Two-stage distribution robust unit combination optimization method considering inertia and wind power dual uncertainty

Abstract

The invention discloses a two-stage distribution robust unit combination optimization method considering inertia and wind power dual uncertainty. The method comprises the following steps: acquiring load prediction data of each synchronous generator set in a power system and a wind power plant power prediction value; considering inertia and wind power dual uncertainty, selecting an inertia prediction error probability density function and a wind power prediction error probability distribution function, and obtaining wind power and inertia multi-discrete scenes and corresponding scene probabilities by using Latin extraction scene generation and K-means scene clustering; deducing a dynamic frequency constraint containing an inertia error of the system in a frequency response process; building a two-stage distribution robust unit combinatorial optimization model containing dynamic frequency constraint; an auxiliary variable is introduced to linearize absolute value constraints in the confidence set, and Camp is adopted; the CG algorithm divides the optimization model into a main problem and a sub-problem for iterative solution. According to the method, the uncertainty of inertia prediction and the uncertainty of wind power are considered, so that the unit combination model has higher frequency safety and economical efficiency.

Description

Technical Field

[0001] The present invention relates to the field of power system optimization operation methods, and in particular to a two-stage distributed blue-rod unit combination optimization method taking into account dual uncertainties of inertia and wind power. Background Art

[0002] With the increase in the penetration rate of new energy sources and the large-scale replacement of conventional power sources by non-synchronous power sources, the inertia level of the power system continues to decrease, and the frequency stability problem becomes increasingly prominent. With the development of virtual inertia technology in high-proportion power systems, the inertia forms of the system have become more diversified, covering all factors involved in suppressing frequency changes before the speed control system takes effect. The generalized inertia of the system is defined as the sum of all inertia forms that can respond inertially, including the rotational inertia of the synchronous machine, the virtual inertia of the voltage source virtual synchronous control, the virtual inertia of the current source inertia control, and the static load at the load level, and the inertia provided by the asynchronous motor. In the research on the unit combination model considering dynamic frequency constraints, most of them only consider the inertia of the synchronous generator with known parameter information of the dispatched department, and do not explore the joint role of other inertia resources in the model.

[0003] With the development of system inertia assessment and system inertia prediction, the current system inertia assessment can establish historical data samples of system equivalent inertia. By subtracting the unit inertia of the parameter information mastered by the dispatching department from the system equivalent inertia, the historical data samples of the remaining inertia of the system can be obtained. At the same time, through artificial intelligence and data-driven technology, the trend prediction of inertia that is not mastered by the dispatching department in the future period can be realized based on information such as new energy and load forecasts. However, due to the uncertainty of the prediction model itself and the prediction data information, the prediction results of inertia also have a high degree of uncertainty. Therefore, it is necessary to consider the impact of this factor in the day-ahead dispatch.

[0004] Based on the above background, incorporating the equivalent inertia that is not known by the dispatching department into the frequency response system, improving the overall inertia system of the system, and considering the uncertainty of the conventional inertia point prediction model and the uncertainty of wind power are of great significance to the frequency safety and stability of day-ahead dispatching. Summary of the invention

[0005] The present invention provides a two-stage distributed robust unit combination optimization method taking into account the dual uncertainties of inertia and wind power. The method comprehensively considers the influence of the overall inertia of the system and the uncertainty of inertia prediction and wind power uncertainty, so that the unit combination model has more frequency safety and economy.

[0006] The technical solution adopted by the present invention is:

[0007] A two-stage distributed blue-rod unit combination optimization method considering dual uncertainties of inertia and wind power includes the following steps:

[0008] Step 1: Obtain the technical parameters, economic parameters, load forecast data and wind farm power forecast value of each synchronous generator set in the power system, as well as the RoCoF (Rate of Change of Frequency) protection setting value and frequency extreme value allowable range in the power system;

[0009] Step 2: Considering the dual uncertainties of inertia and wind power, select the probability density function of inertia prediction error and the probability distribution function of wind power prediction error, use Latin abstraction scenario generation and K-means scenario clustering to obtain multiple discrete scenarios of wind power and inertia and the corresponding scenario probabilities;

[0010] Step 3: In the frequency response process, taking into account the inertia that is not known by the dispatching department, the dynamic frequency constraint of the system including the inertia error is derived according to the dynamic frequency stability range;

[0011] Step 4: Based on the initial scenario probability distribution, the 1-norm and ∞-norm probability distribution uncertain confidence sets are constrained to build a two-stage distributed robust unit commitment optimization model with dynamic frequency constraints;

[0012] Step 5: Auxiliary variables are introduced to linearize the absolute value constraints in the confidence set, and the column-and-constraint generation (C&CG) algorithm is used to divide the two-stage distributed robust unit combination optimization model into the main problem and sub-problems for iterative solution.

[0013] In the step 1, the technical parameters, economic parameters, load forecast data and wind farm power forecast values ​​of the synchronous generator set are as shown in Table 1 of the embodiment section, the inertia and operating parameters of the conventional units of the system, Table 2, the cost parameters of the conventional units of the system, Figure 2 shown.

[0014] RoCoF protection setting value and frequency extreme value allowable range in power system. The maximum RoCoF of China's power system is set to no more than 2 Hz / s to ensure safe operation of the unit. For the incremental distributed power sources, the setting value of RoCoF protection is adjusted to 1 Hz / s, and a delay time of 500 ms is set. Frequency extreme value allowable range: In China's power system, the frequency deviation limit under normal operating conditions is ±0.2 Hz. When the system capacity is small, the deviation limit can be relaxed to ±0.5 Hz. The specific formula descriptions are shown in Equations (26) and (27).

[0015] In step 2, the inertia in the system that is not known by the dispatching department is calculated by using E R Inertia representation, consisting of two parts, one is based on the inertia point prediction Part of it is the inertia prediction error △E R , the expression is as follows:

[0016]

[0017] Gaussian distribution is used as the probability density function of inertia prediction error to characterize the uncertainty of inertia point prediction error;

[0018] △E R ~N(0,σ 2 ) (2);

[0019] Where: σ is the inertia prediction error △E R The standard deviation of N(0,σ 2 ) represents the Gaussian distribution of the predicted value of the inertia point.

[0020] In step 2, wind power is an intermittent energy source, and wind power output is volatile and random. Its grid-connected operation will have a significant impact on the performance of the system. The present invention uses a wind power scenario generation method based on historical statistical data, assuming that the wind power prediction error probability distribution function is a normal distribution, as shown in formula (4); then sampling is performed to generate a series of wind power output prediction error values ​​that obey a given distribution, which are superimposed on the wind power prediction value to obtain a series of wind power output scenarios; random variables are used to characterize the uncertainty of wind power prediction errors. Assume that the wind power output is The wind power forecast is The wind power prediction error is △P Wt , in MW, then:

[0021]

[0022] Normal distribution is used to describe the uncertainty of wind power forecast error.

