High-proportion new energy microgrid optimization operation method considering small interference stability constraint

Abstract

The invention discloses a high-proportion new energy microgrid optimization operation method considering small interference stability constraint. The method comprises the following steps: 1) generating key droop control parameters; 2) training the input convex neural network by using the key droop control parameters to obtain a convexity transformation model of the small interference stability constraint and the state matrix; (3) constructing a micro-grid real-time optimization model by taking minimization of system network loss and minimization of frequency as targets; 4) solving the micro-grid real-time optimization model to obtain a droop control coefficient; 5) converting the droop control coefficient into a stability index by using a convexity transformation model, and judging whether the stability index Zn is greater than or equal to ximin or not, if so, outputting the droop control coefficient, otherwise, optimizing the droop control coefficient by using a droop control parameter optimization model, and outputting the optimized droop control coefficient; according to the method, the stability of the system is guaranteed, efficient consumption of new energy is realized, the overall operation efficiency of the micro-grid is improved, and the calculation complexity is reduced.

Description

Technical Field

[0001] The present invention relates to the field of power grid operation optimization, and specifically to an optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints. Background Technique

[0002] With the increasing global energy demand and the improvement of environmental protection awareness, the traditional energy supply model is facing huge challenges. To cope with the pressure of climate change and the need for sustainable development, countries around the world have started to accelerate the transformation of the energy structure. In 2024, the installed capacity of new energy power generation in China will exceed that of coal-fired power for the first time, accounting for more than 40% of the total installed capacity. The large-scale centralized grid connection of new energy poses challenges to the safe operation of the power system and the balance of power generation and consumption. In recent years, the distributed energy in China has developed rapidly. By the end of 2023, the installed capacity of distributed new energy in the country has exceeded 250 million kilowatts, accounting for more than 10% of the total installed capacity of the power sources in China, and has become an important power supply mode. Against this background, the microgrid, as a flexible power system that can integrate various distributed energy sources, has gradually become one of the key technologies to solve the problems of the traditional power grid and improve energy utilization efficiency. However, the large-scale renewable energy is connected to the microgrid through power electronic devices, which puts forward higher requirements for the stability of the system and is more likely to become unstable under small disturbances. Therefore, the optimal operation method considering small-signal stability constraints is of great significance to the safety and stability of the microgrid.

[0003] Although the small-signal stability analysis method can judge the stability of the system under different states, there are still some deficiencies in the existing technologies in practical applications. First, since a large amount of renewable energy replaces traditional units for power generation, the dimension of the system control parameters increases significantly, making the solution process of the state-space matrix in the small-signal model very complex, resulting in a longer solution time for judging small-signal stability, which is very unfavorable for real-time optimization. Second, small-signal stability analysis mainly judges the stability of the existing power grid state. In modern power systems, there is a large amount of data collection. The use of analytical analysis methods lacks the utilization of historical data. As a large number of distributed energy sources are connected to the power grid through inverters, their droop control parameters are the main factors determining the stable state of the microgrid. Although simple and reliable stability assessment can be carried out through the reduced-order model method, it is still difficult to adapt to the stability determination of droop control parameters under dynamic changes. In addition, the small-signal stability analysis method usually assumes that the dynamic behavior of the system under normal operating conditions can be described by a linearized model. This method ignores the non-linear factors and the possible time-varying characteristics of the system, while new energy microgrids usually exhibit complex non-linear and time-varying characteristics. The real-time operation of the microgrid is essentially an optimal scheduling problem, and the scheduling decision needs to be dynamically adjusted under the real-time changing system state. However, the results of small-signal stability analysis cannot fully reflect this change, resulting in the inability to fully capture the dynamic stability of the system, especially when facing sudden disturbances. Summary of the Invention

[0004] The object of the present invention is to provide an optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, including the following steps:

[0005] 1) Generate small-signal stability samples of the new energy microgrid and perform preprocessing to obtain key droop control parameters;

[0006] 2) Use the key droop control parameters to train the input convex neural network to obtain a convexity conversion model of small-signal stability constraints and the state matrix;

[0007] 3) Construct a real-time optimization model of the microgrid with the goal of minimizing system network loss and frequency;

[0008] 4) Solve the real-time optimization model of the microgrid to obtain the droop control coefficient;

[0009] 5) Use the convexity conversion model to convert the droop control coefficient into a stability index, and judge whether the stability index Z n ≥ξ min holds. If so, output the droop control coefficient; otherwise, use the droop control parameter optimization model to optimize the droop control coefficient and output the optimized droop control coefficient; ξ min is the stability index constraint.

