CN110190599B - A Control Method of Island Microgrid Based on Finite Time Consistency Theory - Google Patents

Abstract

The invention discloses a method for designing an island micro-grid control strategy based on a finite time consistency theory, and belongs to the field of intelligent grid control. In an island micro-grid based on a distributed secondary control strategy, a peer-to-peer sparse network is utilized to communicate with an adjacent agent, and information and local information are obtained through processing by a consistency algorithm; and feeding back the consistency deviation to a frequency and voltage controller designed by a finite time consistency strategy; then, the output of the controller is input to droop control, then transferred to a voltage and current control loop, finally frequency and voltage stabilization is realized and a nominal value is tracked; wherein, when the communication is interrupted, the historical data of the adjacent agent is predicted based on the improved extreme learning machine, and the prediction result is input to a frequency and voltage controller designed by a finite time consistency strategy. So as to realize secondary frequency and voltage regulation under communication fault.

Description

A Control Method of Island Microgrid Based on Finite Time Consistency Theory

technical field

The invention belongs to the field of smart grid control, and in particular relates to a design method of an island microgrid control strategy based on finite time consistency theory.

Background technique

With the increasingly serious problems of energy crisis and environmental pollution, the use of new energy, including distributed generation technology, has received widespread attention from the society, and microgrids have also developed rapidly.

The operation of an AC microgrid can be divided into two modes: grid-connected and islanded. When operating in island mode, use multiple distributed power sources to operate in parallel to improve output power quality. In island mode, hierarchical control is generally used to optimize power quality, and different hierarchical structures are used to define and implement different control objectives. Droop control is often used as the primary controller performing frequency and voltage regulation, and enables power distribution. However, the traditional droop control adjustment effect is not good. When the impedance of the distribution line is not uniform, the power distribution effect is poor. Some existing improvements include virtual impedance to balance line resistance, and some adaptive droop control strategies. Although there are many improved strategies for droop control, simple droop control always fails to accurately track the reference value due to inherent errors. Therefore, in order to obtain a more precise control effect, a secondary control strategy is recommended. At present, the existing secondary control strategies are generally divided into centralized and distributed, and are often used for compensation droop control. The centralized secondary control is formulated by the Microgrid Central Controller (MGCC), which collects and calculates the entire network information in the microgrid, and then issues commands to the physical layer. Although the centralized control has high precision, the communication pressure of the system is great, and the reliability and scalability are poor. Distributed secondary control does not require a central node to communicate, and most use a peer-to-peer sparse network, which only communicates with neighboring agents, so it has better scalability and feasibility. The distributed secondary control strategy can achieve frequency and voltage recovery, accurate power distribution.

However, when the communication fails and is interrupted, it will still cause the instability of the entire system.

SUMMARY OF THE INVENTION

The invention proposes a control method of island microgrid based on the finite time consistency theory, aiming at introducing the finite time theory based on the consistency theory, and combining the two, a new secondary control algorithm is proposed, and the communication is interrupted Taking into account the situation of , based on the improved extreme learning machine (ELM), the historical data of adjacent agents is predicted, and the prediction result is replaced by the information of the communication link that has been interrupted, so as to realize the prediction compensation and improve the system stability.

In order to solve the above-mentioned technical problems, the technical solution provided by the present invention is: a method for controlling an islanded microgrid based on a finite time consistency theory, characterized in that the steps are:

In the island microgrid based on the distributed secondary control strategy, the peer-to-peer sparse network is used to communicate with the adjacent agents, and the information and local information are obtained through the processing of the consensus algorithm; The frequency and voltage controller designed by the characteristic strategy; then, the output of the controller is input to the droop control, and then transferred to the voltage and current control loop, and finally the frequency and voltage are stabilized and tracked to the nominal value;

Among them, when the communication is interrupted, the historical data of adjacent agents is predicted based on the improved extreme learning machine, and the prediction results are input into the frequency and voltage controllers designed by the finite time consistency strategy.

