Scheduling method for source-load-storage collaborative distribution robust optimization of wind power-containing electric power system

Abstract

The invention discloses a scheduling method containing wind power electric power system source load storage collaborative distribution robust optimization in the technical field of electric power systems. A multi-state transition matrix Markov chain model with high flexibility and a day-ahead two-stage distribution robust optimization model based on data driving are included. According to the method in the scheme, the multi-state transition matrix Markov chain model with high flexibility is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power systems, and in particular to a dispatching method for robust optimization of source-load-storage coordinated distribution in an electric power system containing wind power. Background Art

[0002] With the steady implementation of my country's "dual carbon" strategy, wind power, as a renewable clean energy, has become my country's third largest power source. However, due to its randomness and volatility, after large-scale wind power is connected to the power system, the peak-shaving problem of the power system has become increasingly prominent, seriously limiting the absorption capacity of wind power. With the continuous advancement of the market-oriented reform of the power industry, the large-scale implementation of the direct purchase of electricity by large users has brought new means to solve the peak-shaving problem of the power system.

[0003] At present, the research on direct power purchase by large users mainly focuses on the exploration of trading rules, cost sharing calculation, formulation standards of large user transaction electricity prices, optimal power purchase combination models and power purchase strategies for large users, dynamic environmental economic dispatch of direct power purchase, and negotiation methods for direct power purchase by large users. The above research only considers the operating benefits brought by direct power purchase by large users to the power system, but ignores the impact of direct power purchase by large users on the peak load regulation of the power system. In particular, under the premise that the current direct power purchase agreement is mainly bound to thermal power units, the randomness and volatility of wind power make the power system's day-ahead dispatch plan face huge pressure to absorb wind power.

[0004] How to deal with the uncertainty brought by wind power has become the key to the safe and stable operation of the system. For such problems, there are mainly two methods: stochastic optimization (SO) and robust optimization (RO). The literature "Electricity-natural gas operation planning with hourly demand response for deployment of flexible ramp" uses stochastic programming to solve the multi-type energy scheduling problem, but the stochastic programming method relies on the estimation of the probability distribution of the problem, and the real distribution is often difficult to obtain accurately. The robust optimization method in the literature "Two-stage robust optimization economic dispatch method for microgrids" is used to deal with the uncertainty of wind power and analyze the impact of different confidence levels and direct purchase electricity prices on system economic dispatch and wind power consumption. However, the robust optimization method over-considers the worst case, which limits the performance of the model under normal or more favorable conditions. Therefore, in recent years, some scholars have proposed a distributionally robust optimization method (DRO) that combines the advantages of both to solve the shortcomings of SO and RO. DRO uses fuzzy sets composed of probability distributions to describe the characteristics of uncertain variables and find the worst probability distribution decision plan under the confidence set. By sacrificing a certain degree of robustness in exchange for greater economic benefits, the advantages of SO and RO are taken into account. The document "Distributed Robust Optimal Dispatching of Microgrids Considering Risk and Carbon Trading Mechanism" uses Wasserstein distance to construct probability distribution fuzzy sets, and considers extreme scenario correction to reduce the number of scenarios and improve the solution speed of the model. The document "Low-carbon Distributed Robust Optimal Dispatching of AC and DC Distribution Networks Considering Wind Power Uncertainty" constructs distributed robust fuzzy sets based on Kullback-Leibler divergence, and transforms the model into a single-layer linear optimization objective through duality theory for solution. The document "Distributed Robust Optimal Dispatching Method for Power System Considering Hydropower, Photovoltaic and Storage Complementarity and DC Transmission" proposes a two-stage distributed robust optimization scheduling scheme based on moment information. However, the above method relies too much on empirical probability distribution, and the model parameters are difficult to define and the solution is complex. The data-driven distributed robust optimization algorithm realizes accurate modeling of actual situations by mining the statistical characteristics of historical data. It also finds the decision-making plan under the worst probability distribution in the uncertainty probability distribution set under norm constraints, and provides a dispatch plan with robustness and economy.