[0023]

[0024] Where: Represents the normal distribution of wind power forecast error.

[0025] The inertia point prediction value mentioned above is the IEEE10 machine 39 node inertia point prediction value obtained by the existing prediction model. The innovation of the present invention is that the existing inertia point prediction value does not take into account the uncertainty factor of wind power output, and the present invention uses Gaussian distribution as the inertia prediction error probability density function to characterize the uncertainty of the inertia point prediction error. The data and prediction results are shown in Table 1 of the embodiment section. The inertia and operating parameters of the conventional units of the system, Figure 2 shown.

[0026] In step 2, the Latin hypercube sampling scenario is generated as follows:

[0027] The probability distribution functions of wind power error and inertia error are shown in equations (2) and (4). The normal distribution and Gaussian distribution models are still used. Latin hypercube sampling (MCS) is an efficient Monte Carlo simulation method proposed by MD Mckay, RJ Beckman and WJ Conover in 1979. Its essence is stratified sampling based on inverse function transformation. Without changing the original density function, the distribution function values ​​are divided into N non-overlapping equally spaced sub-intervals, and the sampled values ​​in the sub-intervals are transformed by inverse functions based on the probability density function. The obtained sampling results can both meet the characteristics of the probability distribution function and cover the entire distribution interval.

[0028] Suppose there are M random variables x1, x2, x3, … x M ,x1,x2,x3,…x M is a random variable, that is, the dimension of the variable is M, the number of samples required for each variable is N, and the cumulative probability distribution function of each variable is F xi =f i (x i ), x i Indicates that the i-th random variable, f i (x i ) represents the cumulative probability distribution function of each variable, where i = 1, 2, ..., M. The Latin hypercube sampling method is used for sampling, and the process is as follows:

[0029] A1: Define the sampling size as N, and take any random variable x i The cumulative probability distribution function curve F xi =f i (x i ) is divided into N intervals with equal probability, and the width of each interval is

[0030] A2: For a random variable x i The kth sample value x ik , and its corresponding cumulative distribution probability is:

[0031]

[0032] Where: U~N(0,1) obeys uniform distribution; k=1,2,…,N; calculate the cumulative distribution function F xi The inverse function of x can be obtained i The kth sample value xik , satisfying the following formula:

[0033]

[0034] Where: is the cumulative distribution function F xi The inverse function of .

[0035] A3: After sampling, the sampled values ​​of each random variable are arranged into a row of a matrix to form an M×N sampling matrix X MN .

[0036] In step 2, K-means scene clustering is performed as follows:

[0037] K-Means clustering algorithm is a classic partitioning clustering method proposed by Mcqueen in 1967. It has the advantages of simplicity, speed and good scalability. Assuming there is a data set D containing n Euclidean space objects, K-Means clustering is to divide the data objects in D into k clusters (C1, C2, ..., C k ), and for i,j∈{1,2,…,k}, the partitioning must satisfy The basic process of the algorithm is as follows:

[0038] B1: Input a data set D containing n data sample points = {x i |x i ∈R m ,i=1,2,…,n},x i is the data sample point; R m is a real number set of m-dimensional vectors, n is the number of data sample points; the number of clusters that need to be divided is k.

[0039] B2: Randomly select k sample points y1,y2,…,y k There are k clusters C1, C2, …, C k The initial cluster center of

[0040] Among them, y1,y2,…,y k are k sample points; C1, C2, …, C k The clusters into which the dataset is divided.

[0041] B3: Calculate the Euclidean distance d(x i ,y j ), y j To randomly select k sample points, the data sample x i Divide into cluster C j So that x i and i Minimum distance.

[0042]

[0043] B4: Calculate the error square sum criterion function E after clustering. Recalculate the new cluster center of each cluster, such as cluster C j ,j=1,2,…,k, and the new cluster center y can be calculated using formula (9): j ;

[0044]

[0045] In the above formula, E is the error square sum criterion function after clustering; C k The clusters into which the dataset is divided.

[0046] B5: Recalculate the criterion function E′ according to the new cluster center. If |EE′| does not meet the specified threshold, repeat steps B3 and B4. If it does, output the clustering result.

[0047] In step 3, the system equivalent inertia includes the synchronous unit inertia and the E introduced above. R Inertia can be expressed as:

[0048]

[0049] Where: E SYS is the total inertia of the system, in MW·s; E G is the generator inertia, in MW·s; E R is the collection of inertia forms such as virtual inertia load side inertia, and the unit is MW·s; S is the total number of generator sets i; S Ni is the rated capacity of generator i, in MW; H i is the inertia time constant of generator i, in seconds; U i The state quantity representing the start and stop of generator i, 0 means stop, and 1 means start; is the inertia prediction result, in MW·s; △E R is the prediction error, in MW·s.

[0050] The constraint containing the maximum frequency change rate is:

[0051]

[0052] Where: |RoCoF set,max | is the maximum frequency change rate allowed by the system, in Hz / s; △P L is the active unbalance amount caused by the disturbance, in MW; f n is the system rated frequency;

[0053] The constraint containing the frequency minimum point is:

[0054]

[0055] Where: f0 is the original operating frequency of the system, f set,min is the lowest frequency allowed by the system, in Hz; R SSi is the frequency modulation response rate of each synchronous generator, in MW / s; U it The start and stop status of the unit in the time period.

[0056] In step 4, considering the uncertainty of wind turbine output and inertia prediction, as well as the limitations of historical prediction error data, the scene probability distribution obtained by scene clustering has a certain error. Therefore, the distributed robust optimization method is used to limit the probability distribution of the scene. First, the prediction data and the prediction error probability density function are sampled using LHS to obtain N original scene numbers, and then the previous K-Means clustering algorithm is used to obtain N s A typical finite discrete scenario and this N s The initial probability distribution corresponding to a typical finite discrete scenario is the initial probability distribution of a typical finite discrete scenario, and the probability distribution fluctuation range is constrained based on the 1-norm and ∞-norm.

