Two-stage robust optimization scheduling method for source-load power multi-interval uncertainty set

By constructing a multi-interval uncertain set and two-stage robust optimization scheduling method in a DC microgrid, the problem of excessive wide boundary of a single-interval description uncertain set and the probability characteristics of multiple intervals are solved, and the source of the DC microgrid is realized The precise characterization of the load uncertain variable improves the utilization rate of distributed energy and dispatch economy.

CN118920603AInactive Publication Date: 2024-11-08HUNAN UNIV OF TECH

Patent Information

Application Number
CN202410956330.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In the prior art, when building uncertain sets in DC microgrids, single interval uncertain sets describe uncertain variables with too wide fluctuation boundaries, resulting in too conservative optimization results, while multi-interval uncertain sets fail to accurately characterize the fluctuation probability of uncertain factors. Features affect the accuracy and economicality of robust optimization.

Method used

Using a two-stage robust optimization scheduling method of multi-interval uncertain set of source load power, the photovoltaic, wind power and user load power are predicted by constructing a multi-interval uncertain set based on historical data, combining standard Gaussian process model and Bayesian law. Determine the confidence interval and build a DC microgrid operation model to optimize the two-stage robust optimization scheduling scheme of the DC microgrid.

Benefits of technology

The accurate characterization of the uncertain variables of source load in the DC microgrid is achieved, the utilization rate of distributed energy is improved, the load redundancy is reduced, and the economic and accuracy of scheduling is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118920603A_ABST
    Figure CN118920603A_ABST
Patent Text Reader

Abstract

The invention provides a two-stage robust optimization scheduling method for a source-load power multi-interval uncertainty set. The method comprises the following steps: firstly, carrying out mathematical modeling on uncertain variables contained in a power system through a Gaussian distribution mathematical theory and a Gaussian process regression prediction algorithm; according to historical data of an uncertain variable, confidence intervals of the uncertain variable under different confidence coefficients are fitted, and a multi-interval uncertain set containing source-load probability characteristics is constructed by using correlation between probability characteristics of the confidence intervals and fluctuation frequency of the uncertain variable, so that reasonable modeling of the uncertain variable is realized. Compared with other robust optimization scheduling schemes of various uncertain sets, the method has the advantages that uncertain variables are accurately depicted, a high interval coverage rate is achieved with a low interval width, distributed energy and user load fluctuation ranges in the direct-current micro-grid are more focused, the utilization rate of the distributed energy is increased, load redundancy is reduced, and the distributed energy utilization rate is increased. And the economy of the scheduling scheme is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of industrial electrical control, and in particular to a two-stage robust optimization scheduling method for a source-load power multi-interval uncertain set. Background Art

[0002] Distributed energy has strong randomness and intermittency. Its output regulation and distribution under large-scale grid connection is a hot issue for grid stability and economic operation. As a flexible new energy architecture, DC microgrid contains a large number of power electronic devices, which can realize efficient consumption of distributed energy through reasonable scheduling. At present, the accurate description and optimal scheduling of uncertain factors in distributed energy are the key to solving the problem of distributed energy optimization scheduling.

[0003] The uncertainty optimization scheduling algorithms for uncertain factors mainly include robust optimization (RO) and stochastic optimization (SO). Stochastic planning is to reduce the generated uncertain scenarios and perform deterministic analysis on the typical scenarios obtained by reduction, but its planning results are difficult to cope with the extreme operating conditions of DC microgrids. In comparison, robust optimization fully considers the extreme operating scenarios of the power system, and by constructing an uncertainty set to describe the fluctuation of uncertain factors, it ensures that the optimization results can cover all possible scenarios in the power system. Therefore, how to construct an accurate, comprehensive and reasonable uncertainty set plays a decisive role in the effect of robust optimization.