[0005] In order to reduce the impact of the randomness and volatility of wind power output, it is necessary to accurately predict wind power. The document "Wind power output sequence scenario generation method based on month division and specified day type" combines cluster analysis, daily characteristic indicators and Markov chain to make the random characteristics of the generated scenario closer to the statistical laws contained in historical data. The document "Economic dispatch of power system considering wind power uncertainty and direct power purchase by large users" superimposes the wind speed prediction error sequence generated by the ARMA model and Monte Carlo method with the initial predicted wind speed to obtain the corresponding wind power output. However, neither of the above two methods considers the temporal and interval characteristics of wind power uncertainty at the same time. Summary of the invention

[0006] The present invention aims to provide a dispatching method for coordinated distribution and robust optimization of source, load and storage in a wind power system. By constructing a first-order Markov chain state transfer matrix with interval characteristics, the traditional Markov chain model is improved to have a more flexible state transfer matrix, reflecting the interval characteristics of wind power output.

[0007] The dispatching method for coordinated distributed robust optimization of source, load and storage in the wind power power system in this scheme includes a multi-state transfer matrix Markov chain model with high flexibility and a data-driven two-stage day-ahead distributed robust optimization model.

[0008] The working principle of this scheme is as follows: by mining the distribution characteristics of historical data, the elbow method is used to divide the historical data into M categories, and the fitting function of the corresponding prediction interval and the state transfer matrix P between adjacent moments are calculated, thus realizing a highly flexible multi-state transfer matrix Markov chain model.

[0009] Further, the multi-state transfer matrix Markov chain model with high flexibility comprises the following steps:

[0010] 1) Classify the predicted data by the elbow method;

[0011] 2) Calculate the nonparametric kernel density estimation function for different power intervals;

[0012] 3) Calculate the prediction error state transfer matrix at each moment;

[0013] 4) Obtain the upper and lower fluctuation domain of the prediction error at each moment from the state transfer matrix;

[0014] 5) Extract the prediction error value at the next moment according to the interval where the power prediction value is located and the prediction error state at the previous moment;

[0015] 6) If no upper or lower fluctuation domain is generated, continue to step 5). If an upper or lower fluctuation domain is generated, superimpose the wind power forecast;

[0016] 7) When t≠T, continue with step 5); when t=T, construct the day-ahead scenario set;

[0017] 8) kmeans clustering to obtain typical scenarios.

[0018] Furthermore, based on the data-driven two-stage day-ahead distributed blue-robust optimization model: the overall solution process of the two-stage distributed blue-robust optimization model for large-scale direct power purchase by users taking into account the uncertainty of wind power is as follows:

[0019] 1) Let the lower bound of the operating profit of the two-stage optimization model be LB = -∞, the upper bound of the operating profit be UB = +∞, the initial number of iterations be M = 1, and the initial scenario probability distribution be P 0 ;

[0020] 2) Solve the main problem and obtain the optimal solution for the current initial scenario At this time, LB=max{LB,a T x * +b T y0 * +c T ξ0+η *};

[0021] 3) The x obtained in the main problem * Pass to the subproblem. Find the known x * The worst scenario probability distribution P under the condition k * And the optimal objective function value η(x * ), at this time, UB=min{UB,a T x * +b T y0 * +c T ξ0+η(x * )}.

[0022] 4) Given a convergence threshold ε, if UB-LB ≤ ε, stop the iteration and return the optimal solution x * Otherwise, update the probability distribution of the bad scene in the main problem Repeat the above steps.

[0023] Working principle and beneficial effects of this scheme: In summary, this paper uses the improved Markov chain model to consider wind power output and establishes a data-driven two-stage day-ahead distributed robust optimization model. Under the two-stage robust optimization, the first stage of the model integrates wind power forecast scenario information, takes the total revenue of wind farm stations as the goal, and formulates a day-ahead robust scheduling plan for the first stage. In the second stage of the day-ahead, the 1-norm and ∞-norm are used to jointly constrain the uncertainty probability distribution confidence set, and the output of controllable units in the region is adjusted to flexibly respond to the uncertainty of wind power output. When solving, the distributed robust optimization method is used to process the random variables in the model, and the column and constraint generation algorithm are used to alternately iterate. Finally, the effectiveness of the distributed robust optimization algorithm is verified by examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a scatter plot of prediction error distribution;

[0025] Figure 2 is the correlation coefficient matrix;

[0026] Figure 3 Generate a flow chart for the day-ahead scenario;

