Micro-grid double-layer robust optimization scheduling method based on data driving

Abstract

The invention discloses a micro-grid double-layer robust optimization scheduling method based on data driving, and the method employs a K-means clustering method to construct a typical scene which represents a large number of complex scenes, then employs a data driving method to construct a wind power uncertainty set, describes the wind power output distribution, effectively eliminates unnecessary extreme scenes, and improves the scheduling efficiency. And the model conservative degree is reduced. And a two-stage adaptive robust optimization model comprising a micro-grid day-ahead pre-scheduling model and a real-time scheduling model is established, and a column constraint generation algorithm is utilized to decouple and decompose the model into a main problem and a sub-problem for interactive iterative solution, so that the problems of inaccurate modeling and strong robust optimization conservative property caused by uncertain factors of random optimization are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power data processing, and more specifically, to a data-driven double-layer robust optimization scheduling method for a microgrid. Background Art

[0002] With the rapid increase of global population and the rapid development of industrialization, traditional energy sources such as coal and oil are being consumed rapidly, and environmental problems such as global warming and air pollution are becoming increasingly serious. In order to solve the prominent contradiction between traditional energy consumption and environmental pollution, the electric energy field has turned its attention to renewable energy power generation technology with advantages such as low pollution, renewability and high flexibility. With its advantages such as high power supply reliability and flexible operation mode, microgrid has become the main mode for local consumption and grid-connected transmission of renewable energy, and has received widespread attention and research.

[0003] In order to achieve multi-objective coordinated optimization scheduling and safe and stable operation of microgrids, there are several solutions. At present, there are two main methods for optimizing scheduling of uncertain factors in power systems: random optimization and robust optimization. The random scheduling optimization method needs to assume the probability distribution model of random variables, but the probability distribution model cannot accurately describe the complex change law of actual uncertainty factors; the robust scheduling optimization method uses an uncertain set to characterize the change of uncertainty factors, and does not need to assume a probability distribution model, but when considering the optimal solution under the worst scenario, it may lead to conservative robust optimization scheduling results. Therefore, how to improve the accuracy of probability distribution and avoid robust conservatism is the key to improving the stable operation of microgrids.

[0004] In CN116957229A, a two-stage distributed robust optimization scheduling method for microgrids based on Hausdorff distance is disclosed. According to the historical data of renewable energy and load, M groups of wind power historical scenes and M groups of load historical scenes are extracted respectively; the K-means clustering method is used to achieve scene reduction, and each cluster center is a typical scene; the obtained wind power and load typical scenes are arranged and combined to obtain a combined typical scene; a probability distribution uncertainty set with the initial probability distribution as the center and the Hausdorff distance as the constraint condition is constructed; a two-stage distributed robust optimization scheduling model for microgrids based on Hausdorff distance is constructed; and the established optimization scheduling model is iteratively solved. Although this method is distributed robust and fully considers the uncertainty of scene probability, and takes into account the conservatism of robust optimization to achieve good results, it still has certain limitations. In the actual scheduling method, the uncertainty of random variables will cause the accuracy and robustness of modeling to deteriorate, affecting the operating cost of the microgrid, such as the uncertainty of renewable energy generation and load, the pre-scheduling scheme has poor robustness, and the ability to resist the uncertainty of wind power changes is weak, and the regulation cost and wind abandonment cost will increase accordingly. Summary of the invention

[0005] The main technical problem to be solved by the present invention is to provide a data-driven microgrid double-layer robust optimization scheduling method in view of the shortcomings of inaccurate modeling of uncertain factors in random optimization and strong conservatism in robust optimization in the prior art.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A data-driven two-layer robust optimization scheduling method for microgrids, comprising the following steps:

[0008] S1. Construct theoretical framework;

[0009] S11. Establish a microgrid optimization dispatching architecture with model pre-training and rolling training, and construct a two-stage dispatching optimization model, which is expressed as:

[0010]

[0011] In the formula, x is the first-stage decision variable, which represents the unit start-up and shutdown plan; g(x) is the objective function of the first-stage optimization problem; ζ is a random variable, which represents the uncertainty of renewable energy generation and load; y is the second-stage decision variable; f(y) is the objective function of the second-stage optimization problem; Ω(x,ζ) is the feasible domain of y; U is the uncertainty set of the random variable ζ;