[0004] Uncertainty set is a mathematical set used to describe the fluctuation range of distributed energy and user load in DC microgrid. In recent years, domestic and foreign scholars have conducted a lot of research on the construction of uncertainty set, which can be divided into two categories: single interval uncertainty set and multi-interval uncertainty set. Some scholars proposed to describe the uncertainty of distributed energy output by box uncertainty set, but failed to accurately characterize the fluctuation boundary of uncertain variables, and the boundary of uncertainty set was too wide, making the optimization result too conservative. Some literature proposed an ellipsoid uncertainty set for the problem of power grid expansion planning, which better described the fluctuation range of uncertain disturbances, but the solution efficiency was low when dealing with high-dimensional nonlinear uncertainty problems. The above studies all adopted a single interval form to model uncertain factors, and the uncertainty fluctuation concentration of the optimization results appeared at the boundary of the uncertainty set. However, in actual scenarios, the probability of the boundary appearing in the uncertainty fluctuation concentration is low, and the description of the single interval uncertainty set is poorly matched with the actual scenario. In view of the shortcomings of the single interval uncertainty set, some literature constructed a multi-interval uncertainty set based on the ladder characteristics of hydropower resources, but did not consider the accurate division of the boundary of the ladder interval of hydropower resources. Compared with single-interval uncertainty sets, multi-interval uncertainty sets can more accurately describe the fluctuation of uncertain variables in actual scenarios and improve the polymorphism of uncertain factors. However, in the existing research on multi-interval uncertainty sets, the fluctuation probability characteristics of uncertain factors are not characterized, making the robust optimization scheme inaccurate and incomplete. Summary of the invention

[0005] The purpose of the present invention is to provide a two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets, so as to accurately characterize the uncertain variables of source and load in DC microgrid, improve the utilization rate of distributed energy, reduce load redundancy, and improve the economy of scheduling.

[0006] In order to solve the above technical problems, the present invention adopts the following technical method: a two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets, comprising:

[0007] Step S1, constructing a multi-interval uncertainty set considering the uncertain fluctuation probability characteristics of source and load based on the historical data of the DC microgrid;

[0008] Step S2, constructing a DC microgrid operation model including distributed renewable energy;

[0009] Step S3, considering the uncertain factors of distributed renewable energy and user load power, taking the operation economy of the DC microgrid as the optimization goal, introducing the multi-interval uncertainty set obtained in step S1 into the DC microgrid operation model constructed in step S2, and constructing a two-stage robust optimization scheduling model for the DC microgrid;

[0010] Step S4, performing optimization scheduling solution on the DC microgrid two-stage robust optimization scheduling model obtained in step S3, and obtaining a DC microgrid robust optimization scheduling scheme that takes into account the uncertain fluctuation probability characteristics of sources and loads.

[0011] Further, step S1 includes:

[0012] S101, collecting the photovoltaic, wind power and user load power data of the previous day to form a training set and a test set;

[0013] Construct a standard Gaussian process model, using the photovoltaic, wind power and user load power data in the training set as the input vector and output vector of the standard Gaussian process model:

[0014] y=f(x)+ε(1)

[0015] Where x=[x1,x2,...,x n ] T represents the input vector of the standard Gaussian process model, y=[y1,y2,...,y n ] T represents the output vector of the standard Gaussian process model, ε represents the amount of noise that independently satisfies the normal distribution, and its mean is 0 and its variance is σ 2 , denoted as ε~N(0,σ 2 );

[0016] The Gaussian process regression is used to build the photovoltaic, wind and user load power prediction model based on Bayes' theorem, and the photovoltaic, wind and user load day-ahead power data in the test set are used as the input vector and output vector of the prediction model:

[0017]

[0018] Among them, y * Represents the output vector of the prediction model, that is, the prediction data result; x * represents the input vector of the prediction model; K(x,x), K(x,x * )、K(x * ,x) and K(x * ,x * ) represents the covariance matrix; I represents the identity matrix;

[0019] The prediction results satisfying the Gaussian probability distribution are obtained from the above prediction model:

[0020]

[0021] Among them, μ * Represents the mean of the predicted data results; Indicates the variance of the predicted data results;

[0022] Based on the fact that the distributed energy output and user load in the actual DC microgrid are not negative, the confidence interval of the input vector in the prediction model is determined as:

[0023] [L 1-α ,U 1-α ]=[max(0,μ * -z α / 2 σ * ),μ * +z α / 2 σ * ] (4)

[0024] Among them, U 1-α With L 1-α They represent the upper and lower limits of the confidence interval under the condition of confidence level 1-α; α represents the significance level; z α / 2 is the bilateral α quantile of the standard Gaussian distribution;

[0025] S102. Based on the actual fluctuation characteristics of uncertain factors, a multi-interval uncertainty set is constructed. Then, based on the probability distribution information contained in the historical data of uncertain factors, the number of fluctuations of uncertain variables within a day and the fluctuation probability interval are constrained as follows:

[0026]

[0027] Among them, U φ represents the multi-interval probability uncertainty set of equipment φ, which includes wind turbines, photovoltaic power generation equipment and user load equipment; u φ,t represents the uncertain decision value of the device φ in the period t; N i Represents the number of intervals in a multi-interval uncertainty set; N t Indicates the total number of segments in time period t; represents the mean power of the equipment φ in the power system during period t; and They represent the upper and lower boundary values ​​of the confidence interval of device φ in time period i, respectively; and They respectively represent that the device φ is in the upper boundary fluctuation state and the lower boundary fluctuation state of the confidence interval i during period t; μ *φ,t Represents the mean value of the predicted data results of device φ in period t; 1-α i Indicates different confidence levels.