[0010] Furthermore, the small-signal stability samples of the new energy microgrid include network parameter data, load data, and droop control parameter data;

[0011] The network parameter data includes network topology structure and branch impedance;

[0012] The load data includes the load power of each node in the microgrid at a certain moment;

[0013] The droop control parameter data includes randomly generating a group of droop control parameters within the set operating range.

[0014] Furthermore, in step 1), the steps of generating the small-signal stability samples of the new energy microgrid include:

[0015] 1.1) Randomly generate the droop gain Δx of the current operating point and calculate the eigenvalues of the state matrix A s That is:

[0016]

[0017] where A s =(A - BD -1 C) is the state matrix; A is a matrix composed of various parameters related to the voltage and current of the microgrid; B is a coefficient matrix related to the inverter phase angle and inverter power output; C is a coefficient matrix related to the microgrid phase angle and power output, and D is a matrix representing the damping effect of the inverter and the transmission characteristics of the power grid;

[0018] 1.2) Judge whether the operating point is stable through the eigenvalues of the state matrix A s . If it is stable, go to step 1.3), otherwise update the operating point and return to step 1.1); the operating point is stable means that the real part of the largest eigenvalue of the state matrix A s is less than 0;

[0019] 1.3) Calculate the minimum damping ratio and damping ratio sensitivity of the operating point, and judge whether the operating point is near the stability boundary through the minimum damping ratio. If so, record the network parameter data, load data, and droop control parameter data of the current operating point; being near the stability boundary means that the minimum damping ratio range is [-1%, 1%];

[0020] 1.4) Update the operating point and repeat steps 1.1)-1.3) until m small-signal stability samples of the new energy microgrid at different operating points are generated.

[0021] Furthermore, in step 1), the preprocessing means: restricting the output of the small-signal stability samples of the new energy microgrid to [0, 1], and performing dimensionality reduction using the PCA method.

[0022] Furthermore, in step 2), the weight parameters of the input convex neural network hidden layer are set to be non - negative, and the activation function is a non - decreasing convex function.

[0023] Furthermore, the activation function of the input convex neural network is converted into a linear inequality by the big - M method, that is:

[0024]

[0025] Z i ≥0(3)

[0026]

[0027] Z i ≤Mδ i (5)

[0028] where M is an extremely large number, and δ i is a binary variable introduced for linearization; Z i is the output variable of the i - th layer; δ i is the activation function; w i , b i are the weight and bias; X is the input.

[0029] Furthermore, the output Z of the last layer of the input convex neural network is as follows: n as follows:

[0030] Z n =f ICNN (k p ,k q ) (6)

[0031] where k p , k q are the droop control gains. f ICNN is the input convex neural network.

[0032] Furthermore, in step 3), the objective function of the micro - grid real - time optimization model is as follows:

[0033]

[0034] where Δf t is the frequency deviation; α and β are the weighted coefficients of network loss and frequency deviation respectively, is the current of branch i - j, r ij is the resistance of branch i - j.

[0035] Furthermore, in step 3), the constraint conditions of the microgrid real-time optimization model include the Distflow power flow equation, the frequency dynamic differential equation, the inverter frequency and voltage regulation equation, the active and reactive power output constraints of distributed energy sources, and the frequency and voltage constraints;

[0036] Among them, the Distflow power flow equation is as follows:

[0037]

[0038]

[0039] In the formula, represents the active power output of the distributed energy source at time t; is the active power at the head of branch i-j at time t, is the active power of the load at node j at time t, is the active power at the head of branch j-k at time t; represents the reactive power output of the distributed energy source at time t; is the reactive power at the head of branch i-j at time t; is the reactive power of the load at node j at time t, is the reactive power at the head of branch j-k at time t; x ij is the input; is the current of branch i-j; Vit, are the voltages of nodes i and j at time t;

[0040] The frequency dynamic differential equation is as follows:

[0041]

[0042] Among them, P G is the total output of the power generation unit, P L is the load demand, f t is the current frequency, f n is the rated frequency; H and D are the inertia coefficient and equivalent damping of the entire system; f is the frequency; Δt is the time step;

[0043] The inverter frequency and voltage regulation equation is as follows:

[0044]

[0045] Among them, and respectively represent the active and reactive power outputs of the distributed energy source at time t, P i,set and Q i,set are the initial set points of the active and reactive powers for droop control, and represents the droop control gain; Δf represents the difference between the current frequency and the rated frequency; ΔV represents the difference between the current voltage V i t and the rated voltage V n difference;

[0046] The active and reactive power output constraints of distributed energy are as follows:

[0047]

[0048] In the formula, P i,DER,max , P i,DER,min are the upper and lower limits of the active power output of distributed energy at time t; (S i,DER ) 2 is the preset capacity upper limit;

[0049] The frequency and voltage constraints are as follows:

[0050] f min ≤ f t ≤ f max (20)

[0051] V min ≤ V i t ≤ V max (21)

[0052] Further, the droop control parameter optimization model is as follows:

[0053]

[0054] V min ≤ V i,t + ΔV i,t ≤ V max (23)

[0055] f min ≤ f t + Δf t ≤ f max (24)

[0056]

[0057] Among them, and are the correction amounts of the droop control coefficients; V max , V min are the voltage upper and lower limits; f max , f min are the frequency upper and lower limits; k i,q,max , k i,p,min are the upper and lower limits of the droop control gain k i,q,max , ki,q,min k i,p,min is the droop control gain upper and lower limits; ξ min is the lower limit for optimizing the droop control parameter.

[0058] The technical effects of the present invention are beyond doubt, and the beneficial effects of the present invention are as follows:

[0059] 1) The present invention adjusts the droop control parameters of the inverters in distributed energy by embedding small-signal stability constraints in the microgrid operation model, thereby realizing the safe and stable optimal operation of the microgrid.

[0060] 2) Build an islanded microgrid embedded stability surrogate model based on the Input Convex Neural Network (ICNN) to improve the stability and optimal dispatching ability of the microgrid under high proportion of new energy access.

[0061] 3) By establishing the system's frequency dynamic equation and using the central inertial frequency model to characterize the overall inertial characteristics of the microgrid, the frequency control can accurately reflect the dynamic changes of the system. Combining the Euler method to discretize the differential equation ensures that the evolution process of the system frequency can be efficiently calculated during the optimal dispatching process. On this basis, a neural network surrogate model is introduced, and the ICNN is used to model the stability constraints, so that the complex stability judgment problem can be transformed into an efficient convex optimization problem, thereby improving the real-time performance and applicability of small-signal stability judgment.

[0062] 4) By constructing a microgrid optimal operation model and embedding stability constraints into the droop control strategy of the inverter, the system can optimize the active and reactive power distribution while ensuring stability, reduce frequency deviation, and improve the economy and security of system operation. Using convex optimization technology and mixed-integer programming methods, the optimal dispatching problem can maintain good solution performance and system stability in the face of different load conditions, distributed energy fluctuations, and network topology changes.

[0063] 5) The optimization framework adopts a data-driven intelligent optimization strategy and combines historical operation data to train the stability surrogate model, so that the optimal dispatching method can effectively adapt to different operation scenarios and maintain efficient operation in a complex dynamic environment.

[0064] In summary, the present invention considers the stability constraints to optimize the microgrid dispatching strategy, realizes the efficient consumption of new energy while ensuring system stability, improves the overall operation efficiency of the microgrid, and reduces the computational complexity, providing a new optimization method and technical support for the safe and stable operation of new energy microgrids. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1It is the generation process of small interference stable training samples of the present invention;

[0066] Figure 2 It is the structural schematic diagram of the neural network in the present invention;

[0067] Figure 3 It is the flow chart of the present invention. Specific implementation manners

[0068] The present invention will be further described below in conjunction with embodiments, but it should not be understood that the above-mentioned subject scope of the present invention is limited to the following embodiments. Without departing from the above technical idea of the present invention, various substitutions and changes made according to the common general technical knowledge and customary means in the art shall be included within the protection scope of the present invention.

[0069] Embodiment 1:

[0070] Refer to Figures 1 to 3 , a method for optimizing the operation of a high-proportion new energy microgrid considering small interference stability constraints, comprising the following steps:

[0071] 1) Generate small interference stable samples of the new energy microgrid, and perform preprocessing to obtain key droop control parameters;

[0072] 2) Use the key droop control parameters to train the input convex neural network to obtain a convexity conversion model of small interference stability constraints and the state matrix;

[0073] 3) Construct a real-time optimization model of the microgrid with the goal of minimizing system network loss and frequency;

[0074] 4) Solve the real-time optimization model of the microgrid to obtain the droop control coefficient;

[0075] 5) Use the convexity conversion model to convert the droop control coefficient into a stability index, and judge whether the stability index Z n ≥ξ min holds. If so, output the droop control coefficient; otherwise, optimize the droop control coefficient using the droop control parameter optimization model and output the optimized droop control coefficient; ξ min is the stability index constraint.

[0076] The small interference stable samples of the new energy microgrid include network parameter data, load data, and droop control parameter data;

[0077] The network parameter data includes the network topology structure and branch impedance;

[0078] The load data includes the load power of each node in the microgrid at a certain moment;

[0079] The droop control parameter data includes randomly generating a group of droop control parameters within a set operating range.