[0012] S12. Determine the dispatch target with the optimal operating cost in the pre-dispatching stage of the microgrid. The objective function expression of the pre-dispatching stage is:

[0013]

[0014] Where: G is the total cost; T is the total number of time periods, and T=24 in the pre-scheduling stage; is the operating cost of conventional units, gas turbines and fans in the τ time period; is the start-up and shutdown cost of conventional generator set i in time period τ; is the charging and discharging cost of the energy storage device in the τ time period; is the transaction cost between the microgrid and the main grid in the τ period; For reward and punishment expenses;

[0015] S13. According to the change of benefits when regulating the energy storage equipment, the real-time regulation cost and the optimization target of power generation output, the objective function expression of the real-time regulation stage is determined as:

[0016] minF=ΔC G,up +ΔC G,down +ΔC ES +ΔC M +C loss +maxCr

[0017] Where: ΔC G,up Increase the cost of conventional generator sets; ΔC G,down Reduce the cost of conventional generators; ΔC ES Adjust the cost for energy storage equipment; ΔC M Adjust the cost of electricity purchase and sales transactions with the main grid; C loss Adjustment fee for wind power curtailment penalty; C r Penalty fees for gas turbine regulation failures;

[0018] S14. Determine the constraint conditions to ensure the stability of the pre-adjustment stage;

[0019] S2. Construct data-driven uncertainty sets;

[0020] S3. Model solution;

[0021] S31. Based on the two-stage scheduling optimization model, an adaptive robust two-stage optimization model is constructed by combining the pre-scheduling stage and the real-time control stage objectives. According to the economic scheduling scheme and the uncertain probability distribution set, the final objective function is determined, and the expression is:

[0022]

[0023] Where: x is the most economical day-ahead dispatch plan under the worst wind power scenario; y is the most economical real-time control plan corresponding to the day-ahead dispatch plan x.

[0024] S32. Solve the final objective function.

[0025] Furthermore, the expressions of the variables in the objective function of the pre-scheduling stage are:

[0026]

[0027] Where: N c is the number of conventional generator sets in operation during the τ time period; is the power generated by conventional generator set i in the τ time period; a i , b i , c i is the traditional energy consumption coefficient of conventional generator set i; g q,τ is the operating cost of gas turbine q in time period τ; Q r is the number of gas turbines running in the time period τ; f j,τ N is the operating cost of fan j in the time period τ; w is the number of fan operations in the τ time period; u i,τ is the operating status of the conventional generator set i in the τ time period, u = 1 means the set is running, and u = 0 means the set is stopped; and are the start and stop costs of unit i in the τ time period; C ch , C dis as well as are the charge and discharge coefficient and charge and discharge power of the energy storage device respectively; as well as They are the transaction time-of-use electricity price and the purchased and sold electricity power between the microgrid and the main grid in the τ time period; h1 and h2 are the inequality constraint and equality constraint respectively; ω e , is the energy utilization rate at the current stage and the standard energy utilization rate; R τ Reward income for each unit improvement in energy efficiency; The penalty fee for each additional unit of pollutant exhaust gas; H τ , are the pollutant exhaust gas emissions during the operation phase and the allowable standards respectively; K1 and K2 are the penalty factors when the conditions are not met respectively.

[0028] Furthermore, the expressions of the variables in the objective function of the real-time control stage are:

[0029]

[0030] Where: is the penalty cost of increasing the power and unit power of conventional generator set i in the time period τ; is the power reduction and unit power reduction penalty cost of conventional generator set i in the time period τ; Adjust the power for charging and discharging the energy storage device respectively; M To adjust electricity prices; The power adjustment amount of the microgrid purchasing and selling electricity to the main grid; loss Penalty unit price for wind turbine curtailment; are the output and dispatch output of wind turbine j in time period τ respectively; r Penalty fees for unfeasible gas supply; is a slack variable, expressed as the air supply adjustment amount.