[0028] Furthermore, in step S2, when constructing the DC microgrid operation model, a corresponding mathematical model is established for each component in the DC microgrid according to the topological structure of the DC microgrid and the electrical connection relationship of each device, including:

[0029] 1) DC microgrid power balance mathematical model, its power balance constraint formula is as follows:

[0030]

[0031] in, and They represent the active power purchased and sold by the DC microgrid to the external AC grid during period t; P L (t) represents the active power consumed by the user load in the DC microgrid during period t; P PV (t) and P WT (t) represents the active power output of photovoltaic power generation and wind power generation in the DC microgrid during period t; P G (t) represents the active power generated by the controllable generator set in the DC microgrid during period t; and represents the discharge and charge of the energy storage system in the DC microgrid during period t;

[0032] 2) Voltage drop and network loss mathematical model, with the following constraints:

[0033]

[0034] Where: P g,i is the sum of the power flowing into node i; P d,i is the load power of node i; E + ={(i,j)|i<j;i,j∈N} and E - ={(i,j)|i>j; i,j∈N} is a set of directed paths, with the positive direction being i→j; E + is the set of branches flowing into node i; E - is the branch set out of node i; P ij is the power flowing between nodes i and j; V i With V j are the voltages of node i and node j respectively; R ij is the line impedance between nodes i and j; I ij is the current between nodes i and j; P loss represents the line loss between nodes i and j; F represents the vector of nodes; L represents the line between nodes;

[0035] 3) Mathematical model of controllable generator set, and its output power constraint:

[0036]

[0037] in, and They represent the maximum output power and minimum output power of the generator set during period t, and their values ​​must take into account the limitations of the rated power and minimum load rate of the generator set; P G (t) is the output value of the controllable generator at time t; P G_CP represents the ramp power of the controllable generator; the power generation cost C of the generator set in period t G (t) are as follows:

[0038] C G (t) = aP G (t) (9)

[0039] Where: a represents the fuel cost of the generator set;

[0040] 4) Mathematical model of energy storage system, which operates with the following constraints:

[0041]

[0042] in, Indicates the maximum charging and discharging power per unit time of the energy storage system; s SOC (t) represents the charging and discharging state of the energy storage system in the DC microgrid during period t. SOC When (t) is 1, the energy storage system in the DC microgrid is in a charging state. SOC When (t) is 0, the energy storage system in the DC microgrid is in a discharging state; and Respectively represent the minimum remaining capacity and the maximum remaining capacity of the energy storage system; E SOC (0) represents the initial capacity of the energy storage system at time 0; η ch With η dis They represent the operating efficiency of the energy storage system in the charging state and the operating efficiency in the discharging state respectively;

[0043] 5) Mathematical model of energy interaction between DC microgrid and AC grid. The interaction power of this model must satisfy:

[0044]

[0045] in, It represents the maximum power of power interaction between the DC microgrid and the external AC grid. Its value depends on the maximum interactive power of the bidirectional grid-connected inverter at the connection node between the DC microgrid and the external AC grid. M (t) represents the energy interaction state between the DC microgrid and the external AC grid during period t. M When the value of (t) is 1, the bidirectional grid-connected inverter between the DC microgrid and the external AC grid operates in the power purchasing state. MWhen the value of (t) is 0, the bidirectional grid-connected inverter between the DC microgrid and the external AC grid operates in the power selling state; the energy interaction cost C between the DC microgrid and the external AC grid in period t M (t) is:

[0046]

[0047] Among them, K M (t) is the electricity price of the external AC grid during period t;

[0048] 6) The mathematical model of distributed renewable energy output, whose output numerical constraints are:

[0049]

[0050] Where: and Respectively represent the installed capacity of photovoltaic power generation equipment and wind power generation equipment, that is, their power upper limits; P PV (t) and P WT (t) represents the actual output of photovoltaic power generation device and wind power generation device during period t respectively; and They represent the rated output of the photovoltaic power generation device and the wind power generation device in time period t respectively; and They respectively represent the output deviation of the photovoltaic power generation device and the wind power generation device in time period t.