[0057] Probability distribution is the probability distribution of a typical finite discrete scenario, and the confidence can be expressed as:

[0058]

[0059] Where: P r (·) is the probability function; θ1 and θ ∞ is the allowable deviation value of the probability distribution; Initial probability distribution of finite discrete scenarios; N s is the number of typical finite discrete scenes; N is the number of original scenes; 1-2N in formula (13) s exp(-2Nθ1 / N s ) is set to α1, and 1-2N s exp(-2Nθ ∞ ) This part is set to α ∞ , then α1 represents the confidence level satisfied by the probability distribution p based on the 1-norm, α ∞ Denotes the confidence level, θ1 and θ1 satisfied by the probability distribution p based on the ∞-norm respectively. ∞ It can be expressed by the following formula:

[0060]

[0061] The confidence set that can derive the probability distribution is as follows:

[0062]

[0063] Where: ψ is the probability distribution set interval of the scene, which represents the confidence set constrained by the 1-norm and ∞-norm; {p s}The set of probability distributions corresponding to typical finite discrete scenarios.

[0064] In step 5, the wind power and inertia uncertainty set is as shown in formula (15). The 1-norm and ∞-norm constraints in formula (15) contain absolute value terms and are nonlinear constraints. Therefore, auxiliary variables and auxiliary constraints need to be added to transform them into linear constraints. The added auxiliary variables and auxiliary constraints are as follows:

[0065]

[0066] Where: is the probability distribution p of scene s s Relative to The positive offset of is the probability distribution p of scene s s Relative to The negative offset of and They are the 1-norm constraints on p s A 0-1 sign that generates positive and negative offsets; and They are the ∞-norm constraints for p s A 0-1 sign that generates positive and negative offsets; Represents any scene.

[0067] The transformed 1-norm and ∞-norm constraints are shown in equation (17) and equation (18) respectively.

[0068]

[0069] In step 5, the specific process of solving the main problem and sub-problems of the two-stage Bruker unit combination optimization model by stage-by-stage iterative solution using the dualization-free C&CG algorithm is as follows:

[0070] Step 1: Set the lower bound LB = -∞, the upper bound UB = -∞, and the number of iterations n = 1;

[0071] Step 2: Solve MP and get the optimal solution (x * ,η * ), x * , η * Represents the optimal solution of the objective function of the main problem, and updates the lower bound value LB = A T x* +η * , A T Represents the main problem objective function matrix.

[0072] Step 3: Fix the first stage variable x * , solve the subproblem and obtain the probability value in the worst case And the objective function value f sp (x * ). Update the upper bound min{UB,A T x * +f sp (x * )};

[0073] Step 4: If UB-LB < ε, ε represents an infinitesimal number, then stop the iteration and return the optimal solution x * ; Conversely, update the worst probability distribution of the main problem And add new variables to the main problem And add the main question with Relevant constraints.

[0074] step5: Update n=n+1 and return to step2.

[0075] The present invention provides a two-stage distributed blue-rod unit combination optimization method taking into account the dual uncertainties of inertia and wind power, and the technical effects are as follows:

[0076] 1) The two-stage distributed robust unit combination model established by the present invention taking into account the dual uncertainties of inertia and wind power takes into account dynamic frequency constraints and shows better results in frequency stability.

[0077] 2) The two-stage distributed robust unit combination model established by the present invention, which takes into account the dual uncertainties of inertia and wind power, takes into account the influence of inertia from multiple sources, and has better economic advantages compared to the traditional two-stage unit combination model which only takes into account the inertia of the synchronous units on the power supply side.

[0078] 3) The model proposed in the present invention takes into account the dual uncertainties of wind power and inertia. Compared with the deterministic model, that is, the situation where both wind power and inertia are predicted values, it can effectively reduce the wind curtailment and reduce the cost of wind curtailment. At the same time, it can reduce the risk of over-limit of the system frequency change rate and the lowest frequency point when the system is disturbed, thereby ensuring the frequency safety of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 This is the improved IEEE10 machine 39-node wiring diagram.

[0080] Figure 2 It is the wind power and load prediction curve.

[0081] Figure 3 Four typical scenarios of wind farm 1 are shown in the figure.

[0082] Figure 4 Four typical scenarios of wind farm 2 are shown in the figure.

[0083] Figure 5 Schematic diagram of four typical scenarios of inertia.

[0084] Figure 6 This is the unit state change diagram of Scheme 1;

[0085] Figure 7 This is the unit state change diagram of Scheme 2;

[0086] Figure 8 This is the unit state change diagram of Scheme 3.

[0087] Fig. 9 This is a schematic diagram of the maximum frequency change rate and the lowest frequency point in each time period for each scheme in scenario 1.

[0088] Fig.10 Schematic diagram of the maximum change rate of frequency of each scene in the two models.

[0089] Fig.11 Schematic diagram of the lowest frequency point of each scene in the two models. DETAILED DESCRIPTION

[0090] A unit commitment optimization method considering inertia uncertainty and dynamic frequency constraints is proposed. The specific unit commitment model is as follows:

[0091] The first stage is the optimization of unit start-up and shutdown before the day, with the objective function being to minimize the unit start-up and shutdown cost; the second stage is the optimization of unit power dispatch within the day, with the objective function being to minimize the sum of conventional unit operating costs and wind curtailment costs.

[0092] 1. First stage model:

[0093] 1.1 Objective function of the first stage:

[0094] The objective function of the first stage consists of the startup and shutdown costs of conventional units.

[0095]

[0096] Where: T is the number of time periods; N is the total number of conventional units; U it S is the start and stop status of the unit; it (τ it ) is the startup cost of unit i in period t; U i(t-1) is the start and stop status of conventional unit i in time period t-1.

[0097] 1.2 Constraints of the first stage:

[0098] (1) Minimum start-stop time constraints:

[0099]

[0100] Where: are the continuous start-up time and downtime of the i-th generator in the first t-1 period; T i on , T i down are the minimum continuous start and stop time of the i-th generator; U i(t-1) is the start and stop status of conventional unit i in time period t-1.

[0101] 2. Second stage model:

[0102] 2.1 Second stage objective function:

[0103] The objective function of the second stage is composed of the operating cost of conventional units and the wind curtailment cost of wind turbines.

[0104]

[0105] Where: p s N is the occurrence probability of the sth discrete scene after scene clustering screening; s is the total number of discrete scenarios; T is the number of hours in the study period; N is the total number of conventional units; M is the total number of conventional units; U it The start and stop state of the conventional unit i; P it,s is the output active power of conventional unit i in time period t in the sth discrete scenario; i , b i and c i is the unit operating parameter; w is the penalty parameter for wind abandonment; Predict the output value of wind turbines. For The wind power output corresponding to the scenario s generated under the wind power forecast output sampling clustering; P Wjt.s It is the actual output value of the fan.

[0106] 2.2 Second stage constraints:

[0107] (1) System power balance constraints:

[0108]

[0109] Where: P Lt is the load forecast value of the system in period t.