[0031] Furthermore, the constraint conditions include power balance constraint, wind power constraint, unit climbing constraint, landslide power constraint, energy storage device constraint and power interaction constraint.

[0032] Furthermore, the power balance constraint expression is:

[0033]

[0034] The expressions of wind power constraint, unit climbing constraint, landslide power constraint, energy storage device constraint and power interaction inequality constraint are:

[0035]

[0036] Where: They are the continuous operation and shutdown time of τ time period respectively; are the climbing rate and sliding rate of conventional unit i respectively; S ES,min , S ES,max are the minimum and maximum storage capacity of the energy storage system respectively; The energy storage system stores energy at time τ; are the electricity bought and sold at time τ; P ch,max , P dis,max are the maximum charging power and the maximum discharging power of the energy storage device respectively; η c , η d are the charging and discharging coefficients of the energy storage system respectively; It is an auxiliary variable. Its value is 1, indicating that there is electricity transaction with the main grid, and its value is 0, indicating that there is no electricity transaction with the main grid. are the charging and discharging auxiliary variables of the energy storage system at time τ respectively.

[0037] Furthermore, the step of constructing a data-driven uncertainty set includes:

[0038] S21. Conduct training statistical analysis on large-scale historical sample data of wind power output and load demand, and determine the probability error under different scenarios and time periods;

[0039] S22. Use the classic scenario analysis method to reduce a large number of generated scenarios to M wind power sample scenarios;

[0040] S23. Randomly select M scenes as centroids and construct a centroid scene set:

[0041]

[0042] And the remaining scene collection:

[0043] H r ={ζ s ′}s′=1,2,…,KM s

[0044] S24. Calculate the scene distance from the remaining scene to the centroid scene:

[0045]

[0046] S25. Classify the remaining scenes into the nearest centroid to form a cluster set:

[0047] H cl ={C i}i=1,2,…,M s

[0048] S26. Calculate the sum of the pairwise distances between scenes in the cluster set, select new cluster centers and re-determine the centroid set:

[0049]

[0050] S27. Determine whether the centroid and clustering results have changed. If not, solve the initial probability distribution of each discrete scene; otherwise, return to step S24;

[0051] S28. The wind power generation probability distribution is constrained by a comprehensive norm constraint set constructed by 1-norm and ∞-norm.

[0052] Furthermore, the constraint set is expressed as:

[0053]

[0054] Where: θ1, θ ∞ are the allowable deviation values ​​of the 1-norm and infinity-norm probability respectively.

[0055] Furthermore, the matrix form of the final objective function is:

[0056]

[0057]

[0058] In the formula, c T x is the day-ahead scheduling cost function; d T y is the real-time control cost function; u is the uncertainty.

[0059] Furthermore, the final objective function is decomposed into a main problem and sub-problems using a column constraint generation algorithm;

[0060] The main problem expression is:

[0061]

[0062] The sub-problem expression is:

[0063]

[0064] Furthermore, the steps for solving the main problem and the sub-problems include:

[0065] S321. Initialize and set the lower limit value LB = -∞, the upper limit value UB = +∞, and the number of iterations h = 0

[0066] and convergence deviation ε;

[0067] S322. Given the worst scenario u q , solve the main problem and get the optimal solution x q , as the lower limit of wakefulness

[0068] S323. x q Substitute the sub-problems, consider the uncertainty, and calculate the control amount when the real-time control cost is the lowest in each scenario;

[0069] S324. In the uncertainty set Ω, find the maximum expectation of the control cost and its corresponding optimization decision variable, so that the sum of the results of the main problem and sub-problems is the upper limit value

[0070] S325. Update the worst scenario and its probability distribution. If ε, then end; if not, return to S321.

[0071] Compared with the prior art, the beneficial effects are:

[0072] The present invention uses the K-means clustering method to construct typical scenarios to represent a large number of complex scenarios, and then constructs a wind power uncertainty set through a data-driven method to describe the wind power output distribution, effectively eliminating unnecessary extreme scenarios and reducing the model conservatism. Considering factors such as wind power output uncertainty, a two-stage adaptive robust optimization model including a microgrid day-ahead pre-dispatch model and a real-time dispatch model is established. The column constraint generation algorithm (C&CG) is used to decouple the model into main problems and sub-problems for interactive iterative solutions. The two-stage dispatch optimization model of the method described in the present invention fully considers the uncertainty of renewable energy generation and load, reduces the operating cost of microgrid equipment, and improves the utilization rate of new energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 This is a microgrid optimization scheduling architecture diagram based on data-driven model pre-training and rolling training of the present invention.