A further technical solution is that the island microgrid is divided into two parts: a physical layer and a network layer; the physical layer mainly includes a controller and a distributed power source, the controller stabilizes the output of each distributed power source, and then adjusts the output of the distributed power source. The output is connected to the point of common coupling through a static transfer switch; or the controller stabilizes the output of each distributed power source, and then uses the distributed power source to supply power to the load; the network layer exchanges data between different power electronic inverters to complete the entire process. Output synchronization of distributed power sources within a microgrid.

A further technical solution is that the output error of the agent DG _i can be calculated as:

Among them, e _i represents the global output error of agent i, _xi represents the actual output value of agent i, x _j represents the value transmitted by the adjacent agent j of agent i through communication, and x ^ref represents the nominal value of the output of agent i , as the leader node of the consensus algorithm, a _ij represents the weighted adjacency coefficient, that is, the network communication link of agent i, b _i represents the weight leading to the leader, if and only when the agent i can access the leader, b _i >0;

In order to realize the adjustment effect of limited time, a symbolic function is introduced:

The designed finite-time controller is:

u _i =βsig(e _i ) ^α

Among them, α and β are finite-time control parameters, α is an exponential coefficient, 0<α<1, which improves the system convergence performance, β is a proportional coefficient, β>0, determines the step size of the control variable.

A further technical solution is that the steps of predicting the historical data of adjacent agents based on the improved extreme learning machine are as follows:

(1) Establishment of the prediction model

For any N different samples (X _i ,Y _i ), where X _i =[x _i1 x _i2 … x _in ] ^T ∈R ⁿ represents the historical output frequency or voltage of each DG, Y _i =[y _i1 y _i2 … y _im ] ^T ∈ R ^m represents the frequency or voltage of the expected output of each DG. For a single hidden layer neural network with L hidden layer nodes, it can be expressed as:

where g(x) is the excitation function, generally a Gaussian function, Wi = [ _wi1 w _i2 ... w _in ] ^T is the input weight, γ _i is the output weight, c _i is the bias of the _ith hidden layer unit, o _j represents the predicted output;

When considering empirical risk and structural risk, the mathematical model of ES-ELM can be expressed as:

Among them, ||ε|| ² represents the empirical risk, ||γ|| ² represents the structural risk, and η∈R represents the best compromise point determined by the two risk ratio parameters through cross-validation;

The LS-SVM algorithm is introduced to convert the ES-ELM model into the Lagrange equation as:

Among them, λ=[λ ₁ λ ₂ … λ _N ] represents the Lagrange multiplier, H represents the output of the hidden layer, and γ=[γ ₁ γ ₂ … γ _L ] ^T represents the output weight

According to the optimal conditions of KTT, find the gradient of the Lagrange equation and set it to 0, we get

thereby getting

where X ⁺ represents the generalized inverse of matrix X;

(2) Training of the prediction model

Before training, clarify the following contents: training set N different samples (X _i , Y _i ), number of hidden layers L, excitation function g(x),

Step 1: When the data changes greatly, in order to obtain better training results, the data is preprocessed. The specific formula is:

Among them, x(i) and _x '(i) represent the original data and processed data, respectively, and Ex and σ _i represent the mean and standard deviation of the original data;

Step 2: The input weights w _ik and bias c _i are arbitrarily set in the range of (0,1), and the hidden layer output H is calculated;

计算输出权值γ和拉格朗日乘子λ。Step 3: According to

Calculate the output weight γ and the Lagrange multiplier λ.

(3) Result prediction

After ES-ELM training, local prediction of DG _i that has lost communication information is carried out through historical data, and the predicted values of frequency and voltage are obtained respectively.