[0027] Figure 4 is the optimal classification number of the elbow method;

[0028] Figure 5 It is the forecast error distribution fitting diagram;

[0029] Figure 6 For wind power scenario set comparison;

[0030] Figure 7 The direct purchase electricity price is one;

[0031] Figure 8 The direct purchase electricity price is 2;

[0032] Fig. 9 The direct purchase electricity price is three;

[0033] Fig.10 The operating conditions of each unit under different direct purchase electricity amounts. DETAILED DESCRIPTION

[0034] The following is a further detailed description through specific implementation methods:

[0035] The embodiment is basically as shown in the attached Figure 1 As shown:

[0036] 1. Wind power output generation scenario model

[0037] Due to the randomness and volatility of wind power, wind power forecasting is often accompanied by certain errors. Therefore, the wind power output forecast can be regarded as introducing a forecast error term on the basis of deterministic forecasting, and the forecast error term contains the uncertainty characteristics of wind power. In order to more accurately describe the wind power output, this embodiment will deeply analyze the characteristics of wind power forecasting errors.

[0038] 1.1 Analysis of wind power forecast error interval characteristics

[0039] When predicting wind power, the predicted value is often distributed around the actual value. In order to further study the relationship between the predicted value and the prediction error, a scatter plot is made based on the historical data of a wind farm for one year, such as Figure 1 shown.

[0040] Depend on Figure 1 It can be seen that the prediction error is distributed around the predicted value, and different predicted values ​​show completely different error distribution patterns. The overall distribution shows the characteristics of small at both ends and large in the middle. That is, in the low output range, the prediction error is small and mostly positive. In the medium output range, the prediction error is large and asymmetric. In the high output range, the prediction error is small and the distribution is discrete. Therefore, different wind power output prediction values ​​have different distribution error characteristics, and the wind power prediction error is interval-based.

[0041] 1.2 Analysis of the time series characteristics of wind power forecasting errors

[0042] Wind power output has time series correlation, that is, the wind power output at adjacent times has a high correlation. In order to verify the time series correlation of wind power prediction error, the wind power prediction error value X(x1, x2,…, x q )、Y(y1,y2,…,y q ) and construct the correlation coefficient matrix of wind power prediction error, as shown in Figure 2 shown.

[0043]

[0044] In the formula, X and Y are variables, μ x , μ y It is the average value of variables X and Y. The γ value ranges from -1 to 1. The larger the absolute value, the stronger the correlation.

[0045] In summary, when constructing a set of wind power output scenarios, it is necessary to fully consider the interval and time series characteristics of wind power prediction errors, that is, different wind power prediction values ​​have different prediction error distributions, and the prediction errors between adjacent moments have a strong correlation.

[0046] 1.3 Wind power forecast error probability distribution fitting

[0047] In distribution fitting, there are mainly two methods: parameter estimation and non-parametric estimation. Among them, parameter estimation requires estimating specific parameters of the overall distribution from the sample to perform statistical inference, but accurate distribution is often difficult to obtain, resulting in large errors. Non-parametric estimation does not require assumptions about data distribution, but estimates the probability density function of the data based on observed data through kernel density estimation or histogram. Therefore, this paper uses non-parametric kernel density estimation to model and describe wind power uncertainty.

[0048] Among them, the probability density function f of the nonparametric kernel density estimate h (·) and Gaussian kernel function K(·) are:

[0049]

[0050] In the formula, Q WHB,t is the heat energy recovered by the waste heat boiler at time t; η WHB is the heat recovery efficiency of the waste heat boiler. Where, E is the prediction error; E q is the prediction error at the qth moment; h is the bandwidth; n is the number of samples.

[0051] 1.4 Markov chain model

[0052] Markov chain is a mathematical model that describes random processes and can be defined by state space, initial probability distribution and state transition matrix. The state space represents all possible states of the system, the initial probability distribution represents the probability distribution of each state of the system at the beginning, and the state transition probability matrix P describes the probability of the system transferring from one state to another. P is an N×N matrix, where N is the number of prediction error state intervals and the elements P in P are ij is the current state x i ω i , the next moment state x i+1 Transfer to ω j Probability:

[0053] P ij =P(x i+1 =ω j |x i =ω i ) (4)

[0054] 1.5 Improved Markov chain model

[0055] In order to more accurately describe the temporal and interval characteristics of wind power output, this paper proposes a first-order Markov chain state transfer matrix partitioning method with interval characteristics based on the traditional Markov chain model. By mining the distribution characteristics of historical data, the elbow method is used to divide the historical data into M categories, and the fitting function of the corresponding prediction interval and the state transfer matrix P between adjacent moments are calculated, realizing a highly flexible multi-state transfer matrix Markov chain model.

