A power grid scheduling method and device considering wind power uncertainty and a medium

Abstract

The application discloses a power grid scheduling method and device considering wind power uncertainty and a medium, and the method comprises the following steps: constructing an opportunity constraint model based on alternating current flow according to wind power uncertainty; using a sensitivity factor and a first-order Taylor expansion to analyze the opportunity constraint model, and converting the opportunity constraint in the opportunity constraint model into a linear constraint; using an improved double-description method to calculate an original feasible region of the power grid according to the linear constraint; constructing a local nonlinear adjustment optimization model based on the original feasible region, and using a deep learning method to solve the local nonlinear adjustment optimization model to obtain a final feasible region, so as to schedule the power grid according to the feasible region. The application can quickly and accurately calculate the system feasible region under high wind power penetration, and provide early warning information for real-time scheduling of the power grid.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid scheduling, and in particular to a power grid scheduling method and device considering wind power uncertainty and a medium. BACKGROUND

[0002] With the development of new energy technology, the installed capacity of wind power and photovoltaic power is increasing year by year, and wind power has gradually become the main body of power supply. However, the randomness and volatility of wind power make the power grid face the problem of insufficient peak regulation and frequency modulation capacity, and even under extreme weather conditions, there is a contradiction between wind power curtailment and power shortage. Therefore, it is necessary to study a fast solution method of power grid feasible region considering wind power uncertainty to determine whether the future power system operating point is in a safe range and timely perform power grid scheduling.

[0003] The feasible region of a power system can provide important support for research on power markets, voltage control, planning, etc. From the time scale, the power system feasible region is divided into two aspects of steady-state feasible region and transient-state feasible region. However, the current research on the operating feasible region of the power system with high proportion of wind power is still in its infancy, and an efficient solution method for analyzing the operating feasible region has not been established. In the feasible region analysis method, the point domain analysis method can be used to describe the feasible region space under new energy, but it is difficult to obtain its analytical form, and as the scale of the power grid increases, the calculation time consumption increases dramatically. SUMMARY

[0004] To solve the above technical problems, the present application provides a power grid scheduling method and device considering wind power uncertainty and a medium, which can quickly and accurately calculate the system feasible region under high wind power penetration, and provide early warning information for real-time power grid scheduling.

[0005] The present application provides a power grid scheduling method considering wind power uncertainty, comprising:

[0006] According to the wind power uncertainty, an opportunity constraint model based on alternating current flow is constructed;

[0007] The sensitivity factor and the first-order Taylor expansion are used to analyze the opportunity constraint model, and the opportunity constraint in the opportunity constraint model is converted into a linear constraint;

[0008] According to the linear constraint, an improved double description method is used to calculate the original feasible region of the power grid;

[0009] Based on the original feasible region, a local nonlinear adjustment optimization model is constructed, and a deep learning method is used to solve the local nonlinear adjustment optimization model to obtain the final feasible region, so as to perform power grid scheduling according to the feasible region.

[0010] As an improvement of the above-mentioned scheme, the constructing an opportunity constraint model based on alternating current power flow according to wind power uncertainty comprises:

[0011] Modeling new energy power as a sum of wind power ultra-short-term point prediction value and wind power uncertainty deviation;

[0012] Constructing generator power balance constraint, node voltage constraint and branch power flow constraint based on the wind power uncertainty deviation;

[0013] Expressing the generator power balance constraint, the node voltage constraint and the branch power flow constraint as opportunity constraints to obtain an opportunity constraint model.

[0014] As an improvement of the above-mentioned scheme, the using sensitivity factors and first-order Taylor expansion to analyze the opportunity constraint model, and converting the opportunity constraint in the opportunity constraint model into a linear constraint comprises:

[0015] Determining sensitivity factors about wind power uncertainty deviation according to generator power, voltage and branch power flow;

[0016] According to the sensitivity factors, re-expressing the opportunity constraint model to obtain an opportunity constraint containing sensitivity factors;

[0017] Using first-order Taylor expansion to analyze the opportunity constraint containing sensitivity factors to obtain a linear constraint.

[0018] As an improvement of the above-mentioned scheme, the sensitivity factors specifically comprise:

[0019]

[0020]

[0021] Wherein, and respectively represent the sensitivity factors of generator injected active power and injected reactive power; Γ v is the sensitivity factor of voltage; and respectively represent the sensitivity factors of branch active power flow and reactive power flow, ω is wind power uncertainty deviation, p G is the generator injected active power, q G is the generator injected reactive power, v is the node voltage, p L is the branch active power flow, q L is the branch reactive power flow.

[0022] As an improvement of the above-mentioned scheme, the according to the linear constraint, using an improved double description method to calculate an original feasible region of the power grid comprises:

[0023] The linear constraints are used to solve a vertex description problem of a polyhedron, and the vertex description problem of the polyhedron is converted into a corresponding polyhedral cone problem.

[0024] A vertex subset, a set of extreme rays, and a set of adjacent pairs are initialized.

[0025] Adjacent pairs and vertices are updated by gradually adding inequalities in the linear constraints; the updated adjacent pairs include: traversing each extreme ray in geometry, and performing an adjacency test on each extreme ray and vertex to screen updated adjacent pairs.