[0110] (2) Generator output constraints:

[0111] U it P i,min ≤P it,s ≤U it P i,max (twenty three);

[0112] Where: P i.max , P i,min are the upper and lower limits of the output of the i-th generator respectively.

[0113] (3) Generator climbing constraints:

[0114] D i ≤P it -P i(t-1) ≤L i (twenty four);

[0115] Where: L i , D i are the upper and lower limits of the climbing slope of unit i respectively.

[0116] (4) Fan output constraints:

[0117]

[0118] (5) Maximum frequency change rate constraint:

[0119]

[0120] Where: |RoCoF max |RoCoF is the maximum rate of change of frequency obtained, set,max | is the maximum frequency change rate allowed by the system, in Hz / s; △P L is the active unbalance amount caused by the disturbance, in MW; f n is the system rated frequency; N is the total number of generator sets i; S Ni is the rated capacity of generator i, in MW; H i is the inertia time constant of generator i, in seconds; U i is the state quantity that represents the start and stop of generator i (equal to 0 means stop, equal to 1 means start); is the predicted value of inertia, For The inertia output value corresponding to scenario s generated by inertia prediction output and prediction error sampling clustering, in MW·s;

[0121] (6) Frequency minimum point constraint:

[0122]

[0123] Where: f min is the lowest point of the system frequency, f0 is the original operating frequency of the system, and f set,min is the lowest frequency allowed by the system, in Hz; R SSi It is the frequency regulation response rate of each synchronous generator, in MW / s.

[0124] 3 Two-stage distributed robust optimization model:

[0125] The two-stage objective function and constraints mentioned above are written in the form of a matrix, with x representing the variables of the first stage, which are the start and stop states of the conventional units. s Represents the variables of the second stage. The variables of the second stage are unit power, fan power, etc. The data-driven two-stage distributed robust optimization model can be expressed as follows:

[0126]

[0127] Where: x is the first-stage variable; A is the cost coefficient corresponding to the first-stage variable x; p s is the occurrence probability of the sth discrete scene after scene clustering screening; ψ is the probability distribution set interval of the scene; y s is the second-stage variable; Λ is the cost coefficient of the quadratic term in the objective function; B is the cost coefficient of the linear term in the objective function; ξ s It represents the wind power and inertia output value corresponding to the sth discrete scenario after scenario clustering screening, and K is the corresponding cost coefficient;

[0128] Formula (29) represents the first stage constraint, corresponding to the conventional unit start-stop constraint formula (20);

[0129] Formula (30) represents the equality constraints of the variables in the second stage, corresponding to formula (22);

[0130] Equation (31) represents the inequality constraints involving the second-stage variables, corresponding to equations (23) and (24);

[0131] Formula (32) represents the inequality constraints between the second-stage variables and the wind power in the discrete scenario, corresponding to formula (25);

[0132] Formula (33) represents the inequality constraint between the variables in the first stage and the inertia in the discrete scenario, corresponding to formula (26);

[0133] Equation (34) represents the inequality constraint between the first-stage variables and the inertia in the discrete scenario, corresponding to equation (27).

[0134] 4. Model solution:

[0135] 4.1 Linearize the absolute value constraint:

[0136] The wind power and inertia uncertainty set mentioned above is shown in formula (15). The 1-norm and ∞-norm constraints in the formula contain absolute value terms and are nonlinear constraints. Therefore, auxiliary variables and auxiliary constraints need to be added to transform them into linear constraints. The added auxiliary variables and auxiliary constraints are shown in formula (16). The transformed 1-norm and ∞-norm constraints are shown in formula (17) and formula (18), respectively.

[0137] 4.2 The model is divided into main and sub-problems to solve:

[0138] The model is decomposed into main problem and sub-problems, and the C&CG algorithm is used to solve it iteratively in stages.

[0139] (1) Main question:

[0140] The main problem is to find the optimal solution that satisfies the system economy under the premise of known scenario probability distribution p, which can be expressed as:

[0141]

[0142] Where: η is the given threshold; n is the total number of model iterations.

[0143] (2) Sub-questions:

[0144] The subproblem is given the first stage variable x * Then the solution is performed. The superscript "*" indicates the optimal solution of the corresponding variable. The subproblem is a max-min double-layer structure, which can be shown as follows:

[0145]

[0146] The above formula means that the variable y s Under the condition that it can be flexibly adjusted according to the scene changes, when the result of the main problem is known to be x * , find the worst scenario probability distribution within the confidence interval. There is a max-min double-layer optimization structure in the subproblem formula (43), which cannot be solved directly. The traditional approach is to first construct the dual problem of the inner min structure, and combine it with the outer max problem to transform it into a single-layer optimization problem, and then introduce auxiliary variables and auxiliary constraints to transform it into a mixed integer linear programming problem that can be solved directly. However, the discrete probability values ​​of each scenario in the outer max problem in the present invention are independent of the variables in the inner min problem, that is, the feasible domain ψ of the outer max optimization does not overlap with the feasible domain Y of the inner min problem. For each determined p sThe optimal solution of the inner optimization is unique, so there is no need to use the dual theory for transformation. The max-min problem is directly divided into two layers and solved independently, that is, the inner minimum optimization problem is solved first, and then the outer max problem is solved. The process is shown in equations (44) and (45) respectively.

[0147]

[0148] Where: h s The inner optimization result is obtained by bringing the unit start-stop plan solved by the main problem into the formula and then optimizing and solving it.

[0149] The implementation case provided by the present invention is an improved IEEE10 machine 39 node system, where the units at bus 38 and bus 39 are modified to wind turbines with rated capacities of 1200MW and 1800MW respectively. There are 8 conventional generators in the calculation example, the system installed capacity is 4766MW, and the wind farms are located at nodes 38 and 39 respectively. The system network structure wiring diagram is as follows: Figure 1 As shown in Figure 2. The 24h wind power and load forecast curves are shown in Figure 2. Figure 2 As shown, the parameters of conventional units are shown in Tables 1 and 2. The probability density function of wind power and inertia prediction error is set to normal distribution, with a mean of 0 and a standard deviation of 20% of the prediction value. The Latin hypercube sampling method mentioned above is used to generate 1000 scenarios, and the K-means clustering algorithm is used to sample and generate 4 typical scenarios. The probability corresponding to each typical scenario is shown in Table 3. The corresponding output of the four typical scenarios of wind farm 1 is as follows Figure 3 As shown in the figure, the output of wind farm 2 in four typical scenarios is as follows Figure 4 As shown, four typical scenarios of inertia are as follows Figure 5 As shown, α1 and α in the above formula (14) ∞ Set to 0.99. Set the RoCoF protection setting value in the example system |RoCoF set,max | is 0.5Hz / s, the system rated frequency f n is 50Hz, the frequency limit f set,min is 49Hz, and the power of disturbance is 15%P L The simulation uses MATLAB / YALMIP to build the unit combination model and uses the solver Gurobi to solve it.