[0056] This innovation enables the model to perform well in system modeling under changing conditions, providing a more powerful tool for dealing with diverse practical scenarios. It achieves more accurate modeling of wind power output and provides strong support for improving the accuracy of the prediction model. The process flow chart is as follows Figure 3 As shown:

[0057] 2. Data-driven two-stage day-ahead distributed robust optimization model

[0058] 2.1 Objective Function

[0059] The economic dispatch of power systems aims to minimize operating costs and optimize the use of power resources while ensuring the safe and stable operation of the power system and meeting user needs. This paper comprehensively considers factors such as operating costs F1 and electricity sales revenue F2, and uses the distributed robust method to construct an optimization model to analyze the impact of different direct purchase electricity prices on the access of large-scale wind power to the power system. In the first stage of the model, the set of wind power output forecast scenarios and the regulating role of energy storage are fully considered to formulate a robust dispatch plan for the first stage of the day-ahead. Based on the first stage dispatch plan, the second stage of the day-ahead adjusts the output of controllable units in the region to flexibly respond to the uncertainty of wind power output. In addition, the wind abandonment penalty term F is introduced in the objective function. 16 , minimize wind abandonment and improve the utilization rate of wind power resources.

[0060] In summary, the objective function is shown in formula (5):

[0061]

[0062] F1=F 11 +F 12 +F 13 +F 14 +F 15 +F 16 (6)

[0063] Among them, the start-up and shutdown cost F 11 、Fuel cost 12 、Environmental cost 13、 F 14 、Operation cost of energy storage F 15 , Wind abandonment penalty F 16And the electricity sales revenue F2 is:

[0064]

[0065]

[0066] Where: T is the total number of time periods; N G is the number of thermal power units; N W is the number of wind farms; S Ti and S Di are the startup and shutdown costs of thermal power unit i respectively; y i,t and Z i,t is the state parameter of the thermal power unit, y i,t =1 means power on, Z i,t =1 means shutdown, U i,t is the state of thermal power unit i. When it is 1, it indicates that unit i is outputting, and when it is 0, it indicates that unit i is stationary; a i , b i , c i is the power generation cost coefficient of thermal power unit i; P i,t is the active power output of thermal power unit i; γ en,c and γ en,s is the environmental cost coefficient of CO2 and SO2 released by combustion; α c,i , β c,i , c,i is the CO2 emission coefficient of thermal power unit i; α s,i , β s,i , s,i is the SO2 emission coefficient in thermal power unit i; and They are the electricity price sold to the grid and the price of direct purchase of electricity; and They are the non-directly purchased electricity and directly purchased electricity of thermal power units respectively; and are the charging and discharging power of energy storage respectively; η is the charging and discharging efficiency of energy storage; K s is the converted unit charging and discharging cost; δ is the wind abandonment penalty coefficient; and are the predicted output value and the actual dispatched output of wind farm i, respectively; P k is the probability of the kth scene, d i,t is the output adjustment cost of thermal power units; ΔP i,t,k is the output adjustment of thermal power unit; c wind Adjust costs for wind power output.

[0067] 2.2 Constraints

[0068] The distributed robust fuzzy set constructed based on the probability density function does not need to calculate the deviation between the assumed probability distribution and the empirical distribution in advance, which reduces the difficulty of fuzzy set construction. At the same time, this method has more random variable probability distribution information, reduces the requirements for historical data, and improves model accuracy and computational efficiency. Therefore, this paper adopts the idea of ​​non-precise probability estimation to construct the IDM model and establishes a distributed robust CDF fuzzy set based on IDM.

[0069] (1) Power balance constraints;

[0070]

[0071] Where: P I,f,t is the total load of the system at time t.

[0072] (2) Output constraints of thermal power units;

[0073] P i min U i,t ≤P i,t ≤P i max U i,t (15)

[0074] Where: P i max and P i min are the maximum and minimum output values ​​of thermal power unit i.