[0026] Pairs in the adjacent pairs that no longer generate new rays are deleted to obtain a final polyhedral vertex set, and an original feasible region of the power grid is determined.

[0027] As an improvement of the above scheme, a local nonlinear adjustment optimization model is constructed based on the original feasible region, including:

[0028] Based on the original feasible region, a geometric center of a polyhedron corresponding to the original feasible region is calculated; the polyhedron is n-dimensional and has m vertices, where the coordinates of each vertex u i is an n-dimensional vector u i =(u i1 ,u i2 ,…,u in ), and the coordinates of the geometric center C=(C1,C2,…,C n ) are as follows:

[0029]

[0030] N points u old are taken on each edge to construct a local nonlinear adjustment optimization model, specifically:

[0031] min‖u new -u old ‖

[0032] s.t. the linear constraints (21)

[0033] where u new represents a vertex to be solved.

[0034] As an improvement of the above scheme, a deep learning method is used to solve the local nonlinear adjustment optimization model to obtain a final feasible region, including:

[0035] A deep learning method based on an optimization neural network (ONN) is used to batch solve the local nonlinear adjustment optimization model to obtain new feasible region vertices to form a new feasible region; the ONN model is defined as follows:

[0036] y(t; w) = y0+ (1 - e -t )N(t; w) (22)

[0037] where, is the initial solution, is the time horizon, t ∈ [0, T] is the time variable; 1 - e -t for ensuring that the ONN model always satisfies the initial condition y(0; w) = y0;

[0038] The endpoint of the state solution of the ONN model is the approximation of the optimal solution of the local nonlinear adjustment optimization model, that is:

[0039]

[0040] Combining equations (22)-(23) gives:

[0041] P(y(T; w)) ≈ y * (24)

[0042] where P(·) is the projection function that projects the endpoint to the feasible set, and equation (24) indicates that the endpoint of the ONN model together with the projection function is the approximate solution of the optimal solution of the local nonlinear adjustment optimization model;

[0043] The loss function of the ONN model is:

[0044]

[0045] where φ(·) is the ordinary differential equation corresponding to the local nonlinear adjustment optimization model, e -γ*t for redistributing the weight in the loss and giving a higher weight to the time close to the origin, and γ is the weighting hyperparameter;

[0046] The objective function of the ONN model is:

[0047]

[0048] where the objective function E(w) is the integral of the loss function over the time horizon [0, T], and the loss function L(T, w) represents the error of the ONN model at time t;

[0049] The model is trained by minimizing the batch loss:

[0050]

[0051] where, is a set of time t randomly selected from [0, T], denotes the size of the set;

[0052] After obtaining the new feasible region, the geometric center thereof is calculated again, and a loop iteration is performed based on the local non-linear adjustment optimization model and the ONN model until a preset accuracy requirement is met, and a final feasible region is obtained.

[0053] As an improvement of the above scheme, after obtaining the final feasible region, the method further comprises:

[0054] The final feasible region is evaluated by using Hausdorff distance, Minkowski addition and volume ratio, Jaccard index and mean square error of support function respectively, and an evaluation result is obtained.

[0055] The embodiment of the present application further provides a power grid scheduling device considering wind power uncertainty, comprising:

[0056] An opportunity constraint construction module is configured to construct an AC power flow-based opportunity constraint model according to wind power uncertainty;

[0057] A linear constraint conversion module is configured to analyze the opportunity constraint model by using a sensitivity factor and a first-order Taylor expansion, and convert the opportunity constraint in the opportunity constraint model into a linear constraint;

[0058] A feasible region calculation module is configured to calculate an original feasible region of the power grid by using an improved double description method according to the linear constraint;

[0059] A feasible region optimization model is configured to construct a local non-linear adjustment optimization model based on the original feasible region, and solve the local non-linear adjustment optimization model by using a deep learning method to obtain a final feasible region, so as to perform power grid scheduling according to the feasible region.

[0060] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, wherein when the computer program runs, the computer readable storage medium controls a device where the computer readable storage medium is located to perform the power grid scheduling method considering wind power uncertainty.

[0061] Compared with the prior art, the power grid scheduling method, device and medium considering wind power uncertainty provided by the embodiment of the present application have the beneficial effects that: by using the chance constraint analytical method based on the sensitivity factor and the first-order Taylor expansion, the chance constraint can be analyzed to the maximum extent under the condition of considering wind power uncertainty and ensuring no over-limit, and the chance constraint is converted into a form that can be directly processed; by using the improved double description method to solve the feasible region vertex, the vertex calculation problem in the case of large-scale high dimension can be solved, the overall complexity of the algorithm is reduced, and the calculation efficiency is improved; by using the ONN model, the large-scale nonlinear optimization model can be quickly and accurately solved, and the accuracy and calculation efficiency are further improved. The embodiment of the present application can quickly calculate the feasible region vertex on the basis of ensuring the safety of the operating point in the feasible region, and can meet the requirements of real-time scheduling of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 is a flowchart of a power grid scheduling method considering wind power uncertainty provided by the embodiment of the present application;

[0063] Figure 2 is a structural diagram of an improved IEEE case5 system provided by the embodiment of the present application;

[0064] Figure 3 is a calculation result diagram of an improved double description method provided by the embodiment of the present application;