[0080] In step 1), the steps of generating a small-signal stability sample of the new energy microgrid include:

[0081] 1.1) Randomly generate the droop gain Δx at the current operating point and calculate the eigenvalues of the state matrix A s , that is:

[0082]

[0083] In the formula, A s =(A - BD -1 C) is the state matrix; A is a matrix composed of various parameters related to the voltage and current of the microgrid; B is a coefficient matrix related to the inverter phase angle and the inverter power output; C is a coefficient matrix related to the microgrid phase angle and the power output, and D is a matrix representing the damping effect of the inverter and the transmission characteristics of the power grid;

[0084] 1.2) Judge whether the operating point is stable through the eigenvalues of the state matrix A s . If it is stable, go to step 1.3), otherwise update the operating point and return to step 1.1); The operating point is stable means that the real part of the largest eigenvalue of the state matrix A s is less than 0;

[0085] 1.3) Calculate the minimum damping ratio and damping ratio sensitivity of the operating point, and judge whether the operating point is near the stability boundary through the minimum damping ratio. If so, record the network parameter data, load data, and droop control parameter data of the current operating point; Being near the stability boundary means that the minimum damping ratio range is [-1%, 1%];

[0086] 1.4) Update the operating point and repeat steps 1.1)-1.3) until m small-signal stability samples of the new energy microgrid are generated.

[0087] In step 1), the preprocessing means: restricting the output of the small-signal stability sample of the new energy microgrid to [0, 1] and performing dimensionality reduction using the PCA method.

[0088] In step 2), the weight parameters of the input convex neural network hidden layer are set to be non-negative, and the activation function is a non-decreasing convex function.

[0089] The activation function of the input convex neural network is converted into a linear inequality through the big M method, that is:

[0090]

[0091] Z i ≥0(3)

[0092]

[0093] Z i ≤Mδ i (5)

[0094] where M is a very large number, and δ i is a binary variable introduced for linearization; Z i is the output variable of the i-th layer; δ i is the activation function; w i , b i are the weights and biases; X is the input.

[0095] The output Z of the last layer of the input convex neural network is as follows: n as follows:

[0096] Z n = f ICNN (k p , k q ) (6)

[0097] In the formula, k p , k q are the droop control gains. f ICNN is the input convex neural network.

[0098] In step 3), the objective function of the microgrid real-time optimization model is as follows:

[0099]

[0100] In the formula, Δf t is the frequency deviation; α and β are the weighting coefficients of network loss and frequency deviation respectively, is the current of branch i-j, r ij is the resistance of branch i-j.

[0101] In step 3), the constraint conditions of the microgrid real-time optimization model include the Distflow power flow equation, the frequency dynamic differential equation, the inverter frequency and voltage regulation equation, the active and reactive power output constraints of distributed energy sources, and the frequency and voltage constraints;

[0102] Among them, the Distflow power flow equation is as follows:

[0103]

[0104] In the formula, represents the active power output of the distributed energy source at time t; is the active power at the head of branch i-j at time t, is the active power of the load at node j at time t, is the active power at the head of branch j-k at time t; represents the reactive power output of distributed energy at time t; is the reactive power at the head of branch i-j at time t; is the reactive power of the load at node j at time t, is the reactive power at the head of branch j-k at time t; x ij is the input; is the current of branch i-j; Vit, are the voltages of nodes i and j at time t;

[0105] The frequency dynamic differential equation is as follows:

[0106]

[0107] where, P G is the total output of the power generation unit, P L is the load demand, f t is the frequency at the current moment, f n is the rated frequency; H and D are the inertia coefficient and equivalent damping of the entire system;

[0108] The inverter frequency and voltage regulation equation is as follows:

[0109]

[0110] where, and respectively represent the active and reactive power outputs of distributed energy at time t, P i,set and Q i,set are the initial set points of the active and reactive powers for droop control, and represent the droop control gains; Δf represents the difference between the frequency at the current moment and the rated frequency; ΔV represents the difference between the voltage V i t at the current moment and the rated voltage V n ;

[0111] The active and reactive power output constraints of distributed energy are as follows:

[0112]

[0113] In the formula, P i,DER,max 、P i,DER,min are the upper and lower limits of the active power output of distributed energy at time t; (S i,DER ) 2 is the preset capacity upper limit;

[0114] The frequency and voltage constraints are as follows:

[0115] f min ≤ f t ≤ f max (20)

[0116] V min ≤ V i t ≤ V max (21)

[0117] The droop control parameter optimization model is as follows:

[0118]

[0119] V min ≤ V i,t + ΔV i,t ≤ V max (23)

[0120] f min ≤ f t + Δf t ≤ f max (24)

[0121]

[0122]

[0123] Wherein, and are the correction amounts of the droop control coefficients; V max , V min are the upper and lower voltage limits; f max , f min are the upper and lower frequency limits; k i,q,max , k i,p,min are the upper and lower limits of the droop control gain ; k i,q,max , k i,q,min k i,p,min are the upper and lower limits of the droop control gain ; ξ min is the lower limit of the droop control parameter optimization.