[0074] Figure 2 It is a K-means algorithm scenario reduction flow chart of a data-driven microgrid double-layer robust optimization scheduling strategy of the present invention.

[0075] Figure 3 It is a flow chart of solving a two-stage adaptive robust optimization model using a C&CG algorithm based on a data-driven microgrid double-layer robust optimization scheduling strategy of the present invention. DETAILED DESCRIPTION

[0076] The present invention will be further explained and illustrated below in conjunction with the embodiments, but the specific embodiments do not limit the present invention in any form.

[0077] Example 1

[0078] This embodiment provides a data-driven two-layer robust optimization scheduling method for a microgrid, the steps comprising:

[0079] S1. Constructing a theoretical framework

[0080] S11. Establish a microgrid optimization dispatching architecture for model pre-training and rolling training, such as Figure 1 As shown, it includes pre-training (offline training) and rolling training (online training), the pre-training consists of a data set and an offline training module, and the rolling training consists of a data set, online training, and an optimization scheduling module. Offline training mainly uses the idle time before and after the system's online training to train the microgrid's historical interaction data and microgrid simulation generation data, and determines typical scenarios of uncertain parameters to represent a large number of complex scenarios. Online training mainly uses the data from the pre-scheduling stage of offline training, plus the incremental sample data generated in the real-time scheduling stage of the microgrid for real-time rolling training, to achieve the minimum cost optimization in the pre-scheduling stage and the decision optimization of the minimum additional control cost and wind abandonment penalty cost in the real-time control stage, and obtain the optimized scheduling results that meet the real-time working conditions.

[0081] S12. Construct a two-stage scheduling optimization model, expressed as:

[0082]

[0083] In the formula, x is the first-stage decision variable, which represents the unit start-up and shutdown plan; g(x) is the objective function of the first-stage optimization problem; ζ is a random variable, which represents the uncertainty of renewable energy generation and load; y is the second-stage decision variable; f(y) is the objective function of the second-stage optimization problem; Ω(x,ζ) is the feasible domain of y; U is the uncertainty set of the random variable ζ;

[0084] S13. In the day-ahead pre-dispatch stage of the microgrid, determine the day-ahead pre-dispatch plan of the microgrid, take the selected microgrid historical interaction data and wind power forecast value as the benchmark scenario, consider the reward and punishment costs of the microgrid market and the operation and maintenance costs of the equipment, determine the start and stop plan and power generation of the microgrid generator set, the charging and discharging plan of the energy storage system, and the power purchase and sales plan with the main grid, and achieve the dispatching goal of the optimal operating cost in the pre-dispatch stage of the microgrid. The objective function expression of the pre-dispatch stage is:

[0085]

[0086] Where: G is the total cost; T is the total number of time periods, and T=24 in the pre-scheduling stage; is the operating cost of conventional units, gas turbines and fans in the τ time period; is the start-up and shutdown cost of conventional generator set i in time period τ; is the charging and discharging cost of the energy storage device in the τ time period; is the transaction cost between the microgrid and the main grid in the τ period; For reward and punishment expenses;

[0087] Specifically, the expressions of each variable are:

[0088]