[0075] (3) Unit climbing constraints;

[0076]

[0077] Where: and is the ramp rate of unit i. ΔT is the ramp time.

[0078] (4) Minimum start and stop time constraints for units;

[0079]

[0080] Where: T i on and T i off is the continuous start-stop time of unit i; T i U and T i D is the minimum start and stop time of unit i.

[0081] (5) Backup constraints:

[0082]

[0083] Where: R is the reserve rate.

[0084]

[0085] Where: E i The daily direct purchase electricity amount signed for unit i.

[0086] (7) Real-time wind power output constraints;

[0087]

[0088] (8) Energy storage constraints

[0089]

[0090] Where: P s max U is the maximum charge and discharge power allowed by energy storage; s,t It is the charge and discharge status, 1 means discharging, 0 means charging; E s (0) is the energy storage capacity at the initial moment; and The upper and lower limits of the remaining capacity of energy storage during the dispatch process;

[0091] 3. Norm constraints

[0092] The two-stage distributed robust optimization model proposed in this paper aims to solve the problem of wind power output uncertainty in the context of direct power purchase by large users. It is necessary to obtain the probability distribution that maximizes the expected objective function, but the actual probability distribution is difficult to obtain. In order to make the distribution obtained through the scenario set more consistent with the actual operating data while avoiding excessive deviation from the actual probability distribution. Based on the initial probability distribution of each discrete scenario, this paper restricts the probability distribution of wind power scenarios through the 1-norm and ∞-norm.

[0093]

[0094] Where: is the initial probability value of the kth scene; θ1 and θ ∞ are the maximum allowable deviation values ​​of 1-norm and ∞-norm respectively.

[0095] According to the literature "Zhao CY, Guan Y P. Data-driven stochastic unit commitment for integrating wind generation [J]. IEEE Transactions on Power Systems: APublication of the Power Engineering Society, 2016, 31(4): 2587-2596.", "Ding T, Yang Q, Yang Y, et al. A data-driven stochastic reactive power", {P k The confidence of} is shown in the following formula (25):

[0096]

[0097] Let the right side of the above inequality be the confidence level α1 and α of the probability distribution value respectively. ∞ , then formula (25) can be transformed into formula (26):

[0098]

[0099] From equation (25) and equation (26), equation (27) can be obtained as:

[0100]

[0101] Equations (1) to (28) are the two-stage distributed robust optimization model proposed in this paper.

[0102] 4. DRO model solution

[0103] For the two-stage model of the day before (5), this paper adopts the CCG algorithm to solve it. First, the two-stage model proposed in this paper is decomposed into the main problem and sub-problems, and then the CCG algorithm is used to repeatedly iterate the main problem and sub-problems to obtain the optimal solution of the original problem, where the main problem is as follows:

[0104]

[0105] Where: x is the first-stage robust decision variable; y k is the second-stage variable under the kth scenario; ξ0 is the predicted value of wind power output; ξ k is the wind power output value in the kth scenario; m is the number of iterations.

[0106] The sub-problems are of the form:

[0107]

[0108] Since the norm constraint contained in equation (25) is an absolute value constraint, this paper introduces two 0-1 auxiliary variables to make it equivalent to a linear constraint. The absolute value term in equation (27) is used to represent the upper and lower limits of the comprehensive norm constraint and cannot be directly used to solve the linear objective function (5) because the probability P k Relative to The probability offset is unique, by introducing a 0-1 auxiliary variable and The probability P k Relative to The positive and negative offsets are distinguished, so the absolute value term processing process of the 1-norm is:

[0109]

[0110] In addition, the following constraints must be met:

[0111]

[0112]

[0113] The absolute value term processing process of the ∞-norm is the same as that of the 1-norm. The positive offset and negative offset are distinguished by introducing constraints (32) and (33). However, unlike the 1-norm (34), the following constraints need to be added:

[0114]

[0115] Therefore, model (31) is transformed into a mixed linear programming problem that can be quickly solved to obtain the worst probability distribution P * , and P * Passed to the main problem for iterative solution.