[0150] Table 1 System conventional unit inertia and operating parameters

[0151]

[0152] Table 2 System conventional unit cost parameters

[0153]

[0154] Table 3 Probabilities corresponding to four typical scenarios

[0155]

[0156] (1) Analysis of the impact of different numbers of historical data on optimization results:

[0157] In the example, the prediction errors of wind power and inertia are set to follow a normal distribution with a mean of 0 and a standard deviation of 0.2 times the predicted value. Latin cube extraction is used to generate N = 1000 data, and clustering is used to obtain N s = 4 typical scenarios for analysis. In this section, when verifying the influence of different numbers of historical data on the optimization results in formula (14), Latin cube extraction is used to generate 100 to 1000 scenarios, and 4 typical scenarios are still generated by clustering. The results are shown in Table 4. It can be seen that as the number of historical scenarios increases, the total cost gradually decreases from 6788738.23 yuan when the initial historical data is 100 to 6769971.93 yuan when the final historical data is 1000. The total cost continues to decrease with the increase in the number of scenarios. This is mainly because as the number of historical scenarios increases, the information of the initial probability distribution is closer to the true value, the feasible domain of the probability distribution value becomes smaller, the conservatism of the model solution is reduced, and the economic efficiency of the result is better.

[0158] Table 4 Comparison of results with different numbers of historical data

[0159]

[0160] (2) Analysis of the impact of different confidence levels on optimization results:

[0161] It can also be seen from formula (14) that the confidence levels α1 and α ∞ When α1 is different, the optimization results will also change. The comparison results are shown in Table 5 and Table 6. It can be clearly seen from these two tables that when α1 remains unchanged, as α ∞ As α increases, the total cost of the optimized unit also increases. ∞ When α1 is constant, the optimized total cost of the unit also increases as α1 increases. This is because as the confidence level increases, the confidence interval will increase, the uncertainty of the random variable will increase, and a worse expected value of economic cost will be produced when solving the model.

[0162] At the same time, it can be found from Table 5 that when the value of α1 is larger, some cost results do not increase with the increase of α1, which means that the optimization result at this time is not affected by the confidence level α1, that is, the 1-norm constraint, but is mainly affected by the ∞-norm. The comparison between Table 5 and Table 6 shows that when the number of historical data increases to 1000, the system cost increases with α1. ∞ = 0.5 and α ∞=0.8, it does not increase with the increase of α1, which means that the optimization result at this time is mainly affected by the ∞-norm and less affected by the 1-norm.

[0163] Table 5 Optimization results at different confidence levels when N = 100

[0164]

[0165] Table 6 Optimization results at different confidence levels when N = 1000

[0166]

[0167] Further comparison of the difference between the comprehensive norm and only considering the 1-norm or ∞-norm constraints, when only considering the 1-norm, set α1 = 0.5, 0.5 ≤ α ∞ ≤0.99, only considering the ∞-norm, set α ∞ =0.99, 0.2≤α1≤0.99. Under the three conditions of considering the comprehensive norm, considering only the 1-norm, and considering only the ∞-norm, the distributed robust optimization is performed respectively, and the costs of the three cases when the historical data N=100 are shown in Tables 7 and 8 below, and the costs of the three cases when the historical data N=1000 are shown in Tables 9 and 10 below.

[0168] It can be found from Table 7 that, at the same confidence level, the unit cost is lower under the comprehensive norm constraint than when only the 1-norm constraint is considered. It can be found from Table 8 that, at the same confidence level, the unit cost is lower under the comprehensive norm constraint than when only the ∞-norm constraint is considered, and it has better economy. This shows that when only the 1-norm or ∞-norm constraint is considered, the system model is more conservative and the total unit cost solved is also higher, while the comprehensive constraint has a lower degree of conservatism and is more economical. It is studied whether the optimization result will change significantly when the historical data increases. It can be found from Tables 9 and 10 that when the historical data increases, at the same confidence level, the optimization result obtained by considering the comprehensive constraint is still less than the optimization result under the 1-norm or ∞-norm constraint. Therefore, the optimization result still maintains the same rule as the historical data increases.

[0169] Table 7 Optimization results of comprehensive norm and ∞-norm constraints when N = 100

[0170]

[0171]

[0172] Table 8 Optimization results of comprehensive norm and 1-norm constraints when N = 100

[0173]

[0174] Table 9 Optimization results of comprehensive norm and ∞-norm constraints when N = 1000

[0175]

[0176] Table 10 Optimization results of comprehensive norm and 1-norm constraints when N = 1000

[0177]

[0178] (3) Scheduling results under different schemes:

[0179] In order to demonstrate the effectiveness and superiority of the model proposed in the present invention, three schemes are set for comparison (all considering the uncertainty of wind power).

[0180] Scheme 1: The traditional two-stage unit commitment model for the day before and within the day does not consider frequency-related constraints, and its objective function is to minimize the sum of the initial probability distribution of power generation costs;

[0181] Scheme 2: A two-stage unit commitment model within the day before the day that only considers the dynamic frequency constraint of the synchronous generator inertia on the power supply side, and its objective function is to minimize the sum of the initial probability distribution of the power generation cost;

[0182] Solution 3: The method proposed in the present invention is a two-stage day-ahead and day-intraday distributed robust unit combination model with dynamic frequency constraints considering inertia uncertainty, and its objective function is to minimize the sum of the worst probability distribution of power generation costs.

[0183] Table 11 Unit combination cost under each scheme

[0184]

[0185] Table 11 shows the unit combination costs of the above three options. Among them, option 2 has the highest cost, option 3 has the highest cost, and option 1 has the lowest cost. Figure 6 , Figure 7 and Figure 8 They are the unit state change diagrams under three schemes. Fig. 9 is the maximum frequency change rate and the lowest frequency point under the three schemes in scenario 1. Fig. 9It can be found that although the lowest frequency points of the three schemes meet the requirements and the values ​​are all above 49Hz, their maximum frequency change rates do not all meet the requirements. For example, the maximum frequency change rate of scheme one that does not consider the dynamic frequency constraint obviously exceeds the limit of 0.5Hz / s, and there is a frequency safety hazard. The two frequency indicators of schemes two and three that take into account the dynamic frequency constraint are both within the safe range. From the following three unit state change diagrams, it can be found that in order to meet the dynamic frequency constraint and make the system have sufficient inertia, the system will add more units to deal with power disturbances. Compared with scheme two that only takes into account the inertia of synchronous generators, scheme three adds the consideration of the influence of multiple inertia resources, so there are fewer units in operation and the total cost is also smaller. This shows that the scheme of the present invention has good economy while ensuring frequency safety and stability.