[0065] Figure 4 is a structural diagram of a power grid scheduling device considering wind power uncertainty provided by the embodiment of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0067] Please refer to Figure 1 , Figure 1 is a flowchart of a power grid scheduling method considering wind power uncertainty provided by the embodiment of the present application. The power grid scheduling method considering wind power uncertainty comprises the following steps:

[0068] S1: constructing an opportunity constraint model based on alternating current power flow according to wind power uncertainty;

[0069] As one of the optional embodiments, the step of constructing an opportunity constraint model based on alternating current power flow according to wind power uncertainty comprises the following steps:

[0070] The new energy power is modeled as a sum of a wind power ultra-short-term point prediction value and a wind power uncertainty deviation;

[0071] Based on the wind power uncertainty deviation, a generator power balance constraint, a node voltage constraint, and a branch power flow constraint are constructed;

[0072] The generator power balance constraint, the node voltage constraint, and the branch power flow constraint are expressed as chance constraints to obtain a chance constraint model.

[0073] Specifically, without loss of generality, it is assumed that each node has a conventional generator, a new energy power plant, and a load, and the uncertainty is derived from the fluctuation of the new energy output. The conventional generator and the new energy power plant are collectively referred to as the power generation side, and the node where they are located is denoted as The node with uncertainty is denoted as The node and branch sets of the power system are respectively

[0074] Since the nodes are divided into PQ nodes, PV nodes, and balance nodes (i.e., θV nodes), the variables related to the PQ nodes in the following are denoted by subscript PQ, and the rest are the same. The active power P and the reactive power Q of the PQ node are given, and the node voltage and phase (V, δ) are to be solved. The active power injection and voltage amplitude of the PV node are given, and the reactive power injection and voltage phase angle are to be solved. The voltage amplitude and phase of the balance node are given, and the active power injection and reactive power injection are to be solved.

[0075] The new energy power is modeled as a sum of a deterministic prediction value and a deviation:

[0076]

[0077] where p wind is a wind power ultra-short-term point prediction value, ω is a wind power uncertainty deviation, i.e., the source of uncertainty, represents the wind power considering the wind power uncertainty.

[0078] The covariance matrix of the wind power uncertainty is Without assuming that the wind power uncertainty is independent, the non-diagonal elements of ∑ are allowed to be non-zero values. The total deviation caused by the uncertainty is

[0079]

[0080] The standard deviation of the total deviation Ω is

[0081] The participation factor vector is defined as α represents the contribution of each generator to the uncertainty, and is required to ensure that the same

[0082] (1) Generator power balance constraints:

[0083] Assume that the change of active power caused by wind power uncertainty is unknown, denoted by is balanced by the generators at the

[0084]

[0085] Assume that all the generators connected at PV and The output of the generators connected at PQ nodes is unchanged and is deterministic, then for the reactive power:

[0086]

[0087] (2) Node voltage constraints:

[0088] The voltage magnitude at PV and

[0089]

[0090] (3) Branch power flow constraints:

[0091]

[0092] From the above analysis, the chance constraints that constitute the feasible region can be expressed as:

[0093]

[0094]

[0095] θ θV = 0 (11-12)

[0096]

[0097] where γ is called the power ratio, and ε represents the quantile. In constraint (11-1), f(·) is the source-load balance function, is the node voltage phase angle uncertainty surrogate, is the node voltage magnitude uncertainty surrogate, is the branch active power flow uncertainty surrogate, is the reactive power injected by the branch; constraint (11-1) is the power balance constraint of the system; constraints (11-2) and (11-3) are the active power constraints; constraints (11-4) and (11-5) are the reactive power constraints; constraints (11-6) and (11-7) are the voltage constraints; constraints (11-8) and (11-9) are the active power flow constraints of the branch; constraints (11-10) and (11-11) are the reactive power flow constraints of the branch; constraint (11-12) indicates that the voltage phase angle of the balanced node is 0; constraints (11-13) and (11-14) are the active power and reactive power injected by the node, q G is the reactive power injected by the branch; constraint (11-1) is the power balance constraint of the system; constraints (11-2) and (11-3) are the active power constraints; constraints (11-4) and (11-5) are the reactive power constraints; constraints (11-6) and (11-7) are the voltage constraints; constraints (11-8) and (11-9) are the active power flow constraints of the branch; constraints (11-10) and (11-11) are the reactive power flow constraints of the branch; constraint (11-12) indicates that the voltage phase angle of the balanced node is 0; constraints (11-13) and (11-14) are the active power and reactive power injected by the node, q D is the reactive power injected by the branch; constraint (11-1) is the power balance constraint of the system; constraints (11-2) and (11-3) are the active power constraints; constraints (11-4) and (11-5) are the reactive power constraints; constraints (11-6) and (11-7) are the voltage constraints; constraints (11-8) and (11-9) are the active power flow constraints of the branch; constraints (11-10) and (11-11) are the reactive power flow constraints of the branch; constraint (11-12) indicates that the voltage phase angle of the balanced node is 0; constraints (11-13) and (11-14) are the active power and reactive power injected by the node, q

[0098] S2: analyzing the chance constraint model by using a sensitivity factor and a first-order Taylor expansion to convert the chance constraint in the chance constraint model into a linear constraint;

[0099] As one of the optional embodiments, the analyzing the chance constraint model by using a sensitivity factor and a first-order Taylor expansion to convert the chance constraint in the chance constraint model into a linear constraint comprises:

[0100] determining a sensitivity factor about the wind power uncertainty deviation according to the generator power, the voltage, and the branch flow;

[0101] re-representing the chance constraint model according to the sensitivity factor to obtain a chance constraint containing a sensitivity factor;

[0102] analyzing the chance constraint containing the sensitivity factor by using a first-order Taylor expansion to obtain a linear constraint.