The present invention adopts the beneficial effects of the above technical solutions as follows:

(1) In order to achieve more accurate frequency and voltage recovery, the present invention proposes a voltage and frequency control method based on a distributed finite time consistency algorithm. A finite-time strategy and a consensus algorithm are combined to reduce the coupling of frequency and voltage caused by uneven line impedance and reach the reference value within a set finite time.

(2) Using Lyapunov to prove that the proposed finite-time consistency algorithm can accurately realize the secondary control of frequency and voltage, and accurately track the reference value. In addition, through the analysis of the Lyapunov function, the relationship between the convergence rate and parameter selection can be obtained.

(3) When the communication data is lost and the communication is interrupted, the ELM is improved by considering the empirical risk and the structural risk, which is used to train the historical data of frequency and voltage, and predict the lost communication data, and then send the prediction result to Auxiliary controller, thereby suppressing the effects of communication failures.

Description of drawings

1 is a simplified diagram of an islanded microgrid based on a distributed secondary control strategy of the present invention;

Fig. 2 is the inverter block diagram in the i-th DG of the present invention;

Fig. 3 is the control block diagram of the whole system;

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and specific embodiments:

The embodiment of the present invention describes an islanded microgrid control method based on finite time consistency theory, characterized in that the steps are:

In the island microgrid based on the distributed secondary control strategy, the peer-to-peer sparse network is used to communicate with the adjacent agents, and the information and local information are obtained through the processing of the consensus algorithm; The frequency and voltage controller designed by the characteristic strategy; then, the output of the controller is input to the droop control, and then transferred to the voltage and current control loop, and finally the frequency and voltage are stabilized and tracked to the nominal value;

Among them, when the communication is interrupted, the historical data of adjacent agents is predicted based on the improved extreme learning machine, and the prediction results are input into the frequency and voltage controllers designed by the finite time consistency strategy.

The specific content of the present invention is as follows:

(1) Design the microgrid layered control architecture

Microgrid control is mainly divided into two parts: physical layer and network layer, as shown in Figure 1. The physical layer mainly includes the controller and distributed power supply. The controller stabilizes the output of each DG, and then connects the output of the DG to the point of common coupling (PCC) through a static transfer switch (STS), so that the To supply power to the public load attached to the PCC, of course, a DG can also be used to supply power to a specific load. The network layer is mainly to realize data exchange between different power electronic inverters, and realize the synchronization of all DG outputs in the entire microgrid.

Figure 2 shows the block diagram of the inverter inside the DG. It is assumed that there are N DG units in the microgrid, and each DG is defined as DG _i ( _i =1,2,...,N) in a certain order. Connect to PCC via feeder. The controller of the inverter mainly includes: 1) a power calculator; 2) a power controller; 3) a voltage control loop, which adjusts the DG unit AC measurement voltage _voi ; 4) a current control loop, which regulates the DG unit AC measurement current i _li . The three-phase power output by DG _i is converted by dp-frame to obtain v _odi , v _oqi , i _odi , i _oqi , and then v _oi , i _li are obtained by calculation, so as to calculate the active and reactive power components output by DG _i . Power controllers perform droop control in combination with their power ratings:

是系统标称电压，

是逆变器交流测电压v_oi的参考值并作为电压电流环的参考值，P_i和Q_i分别是单元DG_i输出的有功和无功功率分量。where m _i and _ni are the active and reactive droop control coefficients of unit DG _i , ω _i is the output frequency of DG _i , ω ^ref is the system rated frequency (2π×50red/s),

is the nominal system voltage,

is the reference value of the AC voltage v _oi of the inverter and is used as the reference value of the voltage and current loop, _{Pi and Q i are the active and reactive power components output by the unit DG i respectively .}

(2) Design a limited time consistency control strategy to realize the regulation of frequency and voltage.