[0186] (4) Comparative analysis with deterministic methods:

[0187] This section will compare and analyze the decision results of the distributed robust optimization model and the traditional two-stage deterministic model. The deterministic model obtains the conventional unit start-stop plan, conventional unit and wind turbine unit output plan under the condition that the wind power output and inertia are predicted values ​​(this section does not take into account the rotating reserve). The unit cost results and the two-stage unit cost results of the present invention are shown in Table 12. It can be seen that the unit combination model considering dual uncertainty has increased by 0.31% compared with the calculation results of the deterministic unit combination model. Considering uncertainty factors in the unit combination model will increase the expected value of the system operating cost. In order to verify that the unit combination model considering dual uncertainty in the present invention has better robustness after paying the price of reducing economic efficiency, the unit combination scheme obtained by the deterministic model is based on the unit combination scheme obtained by the deterministic model, and is brought into the four typical scenarios of wind power and inertia mentioned above, and economic dispatch is performed respectively. The total operating costs of the unit combination schemes obtained by the two models under different possible system operating scenarios are compared, and the results are shown in Table 13. The wind curtailment costs of the unit combination schemes obtained by the two models under different possible system operation scenarios are compared, and the results are shown in Table 14. The maximum frequency change rate and the minimum frequency point of each scenario of the two models are shown in Table 14. Fig.10 , Fig.11 As shown in Table 13, Table 14 and Fig.10 , Fig.11It can be seen that the two-stage distributed robust model of the present invention comprehensively considers wind power output and inertia in different scenarios for optimization. The frequency indicators of the system are all within the limit value, which can meet the requirements of operation under safe conditions, and the wind abandonment cost in each scenario during the day is zero. The total cost is relatively low, and it can better adapt to the system under uncertain conditions. The unit combination scheme generated by the deterministic model and the start-stop scheme derived from the predicted scenario will generate huge wind abandonment costs when the wind power scenario changes during the day. For example, the total cost under scenario 1 is 38,490,981.67 yuan, an increase of 469% compared to 6,762,824.03 yuan in the model of the present invention. At the same time, from Fig.10 It can be clearly found that the frequency change rate of the deterministic model in scenario 1, scenario 3, and scenario 4 exceeded 0.5Hz / s at 11 o'clock. This shows that the solution generated by the two-stage deterministic model based on the inertia prediction value optimization will have frequency safety risks when the inertia fluctuates during the day, threatening system safety and having poor applicability. Therefore, it is very necessary to consider the dual uncertainty factors of wind power and inertia in the unit combination model.

[0188] Table 12 Unit combination cost under each scheme

[0189]

[0190] Table 13 Total unit combination cost under different scenarios of two models

[0191]

[0192] Table 14 Wind curtailment cost of unit combination under different scenarios of two models

[0193]

[0194] (5) Comparative analysis with uncertainty methods

[0195] This section will compare and analyze the decision results of the distributed robust optimization model and the conventional stochastic optimization model. The conventional stochastic optimization model uses the typical wind power scenarios and inertia scenarios in the examples in this chapter, and adopts the initial probabilities under the corresponding typical scenarios. It optimizes the objective function with the optimal sum of the probability distribution of the cost, and calculates the minimum expected cost of the system. The unit combination cost results of the two models are shown in Table 15 below.

[0196] Table 15 Cost results of two models

[0197]

[0198] It can be seen from Table 15 that the optimization result corresponding to the conventional stochastic programming is more economical than the two-stage distributed robust optimization model. This is because the distributed robust optimization model considers the confidence interval of the probability distribution on the basis of the conventional stochastic optimization model and solves the optimization result under the worst distribution. Therefore, the start-up and shutdown costs and operating costs of the conventional stochastic optimization model and the distributed robust model are the same, but the probability values ​​of each scenario are different. The scenario probability value of the conventional stochastic optimization model is the initial value, while the probability value of each scenario of the distributed robust model is the worst probability value. Therefore, the total cost is increased compared with the conventional stochastic optimization model. Although the distributed robust model is inferior to the conventional stochastic optimization model in terms of economy, it better balances the relationship between system economy and conservatism and shows strong adaptability.

[0199] The present invention aims at the uncertainty problem of renewable energy output and inertia not mastered by the dispatching department in the unit combination. According to the predicted values ​​of wind power and inertia and their error probability density functions, the Latin hypercube sampling method and K-Means clustering are used to obtain multiple discrete scenarios. The 1-norm and ∞-norm constraints of the probability distribution uncertainty confidence set are comprehensively considered. The start-stop output of conventional units and the output of new energy wind power are used as optimization variables. The power generation cost and the amount of wind abandonment are minimized to establish a two-stage distributed blue-robust unit combination model within the day before. The two-stage distributed blue-robust unit combination model established by the invention takes into account the influence of inertia from multiple sources, and has better economic advantages compared with the traditional two-stage unit combination model that only takes into account the inertia of the synchronous unit on the power supply side. The proposed model takes into account the dual uncertainty of wind power and inertia. Compared with the deterministic model, that is, the situation where both wind power and inertia are predicted values, it can effectively reduce the wind abandonment situation and reduce the wind abandonment cost. At the same time, it reduces the risk of over-limit of the system frequency change rate and the lowest frequency point when the system is disturbed, and ensures the frequency safety of the system.

Claims

1. A two-stage distributed blue-rod unit combination optimization method considering the dual uncertainty of inertia and wind power, characterized in that The following steps are involved: Step 1: Obtain the technical parameters, economic parameters, load forecast data and wind farm power forecast values ​​of each synchronous generator set in the power system, as well as the RoCoF protection setting value and the frequency extreme value allowable range in the power system; Step 2: Considering the dual uncertainties of inertia and wind power, select the probability density function of inertia prediction error and the probability distribution function of wind power prediction error, use Latin abstraction scenario generation and K-means scenario clustering to obtain multiple discrete scenarios of wind power and inertia and the corresponding scenario probabilities; Step 3: In the frequency response process, taking into account the inertia that is not known by the dispatching department, the dynamic frequency constraint of the system including the inertia error is derived according to the dynamic frequency stability range; Step 4: Based on the initial scenario probability distribution, the 1-norm and ∞-norm probability distribution uncertain confidence sets are constrained to build a two-stage distributed robust unit commitment optimization model with dynamic frequency constraints; Step 5: Auxiliary variables are introduced to linearize the absolute value constraints in the confidence set, and the column and constraint generation algorithm is used to divide the two-stage robust unit combination optimization model into the main problem and sub-problems for iterative solution.