[0051] Furthermore, in step S3, the uncertain factors of photovoltaic power output, wind power output and user load power are taken into consideration, and the operation economy of the DC microgrid is taken as the optimization goal. The multi-interval uncertainty set obtained in step S1 is introduced into the DC microgrid operation model constructed in step S2, and a DC microgrid two-stage robust optimization scheduling model is constructed as follows:

[0052]

[0053] Where: b T δ represents the operating cost in robust optimization scheduling, b is the system constant matrix, ε and δ are the decision vectors of the first-stage scheduling and the second-stage scheduling respectively; U is a multi-probability interval uncertainty set, which contains the uncertainty description of photovoltaic output, wind power output and user load; u is an uncertain variable, including the output of photovoltaic power generation equipment, wind power generation equipment and user load in the DC power grid system; δ∈F(ε,u) represents the feasible domain of the decision vector δ when ε and u are known, and its expression is as follows:

[0054]

[0055] Among them, D, K, F, G and I uThey correspond to the coefficient matrices of the constraint variables of the robust one-stage and two-stage optimization of the DC microgrid respectively; d, k and h are constant column vectors.

[0056] Furthermore, in step S4, when optimizing and solving the two-stage robust optimization scheduling model of the DC microgrid obtained in step S3, a column constraint generation algorithm is used to decompose the two-stage robust optimization scheduling model into two solvable single-layer optimization problems. By alternately iterating the two decomposed problems, the optimal solution of the two-stage robust optimization scheduling model is finally obtained, that is, the working status and target power value of each device in the DC microgrid.

[0057] Furthermore, step S4 includes:

[0058] S401, the two-stage robust optimization scheduling model is decomposed by using the column constraint generation algorithm, so that the first stage of robust optimization is the outer min problem of the original model, and the second stage of robust optimization is the inner max-min problem of the model;

[0059] The first stage problem of robust optimization makes decisions on the charging and discharging state of the energy storage system and the power purchase and sale state of the bidirectional grid-connected inverter in the DC microgrid. The first stage model of robust optimization is shown in the following formula:

[0060]

[0061] Where l represents the current iteration number of the column constraint generation algorithm; δ s represents the solution of the first stage optimization in the sth iteration; Indicates the value of the uncertain variable u obtained in the sth iteration;

[0062] The objective function of the first stage of robust optimization is In the formula, ε=[s M (t),s SOC (t)];

[0063] Constraints of the first stage of robust optimization: The constraint set of the first stage of robust optimization is the set of all time constraints of equations (6) to (13). The additional conditions for each iteration are In the first iteration, since the second stage has not yet been solved and the objective function is the economic optimization of the system energy scheduling, we first set b T δ≥0, that is, the cost is assumed to be greater than 0;

[0064] The second stage problem of robust optimization is to make decisions on the output power of the generator set and the power purchase and sale value of the bidirectional grid-connected inverter according to the fluctuation of the output of distributed energy resources and the fluctuation of the user load power in the DC microgrid. The second stage model of robust optimization is:

[0065]

[0066] The objective function of the second stage of robust optimization is In the formula,

[0067] Constraints for the second stage of robust optimization:

[0068] The decision variables in the first stage of robust optimization are taken as constant matrices. The Lagrange multiplier matrix is ​​introduced into equations (6) to (13) respectively as equality constraints and inequality constraints. The inner layer problem transformed by KKT conditions is converted into equivalent constraints as follows:

[0069]

[0070] Where: L() is the Lagrangian function obtained by transformation; λ g is the Lagrange multiplier matrix of the inequality constraints; μ h is the Lagrange multiplier matrix of the equality constraint; N g The number of inequality constraints for the second-stage inner min problem for robust optimization;

[0071] The Big-M method is used to linearize the nonlinear term in equation (18), and we get:

[0072]

[0073] In the formula, q g is the introduced 0-1 variable, i.e. q g ∈{0,1}; M is the Big-M coefficient;

[0074] The max-min double-layer problem in the second stage of robust optimization is converted into a single-layer max optimization problem after KKT conditions and Big-M linearization, as shown in formula (20):

[0075]

[0076] S402, the result of the second-stage problem decomposed in S401 is updated to the decision variables and constraints of the first-stage problem, so that the two single-layer optimization problems are iterated alternately, the solution of the first-stage problem and the solution of the second stage are converged, and the optimal solution of the two-stage robust optimization scheduling model is obtained, that is, the working status and target power value of each device in the DC microgrid.