[0116] In summary, the overall solution process of the two-stage distributed blue-robust optimization model for large-scale direct power purchase by users considering the uncertainty of wind power proposed in this paper is as follows:

[0117] Step 1: Let the lower bound of the operating profit of the two-stage optimization model LB = -∞, the upper bound of the operating profit UB = +∞, the initial number of iterations M = 1, and the initial scenario probability distribution P 0 ;

[0118] Step 2: Solve the main problem and obtain the optimal solution for the current initial scenario At this time, LB=max{LB,a T x * +b T y0 * +c T ξ0+η *};

[0119] Step 3: Substitute the x obtained in the main problem * Pass to the subproblem. Find the known x * The worst scenario probability distribution P under the condition k * And the optimal objective function value η(x * ), at this time, UB=min{UB,a T x * +b T y0 * +c T ξ0+η(x * )}.

[0120] Step 4: Given a convergence threshold ε, if UB-LB ≤ ε, stop the iteration and return the optimal solution x * Otherwise, update the probability distribution of the bad scene in the main problem Repeat the above steps.

[0121] 5. Example Analysis

[0122] 5.1 Setting of original data and basic parameters of unit combination

[0123] In order to verify the dispatch model proposed in this paper, taking the thermal power direct purchase model as an example, the 6-unit power system provided in the document "Liu Yixin, Guo Li, Wang Chengshan. Two-stage robust optimization economic dispatch method for microgrids [J]. Proceedings of the CSEE, 2018, 38(14): 4013-4022, 4307." is expanded to a 12-unit system. At the same time, a wind power with an installed capacity of 400MW is introduced to study the impact of large-scale direct purchase of electricity on the system's wind power absorption capacity. Among them, the first type of unit signed a direct purchase of 1600MW, and the second type of unit signed a direct purchase of 400MW. The output of the first and second types of units are added together when statistics are taken. The peak-shaving capacity used in this paper is the difference between the equivalent load of the system and the minimum total startup capacity of the system. That is, the expected peak-shaving capacity is E[tf]:

[0124]

[0125] In order to calculate the electricity sales revenue on the farm side, this paper introduces a time-of-use electricity price strategy, dividing the 24 hours of a day into three time periods. At the same time, in order to mobilize the enthusiasm of power generation companies and large users, the direct purchase electricity price should be lower than the sales electricity price and higher than the grid-connected electricity price, so that the benefits of power generation companies increase and the electricity purchase costs of large users decrease. For specific time-of-use electricity price strategies and electricity price formulation, see the document "Wan Qingzhu, Liu Weina. Robust interval optimization scheduling of direct purchase electricity for large users considering wind load uncertainty [J]. Acta Energiae Solaris Sinica, 2022, 43(7): 347-355.".

[0126] 5.2 Wind power scenario generation

[0127] Since the wind power forecast output is interval-based, historical data needs to be clustered. In order to reduce the human impact, the elbow method is used to determine the optimal number of classifications, such as Figure 4 As shown:

[0128] Here, we select the cluster number k=3 to perform k-means clustering on the historical data. This is a clustering algorithm that divides groups based on distance. For a given sample set R, the samples are divided into I groups so that the distance between points in a group is minimized and the distance between groups is maximized.

[0129]

[0130] Where p is the sample point in the group, μ is the centroid of the group, V is the number of features of the sample point, and v is each feature of the constituent point p.

[0131] The prediction error corresponding to the clustered predicted output is fitted by non-parametric kernel density estimation. The fitting results are as follows: Figure 5 shown.

[0132] from Figure 5 It can be seen that the prediction errors after k-means clustering have different distribution characteristics, and their characteristics are consistent with the interval analysis. Therefore, when fitting the probability distribution of the prediction error, the distribution impact brought by different prediction values ​​should be considered.

[0133] In order to verify the accuracy of the improved Markov chain model proposed in this paper, two days are randomly selected to compare with the traditional Markov chain generation scene, such as Figure 6 shown.

[0134] Depend on Figure 6 It can be seen that the improved Markov chain model can effectively characterize the uncertainty of wind power output. And it has a smaller fluctuation range, which is particularly obvious in the low output range, which shows that different predicted values ​​correspond to different fluctuation domains and there is no obvious disordered fluctuation. In summary, the improved Markov chain model proposed in this paper can effectively reflect the interval and time series characteristics of wind power output.