[0103] Further, the sensitivity factor specifically comprises:

[0104]

[0105] wherein, and respectively represent the sensitivity factors of the active power and the reactive power injected by the generator; Γ v is the sensitivity factor of the voltage; and respectively represent the sensitivity factors of the active power flow and the reactive power flow of the branch, ω is the wind power uncertainty deviation, p G is the active power injected by the generator, qG for the generator, v is the node voltage, p L for the branch active power flow, q L for the branch reactive power flow.

[0106] Specifically, the impact of wind power uncertainty ω is explained using a linearization method based on the first-order Taylor expansion of the day-ahead operating point, based on which the sensitivity factor with respect to the uncertainty ω is defined. The sensitivity factor nonlinearly depends on the day-ahead operating point (θ, v, p, q) and linearly depends on α, γ. At this time, the generator power balance constraint, the node voltage constraint, and the branch power flow constraint can be approximately expressed as:

[0107]

[0108] According to equations (13)-(17), the chance constraint in equation (11) can be rewritten as:

[0109]

[0110] where the active power chance constraint of the PV node depends on the reactive power chance constraint of the PV node depends on the reactive power chance constraint of the PQ node is deterministic because the injected power is fixed. Similarly, the chance constraint of the PQ node voltage depends on Γ v while those of the PV and the PV node are deterministic.

[0111] At this time, according to the first-order Taylor expansion, the chance constraint of equation (18) can be analytically expressed as the following linear constraint:

[0112]

[0113]

[0114] S3: according to the linear constraint, an original feasible region of the power grid is calculated using an improved double description method;

[0115] As one of the optional embodiments, the original feasible region of the power grid is calculated using an improved double description method according to the linear constraint, including:

[0116] The linear constraint and the vertex description problem of the polyhedron to be solved are converted into a corresponding polyhedral cone problem;

[0117] Initialize a vertex subset, a set of extreme rays, and a set of adjacent pairs;

[0118] The adjacent pairs and vertices are updated by adding the inequalities in the linear constraints step by step; the updating of the adjacent pairs comprises: traversing each extreme ray in the geometry, and performing an adjacent test on each extreme ray and vertex to screen the updating of the adjacent pairs;

[0119] The pairs in the adjacent pairs that no longer generate new rays are deleted to obtain a final polyhedral vertex set, and the original feasible region of the power grid is determined.

[0120] Specifically, the above linear inequalities, i.e., formula (19), can be uniformly represented as: where x represents the vertices of the polyhedron. Ax≤b is a unified mathematical representation of all inequalities (19), A can represent the coefficients of all variables to be solved in formula (19), x represents all variables to be solved, and b represents all constant terms.

[0121] Since the designed algorithm is for a polyhedral cone, and the inequality constraints and the construction problem of the feasible region to be solved are described by the vertices of a polyhedron, it is necessary to convert the vertex description problem of the polyhedron into a corresponding polyhedral cone problem. Specifically, a vertex of a polyhedron can be found by converting into a polyhedral cone problem , and then setting x d+1 =1.

[0122] In the double description method, the vertex set needs to be updated each time, and a key step is to check the adjacency of the extreme rays on the cone, which involves judging whether two extreme rays share a face of a polyhedron. The adjacency check includes algebraic check and combinatorial check. The algebraic check means that two extreme rays u1, u2 are adjacent in the extreme ray set U(C) of the polyhedral cone if and only if the rank of the joint tensor of u1 and u2 is The combinatorial check means that two extreme rays u1, u2 are adjacent in the extreme ray set U(C) of the polyhedral cone if and only if

[0123] Condition 1: If {u1, u2} is an edge in G, then |Z(u1)∩Z(u2)|≥r-2.

[0124] Let G be an undirected graph composed of a polyhedral cone C, and its vertex set is the extreme ray set U of C. {u1, u2} forms an edge if and only if |Z(u1)∩Z(u2)|≥r-2. The edge set of G is denoted as E(G). Condition 1 is a necessary condition for the adjacency of extreme rays, which is a simplified adjacency screening condition for fast preliminary screening to significantly reduce the extreme ray pairs that need to be verified and avoid redundant calculations.

[0125] ​​Condition 2: For vertices u1 and u2 in the polar ray adjacency graph G of a polyhedron, condition is true if and only if there is no other vertex in G. Such that {u1,w} and {u2,w} are both edges of G, and When {u1,u2} is an edge in G.

[0126] Condition 2 is a sufficient condition for whether polar rays are adjacent. After initial screening, the final adjacency is determined in a more precise way. According to Condition 2, if two rays are adjacent in G, then they are truly adjacent if and only if there are no other rays w such that... That is, no other rays can be "sandwiched" between these two rays.