[0077] Preferably, in step S401, the second stage constraints of the robust optimization are refined and supplemented as follows:

[0078] 1) Refine formula (18) to:

[0079]

[0080] Where: λ1,...,λ 13 is the Lagrange multiplier matrix of inequality constraints; μ1, μ2 are the Lagrange multiplier matrices of equality constraints; K SOC is the charging and discharging cost of the energy storage system; K M is the external AC grid electricity price;

[0081] 2) Refine formula (19) to:

[0082]

[0083] λ3≤Mq3,P G Δt-P G_CP ≥-M(1-q3) (24)

[0084]

[0085] 3) The KKT conditional expression and multi-interval uncertainty set of the supplementary equality constraint are as follows:

[0086]

[0087] The two-stage robust optimization scheduling method for source-load power multi-interval uncertainty sets provided by the present invention first uses historical data in distributed energy and user loads to model uncertain factors, and then fits confidence intervals of uncertain factors at different confidence levels to associate the probability characteristics of the confidence intervals with the fluctuations of the uncertain factors, thereby constructing a multi-interval uncertainty set containing source-load probability characteristics to achieve accurate modeling of uncertain variables. Compared with traditional robust optimization scheduling schemes based on single probability interval uncertainty sets and other types of multi-interval uncertainty sets, the present invention achieves a higher interval coverage rate with a lower interval width through accurate characterization of uncertain variables, so that the fluctuation range of distributed energy and user load in the DC microgrid is more focused, thereby improving the utilization rate of distributed energy, reducing load redundancy, and improving the economy of the scheduling scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a DC microgrid system architecture diagram in the present invention;

[0089] Figure 2 It is the operation flow chart of the DC microgrid dispatching center in the present invention;

[0090] Figure 3 It is a multi-interval uncertainty set model diagram in the present invention;

[0091] Figure 4 It is a two-stage robust optimization scheduling solution flow chart in the present invention;

[0092] Figure 5is a confidence interval diagram of a multi-interval uncertainty set of user load in an implementation manner of the present invention;

[0093] Figure 6 is a confidence interval diagram of a multi-interval uncertainty set of a photovoltaic power generation device in an embodiment of the present invention;

[0094] Figure 7 is a confidence interval diagram of a multi-interval uncertainty set of a wind power generation device in an embodiment of the present invention;

[0095] Figure 8 It is a user load power scheduling diagram under a two-stage robust optimization scheduling method considering a multi-interval uncertainty set of source load power in an implementation manner of the present invention;

[0096] Fig. 9 It is a photovoltaic power generation equipment output diagram under a two-stage robust optimization scheduling method considering a multi-interval uncertainty set of source and load power in an embodiment of the present invention;

[0097] Fig.10 It is a wind power generation equipment output diagram under a two-stage robust optimization scheduling method considering a multi-interval uncertainty set of source and load power in an implementation manner of the present invention;

[0098] Fig.11 It is a diagram of the output power of the controllable generator set and the charging and discharging power of the energy storage under the two-stage robust optimization scheduling method considering the uncertain set of multiple intervals of source and load power in the implementation mode of the present invention;

[0099] Fig.12 It is an AC / DC power grid interaction power diagram under a two-stage robust optimization scheduling method considering a multi-interval uncertainty set of source and load power in an embodiment of the present invention;

[0100] Fig.13 It is a user load power diagram under the DC microgrid robust optimization scheduling method based on a single probability interval uncertainty set in an implementation manner of the present invention;

[0101] Fig.14 It is a photovoltaic power generation equipment output diagram under a DC microgrid robust optimization scheduling method based on a single probability interval uncertainty set in an implementation mode of the present invention;

[0102] Fig.15 It is a wind power generation equipment output diagram under the DC microgrid robust optimization scheduling method based on a single probability interval uncertainty set in an implementation mode of the present invention;

[0103] Fig.16 It is a user load power diagram under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on residual estimation in an implementation manner of the present invention;

[0104] Fig.17It is a wind power generation equipment output diagram under a DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on residual estimation in an implementation manner of the present invention;

[0105] Fig.18 It is a photovoltaic power generation equipment output diagram under a DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on residual estimation in an implementation manner of the present invention;