[0124] Example 2:

[0125] An optimal operation method for a high - proportion new - energy micro - grid considering small - signal stability constraints, comprising the following steps:

[0126] 1) Generate small - signal stability samples of the new - energy micro - grid and perform pre - processing to obtain key droop control parameters;

[0127] 2) Use the key droop control parameters to train the input convex neural network to obtain a convexity conversion model between small - signal stability constraints and the state matrix;

[0128] 3) Construct a real-time optimization model of the microgrid with the goals of minimizing the system network loss and frequency;

[0129] 4) Solve the real-time optimization model of the microgrid to obtain the droop control coefficient;

[0130] 5) Use the convexity conversion model to convert the droop control coefficient into a stability index, and judge whether the stability index Z n ≥ξ min holds. If so, output the droop control coefficient; otherwise, use the droop control parameter optimization model to optimize the droop control coefficient and output the optimized droop control coefficient; ξ min is the stability index constraint.

[0131] Embodiment 3:

[0132] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, the technical content is the same as that of Embodiment 2. Further, the small-signal stability samples of the new energy microgrid include network parameter data, load data, and droop control parameter data;

[0133] The network parameter data includes the network topology structure and the branch impedance;

[0134] The load data includes the load power of each node in the microgrid at a certain moment;

[0135] The droop control parameter data includes randomly generating a group of droop control parameters within the set operating range.

[0136] Embodiment 4:

[0137] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, the technical content is the same as any one of Embodiments 2-3. Further, in step 1), the steps of generating the small-signal stability samples of the new energy microgrid include:

[0138] 1.1) Randomly generate the droop gain Δx of the current operating point and calculate the eigenvalues of the state matrix A s That is:

[0139]

[0140] where A s =(A - BD -1 C) is the state matrix; A is a matrix composed of the parameters related to the voltage and current of the microgrid; B is a coefficient matrix related to the inverter phase angle and the inverter power output; C is a coefficient matrix related to the phase angle and power output of the microgrid, and D is a matrix representing the damping effect of the inverter and the transmission characteristics of the power grid;

[0141] 1.2) Determine whether the operating point is stable through the eigenvalues of the state matrix A s . If it is stable, proceed to step 1.3); otherwise, update the operating point and return to step 1.1). An operating point being stable means that the real part of the largest eigenvalue of the state matrix A s is less than 0;

[0142] 1.3) Calculate the minimum damping ratio and damping ratio sensitivity of the operating point, and determine whether the operating point is near the stability boundary through the minimum damping ratio. If so, record the network parameter data, load data, and droop control parameter data of the current operating point. Being near the stability boundary means that the minimum damping ratio range is [-1%, 1%];

[0143] 1.4) Update the operating point and repeat steps 1.1) - 1.3) until small-signal stability samples of the new energy microgrid with m operating points are generated.

[0144] Example 5:

[0145] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, with the technical content being the same as any one of Examples 2 - 4. Further, in step 1), the preprocessing means: restricting the output of the small-signal stability samples of the new energy microgrid to [0, 1], and performing dimensionality reduction using the PCA method.

[0146] Example 6:

[0147] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, with the technical content being the same as any one of Examples 2 - 5. Further, in step 2), the weight parameters of the hidden layer of the input convex neural network are set to be non-negative, and the activation function is a non-decreasing convex function.

[0148] Example 7:

[0149] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, with the technical content being the same as any one of Examples 2 - 6. Further, the activation function of the input convex neural network is converted into a linear inequality through the big M method, that is:

[0150]

[0151] Z i ≥0(3)

[0152]

[0153] Z i ≤Mδ i (5)

[0154] where M is an extremely large number, and δi Binary variables introduced for linearization; Z i Output variable of the i-th layer; δ i Activation function; w i , b i Are weights and biases; X is the input.

[0155] Example 8:

[0156] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, the technical content is the same as any one of Examples 2-7. Further, the output Z of the last layer of the input convex neural network n Is as follows:

[0157] Z n = f ICNN (k p , k q ) (6)

[0158] In the formula, k p , k q Are droop control gains.

[0159] Example 9:

[0160] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, the technical content is the same as any one of Examples 2-10. Further, in step 3), the objective function of the microgrid real-time optimization model is as follows:

[0161]

[0162] In the formula, Δf t Is the frequency deviation; α and β are the weighting coefficients of the network loss and the frequency deviation respectively, Is the current of branch i-j, r ij Is the resistance of branch i-j.