[0127] Furthermore, an improved bi-description method is used to determine the vertex set of the polyhedron. The specific steps include:

[0128] S31: The inequality coefficient matrix of the polyhedron constant terms of inequalities As input;

[0129] S32: Based on the input, construct matrix A′=[A,-b] and variable x′=[x T ,y] T This transforms the vertex description problem of a polyhedron into the corresponding polyhedron problem.

[0130] S33: Set the basis matrix U = I d+！ Selecting an initial subset Make |K|=d(rank(A)=d) and det(A) k )≠0;

[0131] S34: Construct the polar ray set U of the initial subset Bx≥0 (B is a submatrix of A, determined by K);

[0132] S35: Initialize the set of adjacency pairs E = {};

[0133] S36: Gradually add inequalities, update poles and adjacent pairs, until K ≠ {1,2,…,m}:

[0134] S361: Choose a new inequality i∈K and construct the set

[0135] S362: Initialize the polar ray set U +- ={};

[0136] S363: Traverse every polar ray u1∈U in U+ + For each polar ray u2∈U in U- -Adjacency test with u1: Using conditions 1 and 2 above for screening, new polar rays are calculated. Then update U +- =U +- ∪{w}, update the set of adjacency pairs E = E∪{{u1,w},{u2,w}};

[0137] S364: Update vertex set U = U + ∪U +- ∪U0

[0138] S365: Delete adjacent pairs that no longer generate new rays and remove processed inequalities;

[0139] S37: Output the set of polyhedron vertices V, which is the original feasible region of the power grid.

[0140] S4: Based on the original feasible region, a local nonlinear adjustment optimization model is constructed, and a deep learning method is used to solve the local nonlinear adjustment optimization model to obtain the final feasible region, so as to carry out power grid scheduling according to the feasible region.

[0141] Specifically, the geometric center of the feasible region of the original high-dimensional polyhedron is first calculated.

[0142] The step of constructing a local nonlinear adjustment optimization model based on the original feasible region includes:

[0143] Based on the original feasible region, calculate the geometric center of the polyhedron corresponding to the original feasible region; assume the polyhedron is n-dimensional and has m vertices, where each vertex u i The coordinates are an n-dimensional vector u i =(u i1 ,u i2 ,…,u in If the geometric center is C = (C1, C2, ..., C...), then the coordinates of the geometric center are C = (C1, C2, ..., C...). n )for:

[0144]

[0145] Take N points u on each edge. old A local nonlinear adjustment optimization model is constructed, specifically as follows:

[0146] min‖u new -u old ‖ (twenty one)

[0147] The linear constraint of st (19)

[0148] Among them, u new This represents the vertex to be solved.

[0149] Furthermore, the step of using deep learning methods to solve the local nonlinear adjustment optimization model to obtain the final feasible region includes:

[0150] The local nonlinear adjustment optimization model is solved in batches using a deep learning method based on Optimization-Neural Networks (ONN) to obtain new feasible region vertices, forming a new feasible region. The ONN model is defined as follows:

[0151] y(t; w) = y0 + (1 - e -t )N(t;w) (22)

[0152] in, As the initial solution, The time range is defined as t∈[0,T], where 1-e is the time variable. -t Used to ensure that the ONN model always satisfies the initial condition y(0; w) = y0;

[0153] The endpoints of the state solutions of the ONN model are approximations of the optimal solutions of the local nonlinear adjustment optimization model, that is:

[0154]

[0155] Combining equations (22) and (23), we get:

[0156] P(y(T;w))≈y * (twenty four)

[0157] Where P(·) is the projection function that projects the endpoints onto the feasible set, and Equation (24) indicates that the endpoints of the ONN model together with the projection function are an approximate solution of the optimal solution of the local nonlinear adjustment optimization model;

[0158] The loss function of the ONN model is:

[0159]

[0160] Where φ(·) is the ordinary differential equation corresponding to the local nonlinear adjustment optimization model, e -γ*t Used to redistribute weights in the loss and assign higher weights to times closer to the origin; γ is a weighted hyperparameter.

[0161] The objective function of the ONN model is:

[0162]

[0163] Wherein, the objective function E(w) is the integral of the loss function over the time range [0,T], and the loss function L(T,w) represents the error of the ONN model at time t;

[0164] The model is trained by minimizing the batch loss:

[0165]

[0166] in, Let t be a set of times randomly selected from [0,T]. Indicates the size of the set;

[0167] After obtaining the new feasible region, its geometric center is recalculated, and iterative processes are performed based on the local nonlinear adjustment optimization model and the ONN model until the preset accuracy requirement is met, thus obtaining the final feasible region.

[0168] As one optional embodiment, after obtaining the final feasible domain, the method further includes:

[0169] The final feasible region was evaluated using the Hausdorff distance, the Minkowski summation to volume ratio, the Jaccard exponent, and the mean square error of the support function, respectively, to obtain the evaluation results.

[0170] The support function describes the maximum extension distance of a polyhedron in a given direction. Given a polyhedron A and a direction vector λ, the support function h(A,λ) is defined as follows:

[0171]

[0172] The support function h(A,λ) is the projected distance of the farthest point of polyhedron A in a given direction λ. To compare two polyhedra A and B, their support function values ​​are calculated in multiple different directions, and the similarity is measured by the mean squared error. The smaller the difference in the function, the more similar the two polyhedra are.