[0106] Fig.19 It is a diagram of the total output of each device under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on residual estimation in an implementation manner of the present invention;

[0107] Fig. 20 It is a total energy consumption diagram of each device under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on residual estimation in an implementation manner of the present invention;

[0108] Fig.21 It is a user load power diagram under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on polynomial fitting in an implementation manner of the present invention;

[0109] Fig. 22 It is a wind power generation equipment output diagram under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on polynomial fitting in an implementation manner of the present invention;

[0110] Fig.23 It is a photovoltaic power generation equipment output diagram under a DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on polynomial fitting in an implementation manner of the present invention;

[0111] Fig.24 It is a diagram of the total output of each device under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on polynomial fitting in an implementation manner of the present invention;

[0112] Fig.25 It is a diagram of the total energy consumption of each device under the DC microgrid robust optimization scheduling method of multiple interval uncertainty sets based on polynomial fitting in an implementation manner of the present invention. DETAILED DESCRIPTION

[0113] In order to facilitate the understanding of those skilled in the art, the present invention is further described below in conjunction with embodiments and drawings. The contents mentioned in the implementation modes are not intended to limit the present invention.

[0114] The DC microgrid system architecture of the present invention is as follows: Figure 1As shown in the figure, in order to ensure the efficient use of energy and safe operation of DC microgrids, a centralized optimization dispatching center for DC microgrids came into being. The dispatching center obtains the working status and power data of each device through the information flow in the DC microgrid, and optimizes and dispatches the electric energy reasonably according to the operating characteristics and energy supply and demand of different devices in the system, so as to ensure the safe and efficient operation of the DC microgrid.

[0115] The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets involved in the present invention refers to Figure 2 , the method includes the following steps.

[0116] Step S1, constructing a multi-interval uncertainty set considering the uncertain fluctuation probability characteristics of source and load based on the historical data of the DC microgrid.

[0117] S101, collecting the photovoltaic, wind power and user load power data of the previous day to form a training set and a test set;

[0118] Construct a standard Gaussian process model, using the photovoltaic, wind power and user load power data in the training set as the input vector and output vector of the standard Gaussian process model:

[0119] y=f(x)+ε (1)

[0120] Where x=[x1,x2,...,x n ] T represents the input vector of the standard Gaussian process model, y=[y1,y2,...,y n ] T represents the output vector of the standard Gaussian process model, ε represents the amount of noise that independently satisfies the normal distribution, and its mean is 0 and its variance is σ 2 , denoted as ε~N(0,σ 2 );

[0121] The Gaussian process regression is used to build the photovoltaic, wind and user load power prediction model based on Bayes' theorem, and the photovoltaic, wind and user load day-ahead power data in the test set are used as the input vector and output vector of the prediction model:

[0122]

[0123] Among them, y * Represents the output vector of the prediction model, that is, the prediction data result; x * represents the input vector of the prediction model; K(x,x), K(x,x * )、K(x * ,x) and K(x * ,x * ) represents the covariance matrix; I represents the identity matrix;

[0124] The prediction result satisfying the Gaussian probability distribution is obtained from formula (2):

[0125]

[0126] Among them, μ * Represents the mean of the predicted data results; Indicates the variance of the prediction data result. Considering that the distributed energy output and user load in the actual DC microgrid cannot be negative, the confidence interval of the input vector in the prediction model is:

[0127] [L 1-α ,U 1-α ]=[max(0,μ * -z α / 2 σ * ),μ * +z α / 2 σ * ] (4)

[0128] Among them, U 1-α With L 1-α They represent the upper and lower limits of the confidence interval under the condition of confidence level 1-α; α represents the significance level; z α / 2 is the bilateral α quantile of the standard Gaussian distribution. Under different confidence levels, the multi-interval uncertainty set model of the present invention is as follows Figure 3 shown.

[0129] S102. In this embodiment, a multi-interval uncertainty set is constructed based on the actual fluctuation characteristics of uncertain factors, and the probability distribution information contained in the historical data of uncertain factors is mined to constrain the number of fluctuations and the fluctuation probability interval of the uncertain variables within a day. The multi-interval uncertainty set condition constraints are as follows:

[0130]

[0131] Among them, U φ represents the multi-interval probability uncertainty set of equipment φ, which includes wind turbines, photovoltaic power generation equipment and user load equipment; u φ,t represents the uncertain decision value of the device φ in the period t; N i Represents the number of intervals in a multi-interval uncertainty set; N t Indicates the total number of segments in time period t; represents the mean power of the equipment φ in the power system during period t; and They represent the upper and lower boundary values ​​of the confidence interval of device φ in time period i, respectively; and They respectively represent that the device φ is in the upper boundary fluctuation state and the lower boundary fluctuation state of the confidence interval i during period t; μ *φ,t Represents the mean value of the predicted data results of device φ in period t; 1-α i Indicates different confidence levels.