2. According to claim 1, a two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power, characterized in that: In step 2, the inertia in the system that is not known by the dispatching department is calculated by using E R Inertia representation, consisting of two parts, one is based on the inertia point prediction Part of it is the inertia prediction error △E R , the expression is as follows: Gaussian distribution is used as the probability density function of inertia prediction error to characterize the uncertainty of inertia point prediction error; △E R ~N(0,σ 2 )(2); Where: σ is the inertia prediction error △E R The standard deviation of N(0,σ 2 ) represents the Gaussian distribution of the predicted value of the inertia point.

3. According to claim 2, a two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power, characterized in that: Using the wind power scenario generation method based on historical statistical data, the probability distribution function of wind power prediction error is assumed to be a normal distribution, as shown in formula (4); then sampling is performed to generate a series of wind power output prediction error values ​​that obey the given distribution, which are superimposed on the wind power prediction value to obtain a series of wind power output scenarios; random variables are used to characterize the uncertainty of wind power prediction error; the wind power output is set to The wind power forecast is The wind power prediction error is △P Wt , in MW, then: Normal distribution is used to describe the uncertainty of wind power forecast error; Where: Represents the normal distribution of wind power forecast error.

4. According to claim 3, a two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power, characterized in that: In step 2, the Latin hypercube sampling scenario is generated as follows: The probability distribution functions of wind power error and inertia error are shown in equations (2) and (4). Suppose there are M random variables x1, x2, x3, … x M ,x1,x2,x3,…x M is a random variable, that is, the dimension of the variable is M, the number of samples required for each variable is N, and the cumulative probability distribution function of each variable is F xi =f i (x i ), x i represents the i-th random variable, f i (x i ) represents the cumulative probability distribution function of each variable, where i = 1, 2, …, M; the Latin hypercube sampling method is used for sampling, and the process is as follows: A1: Define the sampling size as N, and take any random variable x i The cumulative probability distribution function curve F xi =f i (x i ) is divided into N intervals with equal probability, and the width of each interval is A2: For a random variable x i The kth sample value x ik , and its corresponding cumulative distribution probability is: Where: U~N(0,1) obeys uniform distribution; k=1,2,…,N; calculate the cumulative distribution function F xi The inverse function of x can be obtained i The kth sample value x ik , satisfying the following formula: Where: is the cumulative distribution function F xi The inverse function of A3: After sampling, the sampled values ​​of each random variable are arranged into a row of a matrix to form an M×N sampling matrix X MN .

5. According to claim 4, a two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power, characterized in that: In step 2, K-means scene clustering includes the following steps: B1: Input a data set D containing n data sample points = {x i |x i ∈R m ,i=1,2,…,n},x i is the data sample point; R m is a real number set of m-dimensional vectors, n is the number of data sample points; the number of clusters to be divided is k; B2: Randomly select k sample points y1,y2,…,y k There are k clusters C1, C2, …, C k The initial cluster center of Among them, y1,y2,…,y k are k sample points; C1, C2, …, C k Clusters for the dataset; B3: Calculate the Euclidean distance d(x i ,y j ), y j To randomly select k sample points, the data sample x i Divide into cluster C j So that x i and i Minimum distance; B4: Calculate the error square sum criterion function E after clustering; recalculate the new cluster center of each cluster. For cluster C j ,j=1,2,…,k, and use formula (9) to calculate the new cluster center y j ; In the above formula, E is the error square sum criterion function after clustering; C k Clusters for the dataset; B5: Recalculate the criterion function E′ according to the new cluster center. If |EE′| does not meet the specified threshold, repeat steps B3 and B4; if it does, output the clustering result.

6. The two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power according to claim 5, characterized in that: In step 3, the system equivalent inertia includes the synchronous unit inertia and E R Inertia can be expressed as: Where: E SYS is the total inertia of the system, in MW·s; E G is the generator inertia, in MW·s; E R is the collection of inertia forms such as virtual inertia load side inertia, and the unit is MW·s; S is the total number of generator sets i; S Ni is the rated capacity of generator i, in MW; H i is the inertia time constant of generator i, in s; U i The state quantity representing the start and stop of generator i, 0 means stop, and 1 means start; is the inertia prediction result, in MW·s; △E R is the prediction error, in MW·s; The constraint containing the maximum frequency change rate is: Where: |RoCoF set,max | is the maximum frequency change rate allowed by the system, in Hz / s; △P L is the active unbalance amount caused by the disturbance, in MW; f n is the system rated frequency; The constraint containing the frequency minimum point is: Where: f0 is the original operating frequency of the system, f set,min is the lowest frequency allowed by the system, in Hz; R SSi is the frequency modulation response rate of each synchronous generator, in MW / s; U it The start and stop status of the unit in the time period.

7. The two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power according to claim 5, characterized in that: In step 4, the distributed robust optimization method is used to limit the probability distribution of the scene. First, the prediction data and the prediction error probability density function are sampled using LHS to obtain N original scene numbers, and then the K-Means clustering algorithm is used to obtain N s A typical finite discrete scenario and this N s The initial probability distribution corresponding to a typical finite discrete scenario is the initial probability distribution of a typical finite discrete scenario, and the probability distribution fluctuation range is constrained based on the 1-norm and ∞-norm; Probability distribution is the probability distribution of a typical finite discrete scenario, and the confidence is expressed as: Where: Pr(·) is the probability function; θ1 and θ ∞ is the allowable deviation value of the probability distribution; Initial probability distribution of finite discrete scenarios; N s is the number of typical finite discrete scenes; N is the number of original scenes; 1-2N in formula (13) s exp(-2Nθ1 / N s ) is set to α1, and 1-2N s exp(-2Nθ ∞ ) This part is set to α ∞ , then α1 represents the confidence level satisfied by the probability distribution p based on the 1-norm, α ∞ Denotes the confidence level, θ1 and θ1 satisfied by the probability distribution p based on the ∞-norm respectively. ∞ It can be expressed by the following formula: The confidence set that can derive the probability distribution is as follows: Where: ψ is the probability distribution set interval of the scene, which represents the confidence set constrained by the 1-norm and ∞-norm; {p s }The set of probability distributions corresponding to typical finite discrete scenarios.