[0173] In a specific example, the IEEE Case 5 system is used for the experiment. The system source and load information is as follows: the wind turbine is connected to bus B2 with a power output of 300MW, and the grid topology is as follows. Figure 2 As shown, W represents the wind turbine generator, and B1-B5 represent nodes 1-5, respectively. Calculations were performed using the method of this embodiment, resulting in 47 vertices. The feasible region calculation results are as follows: Figure 3 As shown, its Hausdorff distance is 138.97, the Minkowski sum to volume ratio is 4.03, and the Jaccard exponent is 0.91, which meets the requirements for rapid solution of the grid feasible region, can provide reliable information for grid dispatch, and effectively improve the efficiency and accuracy of grid dispatch considering wind power uncertainty.

[0174] This invention, through an analytical approach to opportunity constraints based on sensitivity factors and first-order Taylor expansion, maximizes the analytical transformation of opportunity constraints while considering wind power uncertainties and ensuring no exceedances, converting opportunity constraints into directly manageable methods. By employing an improved dual-description method to solve for feasible region vertices, it addresses the challenge of vertex computation in large-scale, high-dimensional scenarios, reducing overall algorithm complexity and improving computational efficiency. Furthermore, the use of the ONN model enables rapid and accurate solving of large-scale nonlinear optimization models, further enhancing accuracy and computational efficiency. This invention can rapidly calculate feasible region vertices while ensuring 99.99% safety of operating points within the feasible region, meeting the requirements of real-time power grid dispatch.

[0175] Accordingly, the present invention also provides a grid dispatching device that considers the uncertainty of wind power, which can implement all the processes of the grid dispatching method considering the uncertainty of wind power in the above embodiments.

[0176] Please see Figure 4 , Figure 4 This is a schematic diagram of a grid dispatching device considering wind power uncertainty provided in an embodiment of the present invention. The grid dispatching device considering wind power uncertainty includes:

[0177] Opportunity constraint construction module 401 is used to construct an opportunity constraint model based on AC power flow based on wind power uncertainty;

[0178] The linear constraint transformation module 402 is used to analyze the chance constraint model using a sensitivity factor and a first-order Taylor expansion, and to transform the chance constraints in the chance constraint model into linear constraints.

[0179] The feasible region calculation module 403 is used to calculate the original feasible region of the power grid based on the linear constraints and using an improved dual description method.

[0180] The feasible region optimization model 404 is used to construct a local nonlinear adjustment optimization model based on the original feasible region, and to solve the local nonlinear adjustment optimization model using a deep learning method to obtain the final feasible region, so as to perform power grid scheduling according to the feasible region.

[0181] Preferably, the step of constructing an opportunity constraint model based on AC power flow based on wind power uncertainty includes:

[0182] The power of new energy sources is modeled as the sum of the ultra-short-term point prediction value of wind power and the uncertainty deviation of wind power.

[0183] Based on the aforementioned wind power uncertainty deviation, generator power balance constraints, node voltage constraints, and branch power flow constraints are constructed.

[0184] The generator power balance constraint, the node voltage constraint, and the branch power flow constraint are expressed as chance constraints, thus obtaining the chance constraint model.

[0185] Preferably, the step of analyzing the chance constraint model using a sensitivity factor and a first-order Taylor expansion to transform the chance constraints in the chance constraint model into linear constraints includes:

[0186] Based on generator power, voltage, and branch power flow, determine the sensitivity factor for wind power uncertainty deviation;

[0187] Based on the sensitivity factor, the chance constraint model is re-represented to obtain a chance constraint containing the sensitivity factor;

[0188] A first-order Taylor expansion is used to analyze the chance constraint containing the sensitivity factor, resulting in a linear constraint.

[0189] Preferably, the sensitivity factor specifically includes:

[0190]

[0191] Among them, Γ pG With Γ qG These represent the sensitivity factors for injected active power and injected reactive power, respectively; Γ v It is the voltage sensitivity factor; Γ pL With Γ qL These represent the sensitivity factors for active and reactive power flow in the branch circuit, respectively, where ω is the wind power uncertainty deviation, and p G Injecting active power into the generator, q G Injecting reactive power into the generator, where v is the node voltage and p is the reactive power. L For the meritorious trend of the branch road, q L The branch road is not contributing to the current.

[0192] Preferably, the step of calculating the original feasible region of the power grid using an improved dual-description method based on the linear constraints includes:

[0193] The linear constraints and the vertex description problem of the feasible region to be solved are polyhedra. The vertex description problem of the polyhedra is transformed into the corresponding polyhedral cone problem.

[0194] Initialize the vertex subset, polar ray set, and adjacency pair set;

[0195] The adjacency pairs and vertices are updated by progressively adding inequalities to the linear constraints; the updating of adjacency pairs includes: traversing each polar ray in the geometry and performing an adjacency test on each polar ray and vertex to filter and update adjacency pairs;

[0196] Remove adjacent pairs that no longer generate new rays to obtain the final set of polyhedral vertices, thus determining the original feasible region of the power grid.