[0132] In the actual operation of the DC microgrid, the uncertain variables fluctuate only within a narrow range of predicted data values ​​in most periods, and only in a few periods will there be large fluctuations. The present invention uses the confidence expressed by the confidence interval as the probability of the fluctuation of the uncertain factors, constrains the number of occurrences of the uncertain factors by the probability of the fluctuation of the uncertain factors, and constrains the probability of the upward and downward fluctuations of the uncertain factors to be 50%, that is, the number of upward fluctuations and the number of downward fluctuations in the probability interval are the same. By accurately reflecting the fluctuation of uncertain variables in the actual operation scenario, the over-conservatism of the optimized scheduling is reduced, and the purpose of improving economic benefits is achieved.

[0133] In order to balance the rapidity of solution and the accuracy of results, the present invention adopts a three-interval division method to construct the following Figure 5 Schematic diagram of the multi-interval uncertain set model shown, where α1<α2<α3.

[0134] Step S2, constructing a DC microgrid operation model including distributed renewable energy.

[0135] The DC microgrid architecture described in the present invention is mainly composed of: photovoltaic power generation equipment, wind power generation equipment, controllable generator sets, energy storage systems and user loads. The operation mode of the DC microgrid containing distributed renewable energy in the present invention is: the DC microgrid converts solar energy and wind energy into electrical energy through components such as photovoltaic power generation equipment, wind power generation equipment, controllable generator sets, energy storage systems and bidirectional inverters, and optimizes the distribution and scheduling of energy through the energy management system to meet user load requirements. The DC power generated by photovoltaic and wind power generation equipment can be directly supplied to user loads, and the excess electrical energy can be used to charge the energy storage system. When the user load demand exceeds the power supply, additional power can be obtained from the external power grid through a bidirectional inverter. Through flexible energy management and optimization strategies, efficient utilization of renewable energy by DC microgrids is achieved.

[0136] According to the topological structure of the DC microgrid and the electrical connection relationship of each device, the present invention establishes a corresponding mathematical model for each component in the DC microgrid, as follows:

[0137] 1. DC microgrid power balance mathematical model

[0138] The DC microgrid transmits electric energy in the form of DC. Only active power flows during the transmission process. Therefore, only the active power balance constraint of the system needs to be considered during the optimization and scheduling process. The power balance constraint is as follows:

[0139]

[0140] in, and They represent the active power purchased and sold by the DC microgrid to the external AC grid during period t; P L (t) represents the active power consumed by the user load in the DC microgrid during period t; P PV (t) and P WT (t) represents the active power output of photovoltaic power generation and wind power generation in the DC microgrid during period t; P G (t) represents the active power generated by the controllable generator set in the DC microgrid during period t; and Represents the discharge and charge amount of the energy storage system in the DC microgrid during period t.

[0141] 2. Mathematical model of voltage drop and network loss

[0142] In the DC distribution network model, it is necessary to ensure that each node satisfies the active power flow balance shown in equation (1), and it is also necessary to consider the voltage loss and active power flow loss caused by the line impedance between different nodes, which are referred to as voltage drop and network loss respectively. The voltage drop and network loss constraints are as follows:

[0143]

[0144] Where: P g,i is the sum of the power flowing into node i; P d,i is the load power of node i; E + ={(i,j)|i<j;i,j∈N} and E - ={(i,j)|i>j; i,j∈N} is a set of directed paths, with the positive direction being i→j; E + is the set of branches flowing into node i; E - is the branch set out of node i; P ij is the power flowing between nodes i and j; V i With V j are the voltages of node i and node j respectively; R ij is the line impedance between nodes i and j; I ij is the current between nodes i and j; P loss represents the line loss between nodes i and j; F represents the vector of nodes; L represents the line between nodes.

