disturbance, both are discretized matrices.

[0037] The state estimation error vector is defined as in is the state estimate.

[0038] The observer equation designed according to the dynamic system is shown as follows: z k =F*z k-1 +TBu k-1 +K*y k (twenty three) Where: z k is the state of the observer, and F, T, K, and H are designed matrices for decoupling unknown inputs. It can be seen that UIO is essentially a dynamic system that decouples state estimation from the disturbance term in the original system.

[0039] Extensions We get

[0040] It can be seen that if we want to observe the decoupling state, e k Need to be a Me k The function of

[0041] The following equation can be derived 0=(HC-I)E (27) T=I-HC (28) F=A-HCA-K 1 C (29) K 2 =FH (30)

[0042] If all eigenvalues ​​of M are negative, then e(k) will gradually approach zero. The value of the corresponding matrix is ​​obtained. At this time, equation (25) is not E or d k The function of achieves the desired decoupling of state estimation and unknown disturbance input.

[0043] Step 3: Secondary frequency optimization control based on scenario tree stochastic model predictive control Traditional PID control cannot meet the needs of microgrid frequency control, and robustness is difficult to match. Traditional MPC control is generally ineffective in microgrids with high penetration of random new energy sources and cannot meet the frequency control requirements. Based on the Markov chain principle, the present invention designs a scenario tree model to predict the uncertain power changes of new energy sources in microgrids.

[0044] The uncertainty power ΔP of new energy is defined as the difference between the actual output at the current moment and the predicted value. Existing studies generally believe that the probability distribution of output prediction error in a short time scale is close to a normal distribution, and the error size is mainly distributed within 5% of the rated power.

[0045] Based on this, the following steps can be used to select typical scenarios with uncertain power: 1) In order to ensure that the sampling samples cover the entire space of variables, the Latin hypercube sampling method is used to sample the probability distribution of ΔP to obtain a series of sample scenarios and their probabilities of ΔP.

[0046] 2) Since the total number of samples obtained by sampling is generally large, in order to improve the calculation efficiency, the k-means clustering algorithm is used to reduce the sampled scenes, and the typical scenes of uncertain power are as follows: ΔP={ΔP 1 ,ΔP 2 ..........ΔP s} (31)

[0047] Assume that the uncertain power of new energy is obtained based on the actual measured data at the current time k as ΔP(K) = ΔP g The uncertain power at time k+1 is ΔP(K+1)=ΔP h As a random process, the prediction model is established. Then the element ΔP in the state transition probability matrix T(k) of this process is gh The meaning of (k) is: the uncertain power at time k is ΔP g Under the condition of, the uncertain power at time (k+1) is ΔP h In practical applications, the value of the state transition probability matrix can be obtained with the help of historical prior data.

[0048] Taking the current measurement state as the initial value, all scenarios and their probabilities for multi-step prediction can be enumerated. However, since too many scenarios may cause the "curse of dimensionality" problem and reduce computational efficiency, this paper uses a scenario tree design to simplify the process of gradually refining the prediction scenarios as the prediction time domain increases.

[0049] Furthermore, the microgrid LFC model is organized into a state space form as shown below: Where: x is the state variable, u is the control variable, d is the load disturbance, and y is the output variable; A c , B c , C c , D c They are the state constant matrix, control constant matrix, disturbance constant matrix and output constant matrix of the continuous state equation respectively.

[0050] Performance index is an important criterion for judging the performance of the controller. Generally, some objective functions are set and solved to obtain extreme values ​​to determine the optimal solution. The objective function of the microgrid frequency regulation in this paper is to make the difference between the expected value and the actual output value and take the minimum value for optimization, and also to ensure that the output of the microgrid can quickly track the expected value. Its objective function generally has three forms of expression: norm-type performance indicator function expression, quadratic performance indicator function expression and infinite norm performance indicator function expression. The objective function form selected in this paper is quadratic. When the microgrid frequency fluctuates, MPC needs to adjust the system frequency deviation Y p,c The controller issues a control instruction ΔU to restore the frequency to the rated value when there is a frequency error. The specific expression is: J = min{∑(ζΔz T RΔz)-∑(ζΔu T GΔu)} (33) In the formula: In the formula: ζ is the probability of each scene in the scene tree, R and G are the weighted matrices of the objective function. The control structure is as follows Figure 6 shown.