[0089] Where: N c is the number of conventional generator sets in operation during the τ time period; is the power generated by conventional generator set i in the τ time period; a i , b i , c i is the traditional energy consumption coefficient of conventional generator set i; g q,τ is the operating cost of gas turbine q in time period τ; Q r is the number of gas turbines running in the time period τ; f j,τ N is the operating cost of fan j in the time period τ; w is the number of fan operations in the τ time period; u i,τ is the operating status of the conventional generator set i in the τ time period (u=1 indicates that the set is running, u=0 indicates that the set is stopped); and are the start and stop costs of unit i in the τ time period; C ch , C dis as well as are the charge and discharge coefficient and charge and discharge power of the energy storage device respectively; as well as They are the transaction time-of-use electricity price and the purchased and sold electricity power between the microgrid and the main grid in the τ time period; h1 and h2 are the inequality constraint and equality constraint respectively; ω e , is the energy utilization rate at the current stage and the standard energy utilization rate; R τ Reward income for each unit improvement in energy efficiency; The penalty fee for each additional unit of pollutant exhaust gas; H τ , are the pollutant exhaust gas emissions during the operation phase and the allowable standards respectively; K1 and K2 are the penalty factors when the conditions are not met respectively.

[0090] S14. In the real-time control stage of the microgrid, due to the complex uncertainties of wind power generation, there is a deviation between the predicted value of wind power generation and the actual output, resulting in unbalanced power in the system. According to the pre-dispatch plan, wind power generation is controlled in real time to achieve the optimization of real-time control cost and wind power output. Considering the flexible control method of energy storage equipment, ignoring its control cost during real-time control, and only considering the change in income when regulating the energy storage equipment, the objective function expression of the real-time control stage is:

[0091] minF=ΔC G,up +ΔC G,down +ΔC ES +ΔC M +C loss +maxC r

[0092] Where: ΔC G,up Increase the cost of conventional generator sets; ΔC G,down Reduce the cost of conventional generators; ΔC ES Adjust the cost for energy storage equipment; ΔC M Adjust the cost of electricity purchase and sales transactions with the main grid; C loss Adjustment fee for wind power curtailment penalty; C r Penalty fees for gas turbine regulation failures;

[0093] Specifically, the expressions of each variable are:

[0094]

[0095] Where: is the penalty cost of increasing the power and unit power of conventional generator set i in the time period τ; is the power reduction and unit power reduction penalty cost of conventional generator set i in the time period τ; Adjust the power for charging and discharging the energy storage device respectively; M To adjust electricity prices; The power adjustment amount of the microgrid purchasing and selling electricity to the main grid; loss Penalty unit price for wind turbine curtailment; are the output and dispatch output of wind turbine j in time period τ respectively; r Penalty fees for unfeasible gas supply; is a slack variable, expressed as the air supply adjustment amount.

[0096] S15. In order to ensure that all equipment on the energy supply side of the microgrid can operate in a coordinated and stable manner during the pre-dispatch stage, the constraints mainly include power balance constraints, wind power constraints, unit climbing constraints, landslide power constraints, energy storage equipment constraints, and power interaction inequality constraints.

[0097] The power balance constraint expression is:

[0098]

[0099] The expressions of wind power constraint, unit climbing constraint, landslide power constraint, energy storage device constraint and power interaction inequality constraint are:

[0100]

[0101] Where: They are the continuous operation and shutdown time of τ time period respectively; are the climbing rate and sliding rate of conventional unit i respectively; S ES,min , S ES,max are the minimum and maximum storage capacity of the energy storage system respectively; The energy storage system stores energy at time τ; are the electricity bought and sold at time τ; P ch,max , P dis,max are the maximum charging power and the maximum discharging power of the energy storage device respectively; η c , η d are the charging and discharging coefficients of the energy storage system respectively; It is an auxiliary variable. Its value is 1, indicating that there is electricity transaction with the main grid, and its value is 0, indicating that there is no electricity transaction with the main grid. are the charging and discharging auxiliary variables of the energy storage system at time τ respectively.

[0102] S2. Construct data-driven uncertainty sets;

[0103] S21. Conduct training statistical analysis on large-scale historical sample data of wind power output and load demand, and determine the corresponding probability distribution of actual data and typical data in each period of the day.

[0104] S22. Determine the probability error in different scenarios and time periods based on the roulette method.

[0105] S23. Different scenarios are obtained based on the scenario analysis method, and a large number of generated scenarios are reduced to M wind power sample scenarios using the classic scenario analysis method, and the M wind power sample scenarios are reduced and clustered using the K-means clustering algorithm.