2.1 Algebraic graph theory

代表了能够进行信息交换的通信链路。A＝[a_ij]_N×N是加权邻接矩阵系数，a_ii＝0且a_ij≥0，当且仅当(v_i,v_j)∈ε时，a_ij＞0。第i个代理的相邻节点被定义为N_i＝{v_j∈V:(v_i,v_j)∈ε}，有向图G的隶属度矩阵(degree matrix)设为D＝diag{d₁,d₂,...,d_N}，并且

拉普拉斯矩阵Ω＝D-A是一个对称的半正定矩阵。In order to better describe the structure of the communication network of the microgrid, the algebraic directed graph theory is used here. As shown in Figure 3, the communication network in the microgrid consists of N multi-agents (DGs), labeled from 1 to N, and then mapped to a directed graph G(V,ε,A), where the node set V={v ₁ , v ₂ ,...,v _N } represents all DGs, edge sets

Represents a communication link capable of information exchange. A=[a _ij ] _N×N is the weighted adjacency matrix coefficient, a _ii =0 and a _ij ≥0, a _ij >0 if and only if (vi , v _j ) _∈ε . The adjacent nodes of the ith agent are defined as N _i ={v _j ∈V:(v _i ,v _j )∈ε}, and the degree matrix of the directed graph G is set to D=diag{d ₁ ,d ₂ ,...,d _N }, and

The Laplace matrix Ω=DA is a symmetric positive semi-definite matrix.

2.2 Finite Time Consistency Algorithm

Simple droop control cannot avoid the drop in the frequency and voltage of the whole network. Therefore, this paper proposes a consensus algorithm based on finite time, using the previous graph theory analysis, to compensate the output of distributed power, and finally make The output frequency and voltage of each agent are consistent and can continuously track to the set nominal values of frequency and voltage ω ^ref and v ^ref respectively, which improves the dynamic performance of the system.

For, the output error of this agent DG _i can be calculated with the consistency policy as:

Among them, e _i represents the global output error of agent i, _xi represents the actual output value of agent i, x _j represents the value transmitted by the adjacent agent j of agent i through communication, and x ^ref represents the nominal value of the output of agent i , as the leader node of the consensus algorithm, a _ij represents the weighted adjacency coefficient, that is, the network communication link of agent i, b _i represents the weight leading to the leader, if and only when the agent i can access the leader, b _i > 0.

In order to realize the limited time adjustment effect, a sign function is introduced.

It is guaranteed that under certain suitable conditions, the global error of the output of agent i can be eliminated, and finally the output of each agent in the microgrid is consistent and close to the nominal value.

The designed finite-time controller is:

Among them, α and β are finite-time control parameters, α is an exponential coefficient, 0<α<1, which improves the system convergence performance, β is a proportional coefficient, β>0, determines the step size of the control variable.

To verify the feasibility and stability of the designed finite-time controller, the following Lyapunov function is constructed:

therefore,

定义一个有界矩阵

对于连续的δ∈Φ，函数δ^TΓδ也是连续的，并且

存在而且大于0，定义为ρ。Among them, e=[e ₁ e ₂ … e _N ], B=diag{b ₁ b ₂ … b _N }, Γ=Ω+B, Γ is a positive semi-definite matrix, obviously,

define a bounded matrix

For continuous δ∈Φ, the function δT ^Γδ is also continuous, and

exists and is greater than 0, defined as ρ.

therefore,

则may wish to set

but

最后，可以得到：Assume again

Finally, you can get:

将会在有限时间

趋近于0，因此在时间t^*内代理的输出误差e也会趋近于0，即可以实现

According to the above proof

will be for a limited time

approaches 0, so the output error e of the agent will also approach 0 within time t ^* , that is, it can be achieved

that is

The detailed distributed secondary control strategy is shown in Figure 3. As shown in the figure, the information of neighboring DGs is obtained through a peer-to-peer sparse network, and the obtained information and local information are processed through a consensus algorithm. The compliance bias is fed back to the frequency and voltage controllers designed by a finite-time compliance strategy. The output of the controller is then fed into the droop control, which is then transferred to the voltage and current control loops, and finally the frequency and voltage are stabilized and tracked to nominal values.