[0051] In model predictive control, the core is to transform the frequency control challenge in microgrid system into an optimization problem. This process involves setting an objective function and considering the system constraints to find the best solution within these constraints. In particular, considering the stochastic nature of wind and solar output, the frequency control problem of microgrid system is usually transformed into a planning problem, and then the optimization result is sought through mathematical methods.

[0052] The key components of optimization problems include objective functions, constraints, and decision variables. These problems are usually divided into three categories: linear programming, quadratic programming, and quadratic programming with quadratic constraints. This study focuses on solving the frequency control problem of restoring the frequency of the microgrid to the standard value under the premise of satisfying the power constraints of each distributed power source in the microgrid. Due to the constraints in the system, it is usually impossible to directly obtain the optimal solution of the objective function. Therefore, it is necessary to convert the SMPC (stochastic model predictive control) optimization problem with variable constraints into a quadratic programming problem to solve it.

[0053] Convert the constraints of each distributed power source into inequality form: C u ΔU(k)≥b(k+1|k) (34)

[0054] In the model to be optimized in this paper, the optimization variable ΔU represents the power of each distributed power source; the inequality constraints include: the backup power constraint of the diesel engine, the charging and discharging constraint of the energy storage system, the variable constraint of the photovoltaic array, and the variable constraint of the wind turbine.

[0055] In order to solve the optimal problem in the form of quadratic programming, new variables are defined, namely:

[0056] The optimization problem then becomes minρ T ρ, by ρ T ρ=(Az-b) T The extreme value condition of (Az-b) can be derived to obtain the extreme value solution: Z * =(A T A) -1 A T b (36)

[0057] It can be known that the solution to the minimum value, that is, the optimal solution of the control sequence at time k is: ΔU * (k) = K mpc (0-Y p (k+1|k)) (37) Take the first element in the optimal control sequence as the output. Figure 7 shown.

[0058] In order to verify the correctness of the proposed control strategy, a simulation model of an isolated wind-solar-diesel microgrid is established. The control methods adopt PI control and the proposed SMPC control. Two different scenarios are designed.

[0059] Case 1 is a sudden load increase. Assuming that the illumination of the photovoltaic array is constant and the wind speed of the wind turbine is unchanged, the load increases by 0.018 pu in 10 seconds. Figure 8 The solid line in the middle represents the actual load fluctuation, and the dotted line represents the observation and identification of the disturbance by the random observer. When the load changes, the prediction requires the system output value, so there will be a certain time lag in the prediction, but it will eventually be consistent with the actual value. The secondary frequency regulation output of PI control is allocated according to the unit capacity, and each power generation unit outputs in proportion. Compared with PI secondary frequency regulation control, the proposed SMPC control method combines constraints and weights, and coordinates the output of each power generation unit through rolling optimization. It can control the unit output more accurately. At the same time, due to the improvement of observation accuracy, the prediction accuracy is increased, further improving the control effect. Fig. 9 The proposed control method can change the unit output more quickly after a sudden load change, reduce the frequency fluctuation value, and shorten the frequency recovery time.