[0197] Preferably, the step of constructing a local nonlinear adjustment optimization model based on the original feasible region includes:

[0198] Based on the original feasible region, calculate the geometric center of the polyhedron corresponding to the original feasible region; assume the polyhedron is n-dimensional and has m vertices, where each vertex u i The coordinates are an n-dimensional vector u i =(u i1 ,u i2 ,…,u in If the geometric center is C = (C1, C2, ..., C...), then the coordinates of the geometric center are C = (C1, C2, ..., C...). n )for:

[0199]

[0200] Take N points u on each edge. old A local nonlinear adjustment optimization model is constructed, specifically as follows:

[0201] min‖u new -u old ||

[0202] The linear constraint (21) mentioned in st

[0203] Among them, u new This represents the vertex to be solved.

[0204] Preferably, the step of using deep learning methods to solve the local nonlinear adjustment optimization model to obtain the final feasible region includes:

[0205] The local nonlinear adjustment optimization model is solved in batches using a deep learning method based on the Optimized Neural Network (ONN) to obtain new feasible region vertices, forming a new feasible region. The ONN model is defined as follows:

[0206] y(t; w) = y0 + (1 - e -t )N(t;w) (22)

[0207] in, As the initial solution, The time range is defined as t∈[0,T], where 1-e is the time variable. -t Used to ensure that the ONN model always satisfies the initial condition y(0; w) = y0;

[0208] The endpoints of the state solutions of the ONN model are approximations of the optimal solutions of the local nonlinear adjustment optimization model, that is:

[0209]

[0210] Combining equations (22) and (23), we get:

[0211] P(y(T;w))≈y * (twenty four)

[0212] Where P(·) is the projection function that projects the endpoints onto the feasible set, and Equation (24) indicates that the endpoints of the ONN model together with the projection function are an approximate solution of the optimal solution of the local nonlinear adjustment optimization model;

[0213] The loss function of the ONN model is:

[0214]

[0215] Where φ(·) is the ordinary differential equation corresponding to the local nonlinear adjustment optimization model, e -γ*t Used to redistribute weights in the loss and assign higher weights to times closer to the origin; γ is a weighted hyperparameter.

[0216] The objective function of the ONN model is:

[0217]

[0218] Wherein, the objective function E(w) is the integral of the loss function over the time range [0,T], and the loss function L(T,w) represents the error of the ONN model at time t;

[0219] The model is trained by minimizing the batch loss:

[0220]

[0221] in, Let t be a set of times randomly selected from [0,T]. Indicates the size of the set;

[0222] After obtaining the new feasible region, its geometric center is recalculated, and iterative processes are performed based on the local nonlinear adjustment optimization model and the ONN model until the preset accuracy requirement is met, thus obtaining the final feasible region.

[0223] Preferably, the grid dispatching device that considers wind power uncertainties is further used for:

[0224] The final feasible region was evaluated using the Hausdorff distance, the Minkowski summation to volume ratio, the Jaccard exponent, and the mean square error of the support function, respectively, to obtain the evaluation results.

[0225] In specific implementation, the working principle, control process and technical effects of the grid dispatching device considering wind power uncertainty provided in the embodiments of the present invention are the same as those of the grid dispatching method considering wind power uncertainty in the above embodiments, and will not be repeated here.

[0226] This invention also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program, when running, controls the device where the computer-readable storage medium is located to execute the grid dispatching method considering wind power uncertainty as described in any of the above embodiments.

[0227] This invention provides a grid dispatching method, apparatus, and medium that considers wind power uncertainty. Its advantages include: by using an analytical approach to opportunity constraints based on sensitivity factors and first-order Taylor expansion, it can maximize the analytical processing of opportunity constraints while considering wind power uncertainty and ensuring no limit exceedances, transforming opportunity constraints into directly manageable methods; by employing an improved dual-description method to solve for feasible region vertices, it can address the challenge of vertex calculation in large-scale, high-dimensional scenarios, reducing the overall algorithm complexity and improving computational efficiency; and by using an ONN model, it can quickly and accurately solve large-scale nonlinear optimization models, further improving accuracy and computational efficiency. This invention can rapidly calculate feasible region vertices while ensuring the safety of operating points within the feasible region, meeting the requirements of real-time grid dispatching.

[0228] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A power grid scheduling method considering wind power uncertainty, characterized in that, The method comprises the following steps: According to the wind power uncertainty, an AC power flow based chance constraint model is constructed; The chance constraint model is analyzed by using the sensitivity factor and the first-order Taylor expansion, and the chance constraint in the chance constraint model is converted into a linear constraint; According to the linear constraint, an improved double description method is used to calculate the original feasible region of the power grid; wherein the linear constraint and the to-be-solved polyhedral vertex description problem are converted into a corresponding polyhedral cone problem by using the improved double description method; Based on the original feasible region, a local nonlinear adjustment optimization model is constructed, and an optimization neural network (ONN) based deep learning method is used to solve the local nonlinear adjustment optimization model to obtain a final feasible region, so that the power grid is dispatched according to the feasible region. 2.The method of claim 1, wherein, The step of constructing the chance constraint model based on the wind power uncertainty comprises the following steps: The new energy power is modeled as the sum of the wind power ultra-short-term point prediction value and the wind power uncertainty deviation; Based on the wind power uncertainty deviation, a generator power balance constraint, a node voltage constraint and a branch power flow constraint are constructed; The generator power balance constraint, the node voltage constraint and the branch power flow constraint are expressed as a chance constraint to obtain a chance constraint model. 3.The method of claim 1, wherein, The step of analyzing the chance constraint model by using the sensitivity factor and the first-order Taylor expansion to convert the chance constraint in the chance constraint model into a linear constraint comprises the following steps: According to the generator power, the voltage and the branch power flow, the sensitivity factor about the wind power uncertainty deviation is determined; According to the sensitivity factor, the chance constraint model is re-expressed to obtain a chance constraint containing the sensitivity factor; The first-order Taylor expansion is used to analyze the chance constraint containing the sensitivity factor to obtain a linear constraint. 4.The method of claim 3, wherein, The sensitivity factor specifically comprises: wherein, and denote the sensitivity factors of the generator injected active power and injected reactive power, respectively; is the sensitivity factor of the voltage; and denote the sensitivity factors of the branch active power flow and reactive power flow, respectively, is the wind power uncertainty bias, is the generator injected active power, is the generator injected reactive power, is the node voltage, is the branch active power flow, is the branch reactive power flow.