[0163] Example 10:

[0164] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, the technical content is the same as any one of Examples 2-9. Further, in step 3), the constraint conditions of the microgrid real-time optimization model include the Distflow power flow equation, the frequency dynamic differential equation, the inverter frequency and voltage regulation equation, the active and reactive power output constraints of distributed energy sources, and the frequency and voltage constraints;

[0165] Among them, the Distflow power flow equation is as follows:

[0166]

[0167] In the formula, Denote the active power output of distributed energy at time t; is the power at the head of branch i-j at time t, is the load power of node j at time t, is the power at the head of branch j-k at time t;

[0168] The frequency dynamic differential equation is as follows:

[0169]

[0170] where, P G is the total output of the power generation unit, P L is the load demand, f t is the frequency at the current moment, f n is the rated frequency; H and D are the inertia coefficient and equivalent damping of the whole system;

[0171] The inverter frequency and voltage regulation equation is as follows:

[0172]

[0173] where, and respectively represent the active and reactive power outputs of distributed energy at time t, P i,set and Q i,set are the initial set points of active and reactive power for droop control, and represent the droop control gains;

[0174] The active and reactive output constraints of distributed energy are as follows:

[0175]

[0176] The frequency and voltage constraints are as follows:

[0177] f min ≤ f t ≤ f max (20)

[0178] V min ≤ V i t ≤ V max (21)

[0179] Example 11:

[0180] An optimal operation method for a high-proportion new energy microgrid considering small-signal stability constraints, the technical content is the same as any one of Examples 2-10. Further, the droop control parameter optimization model is as follows:

[0181]

[0182] V min V ≤ i,t + ΔV i,t ≤ V max (23)

[0183] f min f ≤ t + Δf t ≤ f max (24)

[0184]

[0185]

[0186] wherein, and are the correction amounts of the droop control coefficients.

[0187] Embodiment 12:

[0188] An optimal operation method for a high - proportion new - energy micro - grid considering small - signal stability constraints, including a data input module, a linear small - signal stability constraint generation module, a micro - grid real - time optimization module, and a droop control strategy correction module.

[0189] The specific steps are as follows:

[0190] Step 1: Data input

[0191] The model obtains the operating - point data of different states within the operating range, including:

[0192] Network parameter data: network topology structure and branch impedance

[0193] Load data: the load power of each node in the micro - grid at a certain moment

[0194] Droop control parameter data: randomly generate a set of droop control parameters (including active and reactive power) within the set operating range.

[0195] Since the damping effect of load dynamics on the droop control of the inverter is not significant, the generated droop control parameters are the main factors affecting stability. At the steady - state operating point, the small - signal model of the micro - grid can be described by the following differential - algebraic equations:

[0196]

[0197] 0 = g(x, u)(2)

[0198] where x is the state variable and u is the input variable. Equation (1) corresponds to the micro - grid dynamic equation, and equation (2) represents the power - flow equation. Linearizing the above equations gives:

[0199]

[0200] 0 = CΔx + DΔu (4)

[0201] Combining (3) and (5) and eliminating Δu, the system state equation can be expressed as:

[0202]

[0203] where A s =(A - BD -1 C) is the state matrix. In small-signal stability analysis, the eigenvalues and damping ratios of the state matrix are important indicators for stability determination. Using a 5th-order electromagnetic model, the microgrid is represented as an AC voltage source with droop control. The state variables of phase angle, frequency, voltage, and branch current in the dq coordinate system are selected. The state matrix can be obtained through existing loads, impedances, and droop gains, thereby evaluating the stability of the system.

[0204] The process of generating small-signal stability samples is as Figure 1 , first, the randomly generated droop gain is used to judge whether the operating point is stable by calculating the eigenvalues of the state matrix. The minimum damping ratio and damping ratio sensitivity of this operating point are calculated. Then, it is judged whether this operating point is near the stability boundary through the minimum damping ratio. If the minimum damping ratio is near the stability boundary, the sampling is stopped, and the information of the operating point near the stability boundary is recorded. If it is not in the boundary region, the adaptation step size is recalculated and the operating point is updated, and this process is continued iteratively until the stability boundary condition is met.

[0205] Step 2: Generation of linear small-signal stability constraints

[0206] Preprocess the generated small-signal stability samples to obtain the key droop control parameters after dimensionality reduction as the training data for the neural network. As Figure 2 , use the input convex neural network (ICNN) to complete the convexity conversion between the small-signal stability constraints and the state matrix. To ensure that the output-input relationship is convex, ICNN sets the hidden layer weight parameters to be non-negative, and the activation function is a non-decreasing convex function. At the same time, to compensate for the possible decrease in the model fitting ability caused by setting all hidden layer weight parameters to be non-negative, ICNN refers to the residual network and adds direct connections from the input layer to each hidden layer. Linearize the output of each layer of ICNN, and use the big M method to convert the ReLU activation function into a series of linear inequalities, and introduce mixed integer variables:

[0207]

[0208] Z i ≥0 (7)

[0209]

[0210] Z i ≤Mδ i (9)

[0211] where M is an extremely large number, and δ i is a binary variable introduced for linearization.