[0060] Case 2 is about the random changes in wind, light and load. In actual work, wind speed, light and system load are in a fluctuating state. In some severe weather conditions, wind speed and light may change rapidly, such as rapid obstruction by clouds and sudden changes in wind speed. This scenario imports 3 minutes of actual wind speed, light data and load fluctuation data into the simulation system. Fig.10 Black represents load disturbance, blue represents photovoltaic power fluctuation, and orange represents wind turbine power fluctuation. It can be seen from the figure that when wind and light fluctuate rapidly, the observer will also observe the general trend of the disturbance waveform. It can improve the control accuracy of MPC, thereby improving the control effect. Compared with PI control, the proposed SMPC control method combines the probability model, and can adjust the unit output more quickly through rolling optimization and feedback correction. At the same time, the proposed unknown input decoupling observer can also maintain the accuracy of the observation results in the case of large uncertainty disturbances, thereby improving the control effect. It makes it possible to quickly restore the frequency to stability in areas with large fluctuations such as 30s and 100s.

[0061] The present invention proposes a random model predictive control method with an unknown input decoupling observer, which effectively improves the frequency control effect of the microgrid system under uncertainty conditions and has practical application value for secondary frequency control of microgrids.

[0062] Variables involved in the formula: The time point t at which the EV enters and leaves the charging station i,enter , t i,leaver ; Single EV charging / discharging power p c ,p d ; The initial energy E of the electric vehicle when it enters the charging station 0 ; Maintain the energy lower limit E of electric vehicle battery life min ; Solar radiation I during period t t ; The mean and variance μ of the lognormal random variable in period t t ,σ t ; Output of photovoltaic power station W during period t s,t ; Solar radiation I t ; Solar radiation under standard environment I std ; A specific solar radiation r c ; Rated output power of the photovoltaic array W sr ; Wind speed v during period t t ; The scale parameter, the first shape parameter, and the second shape parameter a of the Burr distribution t ,c t ,k t ; Wind turbine output W during period t w,t ; Wind speed v t ; Rated wind speed of fan v wr ; Fan cut-in wind speed v in ; Fan cut-out wind speed v out ; Rated output power of fan W wr ; Standard deviation of load active power σ p ; Mean value of active power μ p ; Speed ​​governor gain K n ; Speed ​​regulator time constant T g ; is the turbine gain G Y ; is the turbine volume time constant T Y ; Mechanical power deviation ΔP of the prime mover output m ; The electromagnetic power deviation ΔP transmitted by the generator to the grid e ; Generator rotor angular velocity deviation Δω; Unit inertia constant M; Electromagnetic power deviation ΔP e ; Load power deviation ΔP Load ; External load disturbance Δω; Load damping coefficient D; G pv Photovoltaic output power; wind turbine output power G wt ; The state matrix A of the microgrid model system ob ; Control matrix B of microgrid model system ob ; Output matrix C of microgrid model system ob ; Unknown input disturbance matrix E of microgrid model system ob; Microgrid model system state vector x k ; Microgrid model system output vector y k ; The known input vector u of the microgrid model system k ; Microgrid model system equivalent load disturbance d k ; State estimation error vector e k ; State estimate Observer state z k ; Observer matrix F, T, K, H; Uncertain power ΔP of new energy; Microgrid LFC model state variable x; Microgrid LFC model control variable u; Microgrid LFC model load disturbance d; Microgrid LFC model output variable y; State constant matrix A of continuous state equation of microgrid LFC model c ; Control constant matrix B of continuous state equation of microgrid LFC model c ; The disturbance constant matrix C of the continuous state equation of the microgrid LFC model c ; Output constant matrix D of continuous state equation of microgrid LFC model c ; System frequency deviation Y p,c ; System output control quantity ΔU; The probability ζ of each scene in the scene tree; The weighted matrices R and G of the objective function.