[0135] 5.3 Comparison of Robust Optimization Results under Different Confidence Intervals

[0136] In order to verify the effectiveness of the two-stage day-ahead distributed robust optimization model under the comprehensive norm constraint proposed in this paper, taking the direct purchase electricity price three as an example, by setting different confidence intervals α1 and α ∞ The calculation results of the large-scale direct power purchase system taking into account the uncertainty of wind power are analyzed. The system revenue results under different confidence intervals are compared as shown in Table 1:

[0137] Table 1 Comparison of system benefits at different confidence levels

[0138]

[0139] From Table 2, we can see that when α1 is fixed, as α ∞ As the value of α increases, the system benefits continue to decrease. ∞ When the value is fixed, as the value of α1 increases, the system benefit continues to decrease. At the same time, it can be seen from formula (27) that when the confidence α1 and α ∞ When the value increases, the confidence interval of the system increases, the model faces greater uncertainty in wind power output, and obtains decision-making plans under more severe conditions, resulting in increased costs and reduced benefits. The conclusions obtained from Table 2 and formula (27) show that the proposed model is correct and can be used for the next step of research.

[0140] Furthermore, in order to verify the advantage of the comprehensive norm, the calculation results of the comprehensive norm are compared with those of the separate 1-norm and ∞-norm.

[0141] First, calculate the single ∞-norm α ∞ = 0.99, and then compared with the ∞-norm α ∞ =0.99 is fixed and the calculation results when the 1-norm changes are compared, as shown in Table 2.

[0142] Table 2 Comparison between comprehensive norm and single ∞-norm

[0143]

[0144] Similarly, the total system benefit result of a single 1-norm α1=0.5 is calculated first, and then compared with the calculation results when the 1-norm α1=0.5 is fixed and the ∞-norm changes, as shown in Table 3.

[0145] Table 3 Comparison between comprehensive norm and single 1-norm

[0146]

[0147] It can be seen from Tables 2 and 3 that the benefits obtained by using the comprehensive norm are higher than the total benefits when only the 1-norm or ∞-norm is used. This shows that the conservativeness of the conditional constraints of the comprehensive norm is lower than that when only a single norm constraint is used, and it has better economy.

[0148] 5.4 Impact of direct power purchase on wind power consumption in the system

[0149] The peak load capacity strictly limits the system's wind power absorption capacity. In order to further explore the principle of large-scale direct power purchase for improving wind power absorption capacity, this paper takes the two-stage distributed blue bar optimization model of large-scale direct power purchase with wind power uncertainty as an example to analyze the impact of direct power purchase price on peak load capacity. The distribution of direct power purchase, total expected benefits and system peak load capacity under different direct power purchase prices are as follows: Figure 7 And as shown in Table 4.

[0150] Table 4 Comparison of SO-RO-DRO results

[0151]

[0152] Depend on Figure 7-9 It can be seen that in order to obtain greater economic benefits, power generation companies always tend to use direct purchase of electricity when the price difference is large. Furthermore, it can be seen from Table 4 that when direct purchase of electricity is adopted, the total revenue and peak-shaving capacity of the system are always greater than the total revenue and peak-shaving capacity when there is no direct purchase of electricity. This is because when direct purchase of electricity is used, the units that have signed direct purchase agreements will increase their power generation, thereby squeezing out the output of other units and optimizing the output combination of the units, which will lead to an increase in the total revenue and peak-shaving capacity of the system. Moreover, by comparing direct purchase electricity prices one, two, and three, it can be seen that when different direct purchase electricity prices are adopted, the system has different peak-shaving capacities. By formulating a reasonable direct purchase electricity price, the unit combination can be made more reasonable, with a larger peak-shaving capacity, and the system can better cope with the uncertainty brought by wind power.

[0153] In order to further analyze the principle of how direct power purchase transactions affect the system peak load capacity, this paper compares the unit startup combination when there is no direct power purchase with the unit startup method when the direct power purchase contract is 2000MWh. Fig.10 shown.