[0212] The output of the last layer of the ICNN is the stability index (the real part of the maximum eigenvalue or the minimum damping ratio). After linearizing the activation function, the mapping relationship between the network output and the input is linear, obtaining a stability constraint, which is a linear relationship between the inverter control parameters and the stability index:

[0213] Z n =f ICNN (k p ,k q )≥ξ min (10)

[0214] Step 3: Construction of the microgrid real-time optimization model

[0215] As Figure 3 shown, considering the uncertainty of renewable energy, a day-ahead optimization is carried out. The initial parameter settings of the distributed energy inverters are obtained by using the optimal power flow calculation (OPF) without stability constraints. Considering the small-signal stability constraints, a microgrid real-time optimization is carried out with the goal of minimizing the system network loss and the frequency to the minimum:

[0216]

[0217] For a microgrid considering a radial topology, the Distflow power flow equation is used to describe the power relationship, and second-order cone relaxation is used:

[0218]

[0219] The system frequency is represented by the inertia center model. Based on the system inertia constant, damping coefficient, and power balance relationship, a frequency dynamic differential equation is constructed, and the Euler method is used to discretize the frequency dynamic equation to obtain the frequency at each moment:

[0220]

[0221] where P G is the total output of the power generation unit, P L is the load demand, f t is the frequency at the current moment, and f n is the rated frequency. A linear droop function is used to represent the frequency regulation and voltage regulation process of the inverters in the microgrid, that is

[0222]

[0223] Among them, and respectively represent the active and reactive power outputs of distributed energy at time t, P i,set and Q i,set are the initial set points of active and reactive power for droop control. and represent the droop control gains.

[0224] The active and reactive power outputs of distributed energy should be within the capacity range:

[0225]

[0226] Overall, during the operation of the microgrid, the frequency and voltage should be within the allowable safety range, while considering the stability constraints obtained in Step 2:

[0227] f min ≤ f t ≤ f max (24)

[0228] V min ≤ V i t ≤ V max (25)

[0229] Step 4: Modification of droop control strategy

[0230] As Figure 3 , specifically, we perform stability analysis on the droop control coefficients obtained in Step 3, and use the mapping relationship obtained from the neural network training in Step 2 to generate stability constraints, so as to realize the stability analysis of the droop control coefficients at each moment, and then modify the droop control strategy. Specifically, when the initially set droop control coefficients do not meet the small-signal stability requirements, an optimization problem for a new droop control parameter is constructed, with the minimum change in the droop control parameter as the objective function and the droop control parameter as the only decision variable:

[0231]

[0232] V min ≤ V i,t + ΔV i,t ≤ V max (27)

[0233] f min ≤ f t + Δf t ≤ f max (28)

[0234]

[0235] Among them, and are the correction amounts of the droop control coefficient. After the correction of the droop control, the safety constraints (27)-(28) and the stability constraints (33) still need to be satisfied.

Claims

1. A method for optimizing the operation of a microgrid with a high proportion of renewable energy considering small disturbance stability constraints, characterized in that: The following steps are involved: 1) Generate small disturbance stability samples of new energy microgrid and perform preprocessing to obtain key droop control parameters; 2) Using key droop control parameters to train the input convex neural network, a convex transformation model of small disturbance stability constraints and state matrix is ​​obtained; 3) To minimize the system network loss and frequency, a real-time optimization model of the microgrid is constructed; 4) Solve the real-time optimization model of the microgrid and obtain the droop control coefficient; 5) Use the convexity conversion model to convert the droop control coefficient into a stability index and determine the stability index Z n ≥ξ min Is it true? If so, the droop control coefficient is output. Otherwise, the droop control coefficient is optimized using the droop control parameter optimization model, and the optimized droop control coefficient is output; ξ min is the stability index constraint.

2. A high-proportion new energy microgrid optimization operation method considering small disturbance stability constraints according to claim 1, characterized in that: The new energy microgrid small disturbance stability sample includes network parameter data, load data, and droop control parameter data; The network parameter data includes network topology and branch impedance; The load data includes the load power of each node in the microgrid at a certain moment; The droop control parameter data includes a droop control parameter group randomly generated within a set operating range.