8. The two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power according to claim 7, characterized in that: In the two-stage distributed blue-rod unit combination optimization model, The first stage is the optimization of unit start-up and shutdown on the day before, with the minimum cost of unit start-up and shutdown as the objective function; the second stage is the optimization of unit power dispatch within the day, with the minimization of the sum of conventional unit operation cost and wind curtailment cost as the objective function; 1. First stage model: 1.1 Objective function of the first stage: The objective function of the first stage consists of the startup and shutdown costs of conventional units; Where: T is the number of time periods; N is the total number of conventional units; U it S is the start and stop status of the unit; it (τ it ) is the startup cost of unit i in period t; U i(t-1) is the start and stop status of conventional unit i in time period t-1; 1.2 Constraints of the first stage: (1) Minimum start-stop time constraints: Where: are the continuous start-up time and downtime of the i-th generator in the first t-1 periods respectively; T i on , T i down are the minimum continuous start and stop times of the i-th generator respectively; U i(t-1) is the start and stop status of conventional unit i in time period t-1; 2. Second stage model: 2.1 Second stage objective function: The objective function of the second stage is composed of the operating cost of conventional units and the wind turbine abandonment cost; Where: p s N is the occurrence probability of the sth discrete scene after scene clustering screening; s is the total number of discrete scenarios; T is the number of hours in the study period; N is the total number of conventional units; M is the total number of conventional units; U it P is the start-stop state of the conventional unit i; it,s is the output active power of conventional unit i in time period t in the sth discrete scenario; i , b i and c i is the unit operating parameter; w is the penalty parameter for wind abandonment; Predict the output value of wind turbines. For The wind power output corresponding to the scenario s generated under the wind power forecast output sampling clustering; P Wjt.s is the actual output value of the fan; 2.2 Second stage constraints: (1) System power balance constraints: Where: P Lt is the load forecast value of the system in period t; (2) Generator output constraints: U it Q i,min ≤P it,s ≤U it Q i,max (23); Where: P i.max , P i,min are the upper and lower limits of the output of the i-th generator respectively; (3) Generator climbing constraints: D i ≤P it -P i(t-1) ≤L i (24); Where: L i , D i are the upper and lower limits of the ramp rate of unit i respectively; (4) Fan output constraints: (5) Maximum frequency change rate constraint: Where: |RoCoF max | is the maximum rate of change of frequency, |RoCoF set,max | is the maximum frequency change rate allowed by the system, in Hz / s; △P L is the active unbalance amount caused by the disturbance, in MW; f n is the system rated frequency; N is the total number of generator sets i; S Ni is the rated capacity of generator i, in MW; H i is the inertia time constant of generator i, in s; U i is the state quantity that represents the start and stop of generator i (equal to 0 means stop, equal to 1 means start); is the predicted value of inertia, For The inertia output value corresponding to scenario s generated by inertia prediction output and prediction error sampling clustering, in MW·s; (6) Frequency minimum point constraint: Where: f min is the lowest point of the system frequency, f0 is the original operating frequency of the system, and f set,min is the lowest frequency allowed by the system, in Hz; R SSi is the frequency regulation response rate of each synchronous generator, in MW / s; 3 Two-stage distributed robust optimization model: The two-stage objective function and constraint conditions mentioned above are expressed in the form of a matrix, with x representing the variables of the first stage, which are the start and stop states of the conventional units; y representing the variables of the first stage, which are the start and stop states of the conventional units; s Represents the variables of the second stage. The variables of the second stage are unit power, fan power, etc. The data-driven two-stage distributed robust optimization model can be expressed as follows: st Cx≤c (29); Where: x is the first-stage variable; A is the cost coefficient corresponding to the first-stage variable x; p s is the occurrence probability of the sth discrete scene after scene clustering screening; ψ is the probability distribution set interval of the scene; y s is the second stage variable; Λ is the cost coefficient of the quadratic term in the objective function; B is the cost coefficient of the linear term in the objective function; ξ s It represents the wind power and inertia output value corresponding to the sth discrete scenario after scenario clustering screening, and K is the corresponding cost coefficient; Formula (29) represents the first stage constraint, corresponding to the conventional unit start-stop constraint formula (20); Formula (30) represents the equality constraints of the variables in the second stage, corresponding to formula (22); Equation (31) represents the inequality constraints involving the second-stage variables, corresponding to equations (23) and (24); Formula (32) represents the inequality constraints between the second-stage variables and the wind power in the discrete scenario, corresponding to formula (25); Formula (33) represents the inequality constraint between the variables in the first stage and the inertia in the discrete scenario, corresponding to formula (26); Equation (34) represents the inequality constraint between the first-stage variables and the inertia in the discrete scenario, corresponding to equation (27).

9. The two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power according to claim 7, characterized in that: In step 5, the wind power and inertia uncertainty set is as shown in formula (15). The 1-norm and ∞-norm constraints in formula (15) contain absolute value terms, which are nonlinear constraints. Auxiliary variables and auxiliary constraints are added to transform them into linear constraints. The added auxiliary variables and auxiliary constraints are as follows: Where: is the probability distribution p of scene s s Relative to The positive offset of is the probability distribution p of scene s s Relative to The negative offset of and They are the 1-norm constraints on p s A 0-1 sign that generates positive and negative offsets; and They are the ∞-norm constraints for p s A 0-1 sign that generates positive and negative offsets; Represents any scene; The transformed 1-norm and ∞-norm constraints are shown in equation (17) and equation (18) respectively; 10. The two-stage distributed robust unit commitment optimization method taking into account the dual uncertainties of inertia and wind power according to claim 9, characterized in that: In step 5, the specific process of solving the main problem and sub-problems of the two-stage Bruker unit combination optimization model by stage-by-stage iterative solution using the dualization-free C&CG algorithm is as follows: Step 1: Set the lower bound LB = -∞, the upper bound UB = -∞, and the number of iterations n = 1; Step 2: Solve MP and get the optimal solution (x * ,η * ), x * , η * Represents the optimal solution of the objective function of the main problem, and updates the lower bound value LB = A T x * +η * , A T Represents the main problem objective function matrix; Step 3: Fix the first stage variable x * , solve the subproblem and obtain the probability value in the worst case and the objective function value f sp (x * ); Update the upper bound min{UB,A T x * +f sp (x * )}; Step 4: If UB-LB < ε, ε represents an infinitesimal number, then stop the iteration and return the optimal solution x * ; Conversely, update the worst probability distribution of the main problem And add new variables to the main problem And add the main question with Related constraints; step5: Update n=n+1 and return to step2.

Images