Claims

1. A wind-solar-diesel-V2G microgrid active frequency stochastic model predictive control method, characterized by: The steps are: S1. Establish a charging / discharging power model that takes into account the charging needs of EV users and establish a microgrid model that includes wind, solar, and diesel charging piles S11. Real-time energy and charge / discharge power boundaries and real-time status of single EV Output power limit of single EV: （1） （2） Where t enter With t leave is the time point when EV enters and leaves the charging station, P c With P d is the charge / discharge power of a single EV; Energy boundary of single EV: （3） （4） The overall control margin of the charging station: （5） （6） S12. Photovoltaic unit output probability model Probability density function: （7） （8） Where: is the solar radiation during period t; are the mean and variance of the lognormal random variable in period t respectively; Photovoltaic power station output during period t Solar irradiance The correlation is expressed as: （9） Where: is the solar radiation under standard environment; is a specific solar irradiance; is the rated output power of the PV array; S13. Probabilistic model of distributed wind turbine output Three-parameter Burr distribution function of wind speed probability density function （10） Where: is the wind speed during period t; are the scale parameter, first shape parameter and second shape parameter of the Burr distribution in period t respectively; Wind turbine output during period t With wind speed The correlation is expressed as （11） Where: They are the wind turbine rated wind speed, cut-in wind speed and cut-out wind speed respectively; is the rated output power of the fan; Load fluctuation can be expressed by the normal distribution function shown below （12） in: is the standard deviation of load active power; is the mean value of active power; S14. Establishment of LFC with wind-solar-diesel-V2G microgrid model S141, Diesel Generator Set The transfer function of the speed regulator is: （13） Where: K n is the speed regulator gain; T g is the speed regulator time constant; The model of the prime mover is represented by the first-order inertia link: （14） Where: K Y is the turbine gain; T Y is the steam capacity time constant; Power relationship between prime mover and generator: （15） Where: ΔP m is the mechanical power deviation of the prime mover output; ΔP e is the electromagnetic power deviation transmitted by the generator to the grid; Δω is the rotor angular velocity deviation of the generator; M is the inertia constant of the unit; By Laplace transforming equation (15), we can obtain the power relationship in the frequency domain: （16） Electromagnetic power deviation ΔP e Deviation from load power ΔP Load The relationship between （17） Where: ΔP Load is the external load disturbance; D is the load damping coefficient; Combining (16) and (17): （18） S142, wind power and photovoltaic unit model Control to output transfer function: （19） Only consider the relationship between fan input and active power output （20） S2. Establish unknown input observer of improved Lumberg observer The state equation of the microgrid model is written in the following state equation form: （21） in is the state matrix of the system, is the control matrix of the system, excluding unknown disturbance terms, is the output matrix of the system, is the matrix of unknown input perturbations, is the state vector, is the output vector, is a known input vector, is the equivalent load disturbance, both are discretized matrices; The state estimation error vector is defined as （22） in is the state estimate; Observer designed based on dynamic system （23） （24） Where: is the state of the observer, F, T, K, H are the designed matrices; Extensions get （25） Become a Function （26） The following equation is derived （27） （28） （29） （30） S3, Secondary frequency optimization control based on scenario tree random model predictive control Typical scenarios for uncertain power （31） Assume that at the current moment k, the uncertain power of new energy is obtained according to the actual measurement data: , the uncertain power at time k+1 As a random process, the prediction model is established. Then the elements in the state transition probability matrix T(k) in this process are The meaning is: the uncertain power at time k is Under the condition of, the uncertain power at time (k+1) is probability; The microgrid LFC model is organized into a state space form (32) Where: x is the state variable, u is the control variable, d is the load disturbance, and y is the output variable; A c , B c , C c , D c They are the state constant matrix, control constant matrix, disturbance constant matrix and output constant matrix of the continuous state equation respectively; The objective function is selected as quadratic （33） In the formula: In the formula: is the probability of each scene in the scene tree, R and G are the weighted matrices of the objective function; Convert the constraints of each distributed power source into inequality form: （34） Define new variables, namely: （35） The optimization problem then becomes min ρ T ρ, by ρ T ρ = (Az-b) T The extreme value condition of (Az-b) can be derived to obtain the extreme value solution: （36） The optimal solution of the control sequence at time k is: （36） （37） Take the first element in the optimal control sequence as output.

Images