5. The method for power grid dispatching considering wind power uncertainty according to claim 1, characterized in that, The step of calculating the original feasible region of the power grid by using the improved double description method according to the linear constraint comprises the following steps: The linear constraint and the to-be-solved polyhedral vertex description problem are converted into a corresponding polyhedral cone problem; A vertex subset, a set of extreme rays and a set of adjacent pairs are initialized; By adding the inequality in the linear constraint step by step, the adjacent pairs and the vertices are updated; the updating of the adjacent pairs comprises: traversing each extreme ray in the geometry, testing the adjacency of each extreme ray and the vertex to screen the updated adjacent pairs; The pairs in the adjacent pairs that no longer generate new rays are deleted to obtain a final polyhedral vertex set, and the original feasible region of the power grid is determined.

6. The method for power grid dispatching considering wind power uncertainty according to claim 5, characterized in that, The step of constructing the local nonlinear adjustment optimization model based on the original feasible region comprises the following steps: Based on the original feasible region, calculate the geometric center of the polyhedron corresponding to the original feasible region; assume the polyhedron is n-dimensional and has m vertices, where each vertex... The coordinates are an n-dimensional vector Then the coordinates of the geometric center for: N points are taken on each side A local nonlinear adjustment optimization model is constructed, specifically: wherein, represents the vertex to be solved.

7. The method for power grid dispatching considering wind power uncertainty according to claim 6, characterized in that, The step of solving the local nonlinear adjustment optimization model by using the deep learning method to obtain the final feasible region comprises the following steps: The ONN model is defined as follows: (22) wherein, is the initial solution, is the time horizon, is the time variable; for ensuring that the ONN model always satisfies the initial condition ; The endpoint of the state solution of the ONN model is the approximation of the optimal solution of the local nonlinear adjustment optimization model, that is: (23) Combined with equations (22)-(23), we have (24) wherein, is a projection function that projects the end point onto the feasible set, and equation (24) represents that the end point of the ONN model together with the projection function is an approximate solution of the optimal solution of the local non-linear adjustment optimization model. A loss function of the ONN model is: (25) wherein, is the ordinary differential equation corresponding to the local non-linear adjustment optimization model, for re-distributing the weights in the loss and giving higher weights to times close to the origin, is a weighted hyper-parameter; An objective function of the ONN model is: (26) wherein the objective function is an integral over a time range of a loss function representing an error of the ONN model at time t; The model is trained by minimizing the batch loss: (27) wherein, is a set of times t randomly drawn from denotes the set size;​ After obtaining the new feasible region, the geometric center thereof is calculated again, and a loop iteration is performed based on the local nonlinear adjustment optimization model and the ONN model until a preset accuracy requirement is met, so as to obtain a final feasible region. 8.The method of claim 1, wherein, After obtaining the final feasible region, the method further comprises: The final feasible region is evaluated by using Hausdorff distance, Minkowski addition and volume ratio, Jaccard index, and mean square error of support function respectively, so as to obtain an evaluation result.

9. A power grid scheduling device considering wind power uncertainty, characterized in that, Comprise: An opportunity constraint construction module is configured to construct an AC power flow-based opportunity constraint model according to wind power uncertainty; A linear constraint transformation module is configured to analyze the opportunity constraint model by using a sensitivity factor and a first-order Taylor expansion, and transform the opportunity constraint in the opportunity constraint model into a linear constraint; A feasible region calculation module is configured to calculate an original feasible region of the power grid by using an improved double-description method according to the linear constraint; wherein the linear constraint and a vertex description problem of a polyhedron to be solved are polyhedral vertex description problems, and the improved double-description method transforms the polyhedral vertex description problem into a corresponding polyhedral cone problem; A feasible region optimization model is configured to construct a local nonlinear adjustment optimization model based on the original feasible region, and solve the local nonlinear adjustment optimization model by using an ONN-based deep learning method, so as to obtain a final feasible region, and perform power grid dispatching according to the feasible region.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when a device in which the computer-readable storage medium is located executes the computer program, the power grid dispatching method considering wind power uncertainty is implemented. The computer-readable storage medium stores a computer program, wherein when a device in which the computer-readable storage medium is located executes the computer program, the power grid dispatching method considering wind power uncertainty is implemented.

Images