[0106] S624. The wind power generation probability distribution is constrained by a comprehensive norm constraint set constructed by 1-norm and ∞-norm. The constraint set expression is:

[0107]

[0108] In the formula, θ1, θ ∞are the allowable deviation values ​​of the 1-norm and infinity-norm probability respectively.

[0109] S3. Model solution;

[0110] S31. Based on the two-stage dispatch optimization model, an adaptive robust two-stage optimization model is constructed by combining the pre-dispatch stage and the real-time control stage objectives. The most economical day-ahead dispatch scheme under the worst wind power output scenario is considered, and the uncertain probability distribution set of wind power generation driven by data is considered at the same time. The final objective function is determined, and the expression is:

[0111]

[0112] Where: x is the most economical day-ahead dispatch plan under the worst wind power scenario; y is the most economical real-time control plan corresponding to the day-ahead dispatch plan x.

[0113] S32. Solve the final objective function.

[0114] S321. The final objective function is written in compact matrix form as:

[0115]

[0116]

[0117] S322. Decompose the main problem and sub-problems using the column constraint generation algorithm;

[0118] The main problem expression is:

[0119]

[0120] The sub-problem expression is:

[0121]

[0122] Where m is the number of iterations.

[0123] S323. Call the CPLEX solver in Matlab to solve the problem directly.

[0124] Example 2

[0125] This embodiment provides a calculation method for reducing and clustering M wind power sample scenarios using a K-means clustering algorithm, such as Figure 2 , the steps include:

[0126] (1) Randomly select M scenes as the centroid and construct a centroid scene set: And the remaining scene collection: H r ={ζ s ′}s′=1,2,…,KMs ;

[0127] (2) Calculate the scene distance from the remaining scene to the centroid scene:

[0128] (3) Classify the remaining scenes into the nearest centroid to form a cluster set: H cl ={C i}i=1,2,…,M s ;

[0129] (4) Calculate the sum of the pairwise distances between scenes in the cluster set: Select new cluster centers and redefine the centroid set;

[0130] (5) Determine whether the centroid and clustering results have changed. If not, solve the initial probability distribution of each discrete scene; otherwise, return to step (2).

[0131] Example 3

[0132] This embodiment provides the steps for solving the main problem and sub-problems, such as Figure 3 ,include:

[0133] (1) Initialize the lower limit LB = -∞, the upper limit UB = +∞, and the number of iterations h = 0

[0134] and convergence deviation ε;

[0135] (2) Given the worst scenario u q , solve the main problem and get the optimal solution x q , as the lower limit of wakefulness

[0136] (3) x q Substitute the sub-problems, consider the uncertainty, and calculate the control amount when the real-time control cost is the lowest in each scenario;

[0137] (4) In the uncertainty set Ω, find the expectation of the maximum control cost and its corresponding optimization decision variable, so that the sum of the results of the main problem and sub-problems is used as the upper limit value

[0138] (5). Update the worst scenario and its probability distribution, if. ε, then end; if not, return to S71.

[0139] Experimental example

[0140] (1) The ARO method of the present invention is compared with the traditional robust method, and the results are shown in Table 1 below:

[0141] Table 1

[0142] Comparison of optimization methods Microgrid revenue / yuan Objective function deviation ratio / % Iterations Calculation time / s ARO Method 8960 0.13 4 158.3 Traditional robust optimization 4750 0.75 30 537.5

[0143] As can be seen from Table 1, the deviation of the objective function result obtained by the ARO optimization method of the present invention is lower than that of the traditional robust optimization method, and the deviation range of the result is between 0.1% and 0.85%. The traditional robust optimization needs more iterations to reach the convergence condition, and the calculation time is longer.

[0144] (2) The cost comparison results of the ARO method of the present invention and the traditional scheduling method are shown in Table 2 below:

[0145] Table 2

[0146]

[0147]

[0148] It can be seen from Table 2 that the total cost of the traditional dispatching method is higher than the total cost of the ARO dispatching of the present invention. When formulating the pre-dispatching plan, the traditional dispatching method fails to consider the influence of wind power uncertainty and other factors. Its pre-dispatching plan has poor robustness and weak ability to resist the uncertainty of wind power changes. It can be seen that the regulation cost and wind abandonment cost in the subsequent regulation stage will increase accordingly. The wind power generation uncertainty set established by the method proposed in the article excludes impossible extreme scenarios, so that the optimal solution is solved in a smaller and more accurate feasible domain, which can better reflect the possible actual wind power scenarios, and has a strong ability to resist uncertain parameters, so that the regulation cost and wind abandonment cost are lower, so that the total cost of the two stages of the ARO dispatching method is lower than the total cost of the two stages of the traditional dispatching method, so that the microgrid obtains better economic benefits.