(3) Design a predictive compensation strategy for communication interruption

The traditional ELM is based on the principle of empirical risk minimization to minimize the training error, but during the training process, there will be an overfitting problem, which reduces the generalization ability of the model. Therefore, we introduce the theory of structural risk minimization, reasonably weigh empirical risk and structural risk, and propose an improved ELM model (ES-ELM). When the communication data is lost and interfered, the improved extreme learning machine is used to train the historical data of voltage and frequency, so as to predict the original data, and finally input the predicted data into the consistency control. Compared with traditional extreme learning machine, ES-ELM has better generalization ability and robustness.

For any N different samples (X _i ,Y _i ), where X _i =[x _i1 x _i2 … x _in ] ^T ∈R ⁿ represents the historical output frequency or voltage of each DG, Y _i =[y _i1 y _i2 … y _im ] ^T ∈ R ^m represents the frequency or voltage of the expected output of each DG. For a single hidden layer neural network with L hidden layer nodes, it can be expressed as:

where g(x) is the excitation function, generally a Gaussian function, Wi = [ _wi1 w _i2 ... w _in ] ^T is the input weight, γ _i is the output weight, c _i is the bias of the _ith hidden layer unit, o _j represents the predicted output.

When considering empirical risk and structural risk, the mathematical model of ES-ELM can be expressed as:

Among them, ||ε|| ² represents the empirical risk, ||γ|| ² represents the structural risk, and η∈R represents the best compromise point determined by the cross-validation of the two risk ratio parameters.

The LS-SVM algorithm is introduced to convert the ES-ELM model into the Lagrange equation as:

Among them, λ=[λ ₁ λ ₂ … λ _N ] represents the Lagrange multiplier, H represents the output of the hidden layer, and γ=[γ ₁ γ ₂ … γ _L ] ^T represents the output weight

According to the optimal conditions of KTT, find the gradient of the Lagrange equation and set it to 0, we get

thereby getting

where X ⁺ denotes the generalized inverse of matrix X.

To sum up, ES-ELM considers both empirical risk and structural risk, so that the overfitting of the model is reduced and the generalization ability is excellent.

Before training, clarify the following contents: N different samples (X _i , Y _i ) in the training set, the number of hidden layers L, and the excitation function g(x). The ES-ELM-based prediction process mainly includes the following steps:

Step 1: When the data changes greatly, in order to obtain better training results, the data is preprocessed. The specific formula is:

Among them, x(i) and _x '(i) represent the original data and processed data, respectively, and Ex and σ _i represent the mean and standard deviation of the original data.

Step 2: The input weights w _ik and bias c _i are arbitrarily set in the range of (0,1), and the hidden layer output H is calculated.

Step 3: Calculate the output weight γ and the Lagrange multiplier λ according to formula (14).

After ES-ELM training, the local prediction of the lost communication information DG _i is carried out through historical data, and the predicted values of frequency and voltage are obtained respectively. A shorter prediction time can reduce the system instability caused by data loss.

Based on the above analysis, in the case of communication interruption, the prediction compensation mechanism is used to improve the stability of the entire system.

Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, etc., All should be included within the protection scope of the present invention.

Claims (3)

1. An island micro-grid control method based on finite time consistency theory is characterized by comprising the following steps:

in an island micro-grid based on a distributed secondary control strategy, a peer-to-peer sparse network is utilized to communicate with an adjacent agent, and information and local information are obtained through processing by a consistent algorithm; and feeding back the consistency deviation to a frequency and voltage controller designed by a finite time consistency strategy; then, the output of the controller is input to droop control, then transferred to a voltage and current control loop, finally frequency and voltage stabilization is realized and a nominal value is tracked;

when communication is interrupted, predicting historical data of adjacent agents based on an improved extreme learning machine, and inputting a prediction result to a frequency and voltage controller designed by a finite time consistency strategy;

the step of predicting the historical data of the adjacent agents based on the improved extreme learning machine specifically comprises the following steps:

(1) establishment of prediction model

For any N different samples (X)_i,Y_i) Wherein X is_i＝[x_i1 x_i2 … x_in]^T∈RⁿRepresenting the historical output frequency or voltage, Y, of each DG_i＝[y_i1 y_i2 … y_im]^T∈R^mThe frequency or voltage representing the desired output of each DG is expressed for a single hidden neural network having L hidden nodes as:

wherein g (x) is an excitation function and is highA function of si, W_i＝[w_i1 w_i2 … w_in]^TAs input weights, γ_iAs output weight, c_iIs the bias of the ith hidden layer unit, o_jRepresenting a prediction output;

when considering empirical and structural risks, the mathematical model of ES-ELM is expressed as:

wherein | | | purple hair²Representing empirical risk, | γ | luminance²Representing structural risk, wherein eta epsilon R represents the optimal break point determined by two risk proportion parameters in a cross validation mode;

an LS-SVM algorithm is introduced, and an ES-ELM model is converted into a Lagrange equation as follows:

wherein λ ═ λ₁ λ₂ … λ_N]Representing lagrange multiplier, H representing hidden layer output, γ ═ γ₁ γ₂ … γ_L]^TRepresenting output weights

According to the KTT optimal condition, solving the gradient of the Lagrange equation and making the gradient be 0 to obtain

Thereby obtaining

Wherein X⁺A generalized inverse matrix representing matrix X;

(2) training of predictive models

The following is made clear prior to training: training set N different samples (X)_i,Y_i) The number of hidden layers L, the excitation function g (x),

step 1, when the data change is large, in order to obtain a better training result, preprocessing the data, wherein a specific formula is as follows:

wherein x (i) and x' (i) represent the original data and the processed data, respectively, E_xAnd σ_iMeans and standard deviations representing the raw data;

step 2: input weight w_ikAnd bias c_iArbitrarily setting in the range of (0,1), and calculating hidden layer output H;

step 3: according to

Calculating an output weight gamma and a Lagrange multiplier lambda;

(3) result prediction

DG with lost communication information through historical data after ES-ELM training_iAnd carrying out local prediction to respectively obtain predicted values of the frequency and the voltage.

2. The island microgrid control method based on the finite time consistency theory is characterized in that the island microgrid is divided into a physical layer part and a network layer part; the physical layer mainly comprises a controller and distributed power supplies, wherein the controller stabilizes the output of each distributed power supply and then connects the output of each distributed power supply to a common coupling point through a static transfer switch; or the controller stabilizes the output of each distributed power supply and then supplies power to the load by using the distributed power supplies; and the network layer performs data exchange among different power electronic inverters to complete output synchronization of the distributed power supply in the whole microgrid.

3. The island microgrid control method based on the finite time consistency theory of claim 1, characterized in that DG is applied to an agent_iThe output error consistency algorithm is as follows:

e_i＝∑a_ij(x_j-x_i)+b_i(x^ref-x_i)

wherein e is_iRepresenting the global output error, x, of agent i_iRepresenting the actual output value, x, of agent i_jValue, x, communicated from a neighboring agent j representing agent i^refNominal value representing the output of agent i, Leader node as consistency algorithm, a_ijNetwork communication links representing weighted adjacency coefficients, i.e. agents i, b_iRepresenting the weight leading to the Leader, b if and only if agent i accesses the Leader_i＞0；

To achieve a time-limited adjustment effect, a sign function is introduced:

the designed finite time controller is as follows:

u_i＝βsig(e_i)^α

wherein, alpha and beta are finite time control parameters, alpha is an exponential coefficient, alpha is more than 0 and less than 1, the system convergence performance is improved, beta is a proportionality coefficient, beta is more than 0, and the step length of a control variable is determined.

Images