[0154] Fig.10 The startup status of the system units in these two cases is shown. In this example, the lowest value of the system equivalent load occurs at t = 3. In the absence of direct power purchase, the units started at t = 3 include units 1, 2, 7, 8, and 12, and the system peak-shaving capacity is 173.34MW. However, when the electricity price three is adopted, the output of units 1 and 2 at this moment increased from 258.38MW when there is no direct power purchase to 278.92MW, which has a certain squeeze on the output of units 8 and 12, resulting in a delay in their startup time. Therefore, the number of units started at t = 3 is reduced, and the system peak-shaving capacity is increased.

[0155] Furthermore, from the calculation results in Table 4, it can be seen that the total revenue and peak-shaving capacity of the distributed robust optimization algorithm are between the random programming and robust methods, indicating that the robustness of the distributed robust algorithm adopted in this paper is higher than that of the random programming, and has higher economy than the robust optimization. Combining the economy of random programming and the robustness of robust optimization, the distributed robust algorithm proposed in this paper has achieved a balance between economy and robustness.

[0156] 6. Conclusion

[0157] Based on the time series and interval characteristics of wind power forecast errors, this paper proposes an improved Markov chain model to generate a set of wind farm output scenarios that consider the time series and interval characteristics of wind power. The uncertainty of wind power output is handled by a two-stage robust optimization dispatch model based on the day-ahead. This model is compared with traditional stochastic optimization methods and robust optimization methods to analyze the impact of different direct purchase electricity prices on the system's wind power absorption capacity when large users directly purchase electricity to participate in the power system. The conclusions are as follows:

[0158] 1) An improved Markov chain model considering the interval nature of wind power prediction error is proposed, which makes the wind power scenario more accurate and reduces the uncertainty brought about by wind power access to the system.

[0159] 2) The direct purchase electricity price can change the power generation behavior of power generation companies and formulate a reasonable direct purchase electricity price system, so that power generation companies and users can obtain greater benefits, reduce the minimum total number of system startups, increase the system peak load capacity, and facilitate the absorption of wind power.

[0160] 3) Since the distributed rod optimization takes certain uncertainties into consideration in advance and reserves a certain peak-shaving capacity to cope with the uncertainty of wind power output, it has a better performance in balancing economy and conservatism.

[0161] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A dispatching method for coordinated distribution and robust optimization of source, load and storage in a wind power system, characterized by: It includes a multi-state transfer matrix Markov chain model with high flexibility and a data-driven two-stage day-ahead distributed robust optimization model.

2. The dispatching method for coordinated distribution and robust optimization of wind power system source-load-storage according to claim 1 is characterized by: The multi-state transfer matrix Markov chain model with high flexibility comprises the following steps: 1) Classify the predicted data by the elbow method; 2) Calculate the nonparametric kernel density estimation function for different power intervals; 3) Calculate the prediction error state transfer matrix at each moment; 4) Obtain the upper and lower fluctuation domain of the prediction error at each moment from the state transfer matrix; 5) Extract the prediction error value at the next moment according to the interval of the power prediction value and the prediction error state at the previous moment; 6) If no upper or lower fluctuation domain is generated, continue to step 5). If an upper or lower fluctuation domain is generated, superimpose the wind power forecast; 7) When t≠T, continue with step 5); when t=T, construct the day-ahead scenario set; 8) kmeans clustering to obtain typical scenarios.

3. The dispatching method for coordinated distribution and robust optimization of wind power system source-load-storage according to claim 2 is characterized by: The overall solution process of the two-stage distributed blue-robust optimization model based on data-driven day-ahead: Taking into account the uncertainty of wind power, the two-stage distributed blue-robust optimization model for large-scale direct power purchase is as follows: 1) Let the lower bound of the operating profit of the two-stage optimization model be LB = -∞, the upper bound of the operating profit be UB = +∞, the initial number of iterations be M = 1, and the initial scenario probability distribution be P 0 ; 2) Solve the main problem and obtain the optimal solution for the current initial scenario At this time, LB=max{LB,a T x * +b T y0 * +c T ξ0+η * }; 3) The x obtained in the main problem * Pass to subproblem; find the known x * The worst scenario probability distribution under the conditions And the optimal objective function value η(x * ), at this time, UB=min{UB,a T x * +b T y0 * +c T ξ0+η(x * )}; 4) Given a convergence threshold ε, if UB-LB ≤ ε, stop the iteration and return the optimal solution x * ; Otherwise, update the probability distribution of the bad scenarios in the main problem Repeat the above steps.

Images