3. The method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: In step 1), the steps of generating a small disturbance stable sample of a new energy microgrid include: 1.1) Randomly generate the droop gain Δx of the current operating point and calculate the state matrix A s The characteristic value of is: In the formula, A s =(A-BD -1 C) is the state matrix; A is a matrix composed of various parameters related to the voltage and current of the microgrid; B is a coefficient matrix related to the inverter phase angle and the inverter power output; C is a coefficient matrix related to the phase angle and power output of the microgrid, and D is a matrix representing the damping effect of the inverter and the transmission characteristics of the power grid; 1.2) Through the state matrix A s The eigenvalue of determines whether the operating point is stable. If it is stable, proceed to step 1.3). Otherwise, update the operating point and return to step 1.1). The stability of the operating point means that: the state matrix A s The real part of the maximum eigenvalue of is less than 0; 1.3) Calculate the minimum damping ratio and damping ratio sensitivity of the operating point, and determine whether the operating point is near the stable boundary by the minimum damping ratio. If so, record the network parameter data, load data, and droop control parameter data of the current operating point; near the stable boundary means: the minimum damping ratio range is [-1%, 1%]; 1.4) Update the operating point and repeat steps 1.1) to 1.3) until small disturbance stability samples of new energy microgrids at m operating points are generated.

4. The method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: In step 1), the preprocessing refers to: limiting the small disturbance stable sample output of the new energy microgrid to [0,1], and using the PCA method to reduce the dimension.

5. The method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: In step 2), the weight parameters of the hidden layer of the input convex neural network are set to be non-negative, and the activation function is a monotonically non-decreasing convex function.

6. A high-proportion new energy microgrid optimization operation method considering small disturbance stability constraints according to claim 1, characterized in that: The activation function of the input convex neural network is converted into a linear inequality through the big M method, namely: Z i ≥0(3) WITH i ≤Mδ i (5) Among them, M is a very large number, δ i Binary variable introduced for linearization; Z i is the output variable of the i-th layer; δ i is the activation function; w i 、b i are weights and biases; X is the input.

7. A method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: The last layer of the input convex neural network outputs Z n As shown below: Z n =f ICNN (k p ,k q )(6) In the formula, k p , k q is the droop control gain; f ICNN is the input of the convex neural network.

8. The method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: In step 3), the objective function of the microgrid real-time optimization model is as follows: Where Δf t is the frequency deviation; α and β are the weighting coefficients of network loss and frequency deviation respectively. is the current of branch ij, r ij is the resistance of branch ij.

9. The method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: In step 3), the constraints of the microgrid real-time optimization model include Distflow power flow equation, frequency dynamic differential equation, inverter frequency and voltage regulation equation, distributed energy active and reactive output constraints, frequency and voltage constraints; Among them, the Distflow power flow equation is as follows: In the formula, represents the active power output of distributed energy at time t; is the active power at the head end of branch ij at time t, is the load active power of node j at time t, is the active power at the head end of branch jk at time t; represents the reactive power output of distributed energy at time t; is the reactive power at the head end of branch ij at time t; is the load reactive power of node j at time t, is the reactive power at the head end of branch jk at time t; ij is the input; is the current of branch ij; is the voltage of nodes i and j at time t; The frequency dynamic differential equation is shown below: Among them, P G is the total output of the power generation unit, P L is the load demand, f t is the current frequency, f n is the rated frequency; H and D are the inertia coefficient and equivalent damping of the whole system; f is the frequency; Δt is the time step; The inverter frequency and voltage regulation equation is as follows: in, and They represent the active and reactive power output of distributed energy at time t, P i,set and Q i,set are the initial set points for active and reactive power for droop control, and represents the droop control gain; Δf represents the difference between the current frequency and the rated frequency; ΔV represents the current voltage V i t With rated voltage V n difference; The active and reactive output constraints of distributed energy are as follows: Where P i,DER,max , P i,DER,min is the upper and lower limits of the active power output of distributed energy at time t; (S i,DER ) 2 is the preset capacity limit; The frequency and voltage constraints are as follows: f min ≤f t ≤f max (20) In min ≤V i t ≤V max (21) 10. The method for optimizing operation of a microgrid with a high proportion of new energy considering small disturbance stability constraints according to claim 1, characterized in that: The droop control parameter optimization model is as follows: V min ≤V i,t +ΔV i,t ≤V max (23) f min ≤f t +Δf t ≤f max (24) in, and is the correction value of the droop control coefficient; V max 、V min is the voltage upper and lower limits; f max 、f min k is the upper and lower limits of frequency; i,q,max , k i,p,min is the droop control gain The upper and lower limits of k i,q,max , k i,q,min k i,p,min is the droop control gain The upper and lower limits of ξ min Optimize lower limits for droop control parameters.

Images