[0149] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A data-driven two-layer robust optimization scheduling method for microgrids, characterized in that the steps include: S1. Construct theoretical framework; S11. Establish a microgrid optimization dispatching architecture with model pre-training and rolling training, and construct a two-stage dispatching optimization model, which is expressed as: In the formula, x is the first-stage decision variable, which represents the unit start-up and shutdown plan; g(x) is the objective function of the first-stage optimization problem; ζ is a random variable, which represents the uncertainty of renewable energy generation and load; y is the second-stage decision variable; f(y) is the objective function of the second-stage optimization problem; Ω(x,ζ) is the feasible domain of y; U is the uncertainty set of the random variable ζ; S12. Determine the dispatch target with the optimal operating cost in the pre-dispatching stage of the microgrid. The objective function expression of the pre-dispatching stage is: Where: G is the total cost; T is the total number of time periods, and T=24 is taken in the pre-scheduling stage; is the operating cost of conventional units, gas turbines and fans in the τ time period; is the start-up and shutdown cost of conventional generator set i in time period τ; is the charging and discharging cost of the energy storage device in the τ time period; is the transaction cost between the microgrid and the main grid in the τ period; For reward and punishment expenses; S13. According to the change of benefits during energy storage equipment regulation, the real-time regulation cost and power generation output optimization goals, the objective function expression of the real-time regulation stage is determined as: minF=ΔC G,up +ΔC G,down +ΔC ES +ΔC M +C loss +maxC r Where: ΔC G,up Increase the cost of conventional generator sets; ΔC G,down Reduce the cost of conventional generators; ΔC ES Adjust the cost for energy storage equipment; ΔC M Adjust the cost of electricity purchase and sales transactions with the main grid; C loss Adjustment fee for wind power curtailment penalty; C r Penalty fees for gas turbine regulation failures; S14. Determine the constraint conditions to ensure the stability of the pre-adjustment stage; S2. Construct data-driven uncertainty sets; S3. Model solution; S31. Based on the two-stage scheduling optimization model, an adaptive robust two-stage optimization model is constructed by combining the pre-scheduling stage and the real-time control stage objectives. According to the economic scheduling scheme and the uncertain probability distribution set, the final objective function is determined, and the expression is: Where: x is the most economical day-ahead dispatch plan under the worst wind power scenario; y is the most economical real-time control plan corresponding to the day-ahead dispatch plan x. S32. Solve the final objective function.

2. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 1 is characterized in that: The expressions of the variables in the objective function of the pre-scheduling stage are: Where: N c is the number of conventional generator sets in operation during the τ time period; is the power generated by conventional generator set i in the τ time period; a i , b i , c i is the traditional energy consumption coefficient of conventional generator set i; g q,τ is the operating cost of gas turbine q in time period τ; Q r is the number of gas turbines running in the time period τ; f j,τ N is the operating cost of fan j in the time period τ; w is the number of fan operations in the τ time period; u i,τ is the operating status of the conventional generator set i in the τ time period, u = 1 means the set is running, and u = 0 means the set is stopped; and are the start and stop costs of unit i in the τ time period; C ch , C dis as well as are the charge and discharge coefficient and charge and discharge power of the energy storage device respectively; as well as They are the transaction time-of-use electricity price and the purchased and sold electricity power between the microgrid and the main grid in the τ time period; h1 and h2 are the inequality constraint and equality constraint respectively; ω e ,ω e0 is the energy utilization rate at the current stage and the standard energy utilization rate; R τ Reward income for each unit improvement in energy efficiency; The penalty fee for each additional unit of pollutant exhaust gas; H τ , are the pollutant exhaust gas emissions during the operation phase and the allowable standards respectively; K1 and K2 are the penalty factors when the conditions are not met respectively.

3. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 1 is characterized in that: The expressions of each variable in the objective function of the real-time control stage are: Where: is the penalty cost of increasing the power and unit power of conventional generator set i in the time period τ; is the power reduction and unit power reduction penalty cost of conventional generator set i in the time period τ; Adjust the power for charging and discharging the energy storage device respectively; M To adjust electricity prices; The power adjustment amount of the microgrid purchasing and selling electricity to the main grid; loss Penalty unit price for wind turbine curtailment; are the output and dispatch output of wind turbine j in time period τ respectively; r Penalty fees for unfeasible gas supply; is a slack variable, expressed as the air supply adjustment amount.

4. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 1 is characterized in that: The constraints include power balance constraints, wind power constraints, unit climbing constraints, landslide power constraints, energy storage equipment constraints and power interaction constraints.

5. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 4 is characterized in that: The power balance constraint expression is: The expressions of wind power constraint, unit climbing constraint, landslide power constraint, energy storage device constraint and power interaction inequality constraint are: Where: They are the continuous operation and shutdown time of τ time period respectively; are the climbing rate and sliding rate of conventional unit i respectively; S ES,min , S ES,max are the minimum and maximum storage capacity of the energy storage system respectively; The energy storage system stores energy at time τ; are the electricity bought and sold at time τ; P ch,max , P dis,max They are the maximum charging power and the maximum discharging power of the energy storage device respectively; η c , η d are the charging and discharging coefficients of the energy storage system respectively; It is an auxiliary variable. Its value is 1, indicating that there is electricity transaction with the main grid, and its value is 0, indicating that there is no electricity transaction with the main grid. are the charging and discharging auxiliary variables of the energy storage system at time τ respectively.

6. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 1 is characterized in that: The step of constructing a data-driven uncertainty set includes: S21. Conduct training statistical analysis on large-scale historical sample data of wind power output and load demand, and determine the probability error under different scenarios and time periods; S22. Use the classic scenario analysis method to reduce a large number of generated scenarios to M wind power sample scenarios; S23. Randomly select M scenes as centroids and construct a centroid scene set: And the remaining scene collection: H r ={ζ s ′}s′=1,2,…,KM s S24. Calculate the scene distance from the remaining scene to the centroid scene: S25. Classify the remaining scenes into the nearest centroid to form a cluster set: H cl ={C i }i=1,2,…,M s S26. Calculate the sum of the pairwise distances between scenes in the cluster set, select new cluster centers and re-determine the centroid set: S27. Determine whether the centroid and clustering results have changed. If not, solve the initial probability distribution of each discrete scene; otherwise, return to step S24; S28. The wind power generation probability distribution is constrained by a comprehensive norm constraint set constructed by 1-norm and ∞-norm.

7. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 6 is characterized in that: The constraint set is expressed as: Where: θ1, θ ∞ are the allowable deviation values ​​of the 1-norm and infinity-norm probability respectively.

8. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 1 is characterized in that: The matrix form of the final objective function is: In the formula, c T x is the day-ahead scheduling cost function; d T y is the real-time control cost function; u is the uncertainty.

9. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 8 is characterized in that: The final objective function is decomposed into a main problem and sub-problems using a column constraint generation algorithm; The main problem expression is: The sub-problem expression is:

10. The data-driven two-layer robust optimization scheduling method for microgrids according to claim 9 is characterized in that: The steps to solve the main problem and subproblems include: S321. Initialize and set the lower limit LB = -∞, the upper limit UB = +∞, the number of iterations h = 0 and the convergence deviation ε; S322. Given the worst scenario u q , solve the main problem and get the optimal solution x q , as the lower limit of wakefulness S323. x q Substitute the sub-problems, consider the uncertainty, and calculate the control amount when the real-time control cost is the lowest in each scenario; S324. In the uncertainty set Ω, find the maximum expectation of the control cost and its corresponding optimization decision variable, so that the sum of the results of the main problem and sub-problems is the upper limit value S325. Update the worst scenario and its probability distribution. If ε, then end; if not, return to S321.

Images