[0145] 3. Mathematical model of controllable generator set

[0146] The DC grid system is equipped with a controllable generator set to supply flexibly adjustable electric energy to the DC grid system. Considering the ramp rate constraint of the controllable generator set, its output power constraint is as follows:

[0147]

[0148] in, and They represent the maximum output power and minimum output power of the generator set during period t, and their values ​​must take into account the limitations of the rated power and minimum load rate of the generator set; P G (t) is the output value of the controllable generator at time t; P G_CP represents the ramp power of the controllable generator. The power generation cost C of the generator set in period t G (t) are as follows:

[0149] C G (t) = aP G (t) (9)

[0150] Where a represents the fuel cost of the generator set.

[0151] 4. Mathematical model of energy storage system

[0152] In a DC microgrid, reasonable regulation of the energy storage system is conducive to improving the overall economic efficiency of the grid operation. However, in order to increase the service life of the energy storage system, the following constraints are imposed on its operation:

[0153]

[0154] in, Indicates the maximum charging and discharging power per unit time of the energy storage system; s SOC (t) represents the charging and discharging state of the energy storage system in the DC microgrid during period t. SOC When (t) is 1, the energy storage system in the DC microgrid is in a charging state; when s SOC When (t) is 0, the energy storage system in the DC microgrid is in a discharging state. and They represent the minimum remaining capacity and the maximum remaining capacity of the energy storage system respectively. This constraint can effectively prevent the energy storage system from overcharging and over-discharging, increase the service life of the energy storage system, and thus improve the overall economy of the system. SOC (0) represents the initial capacity of the energy storage system at time 0; η ch With η dis They represent the operating efficiency of the energy storage system in the charging state and the operating efficiency in the discharging state respectively.

[0155] 5. Mathematical model of energy interaction between DC microgrid and AC grid

[0156] The energy interaction between the microgrid and the external AC grid is completed by the bidirectional inverter, and its interaction power must meet the following requirements:

[0157]

[0158] in, It represents the maximum power of power interaction between the DC microgrid and the external AC grid. Its value depends on the maximum interactive power of the bidirectional grid-connected inverter at the connection node between the DC microgrid and the external AC grid. M (t) represents the energy interaction state between the DC microgrid and the external AC grid during period t. M When the value of (t) is 1, the bidirectional grid-connected inverter between the DC microgrid and the external AC grid operates in the power purchasing state; when s M When the value of (t) is 0, the bidirectional grid-connected inverter between the DC microgrid and the external AC grid operates in the power selling state. The energy interaction cost C between the DC microgrid and the external AC grid in period t M (t) is:

[0159]

[0160] Among them, K M (t) is the electricity price of the external AC grid during period t.

[0161] 6. Mathematical model of distributed renewable energy output

[0162] The renewable energy included in the DC power grid system includes photovoltaic power generation output and wind power generation output. Renewable energy itself is random and intermittent, and its output in different periods may not be a certain value. The present invention describes the specific output value constraint of renewable energy in the form of output power error:

[0163]

[0164] Where: and Respectively represent the installed capacity of photovoltaic power generation equipment and wind power generation equipment, that is, their power upper limits; P PV (t) and P WT (t) represents the actual output of photovoltaic power generation device and wind power generation device during period t respectively; and They represent the rated output of the photovoltaic power generation device and the wind power generation device in time period t respectively; and They respectively represent the output deviation of the photovoltaic power generation device and the wind power generation device in time period t.

[0165] Step S3, constructing a two-stage robust optimization scheduling model for DC microgrid.

[0166] First, the present invention considers the uncertain factors of photovoltaic output, wind power output and user load power, takes the operation economy of DC microgrid as the optimization goal, and designs a two-stage robust optimization scheduling model as follows:

[0167]

[0168] Where: b T δ represents the operating cost in robust optimization scheduling, b is the system constant matrix, ε and δ are the decision vectors of the first-stage scheduling and the second-stage scheduling respectively; U is a multi-probability interval uncertainty set, which contains the uncertainty description of photovoltaic output, wind power output and user load; u is an uncertain variable, including the output of photovoltaic power generation equipment, wind power generation equipment and user load in the DC power grid system; δ∈F(ε,u) represents the feasible domain of the decision vector δ when ε and u are known, and its expression is as follows:

[0169]

[0170] Among them, D, K, F, G and I u The coefficient matrices corresponding to the constraints of the robust one-stage and two-stage optimization of the DC microgrid are as follows; d, k and h are constant column vectors; the uncertain variable value u can be converted into a simplified model as shown in formula (15), and the feasible domain is searched until the worst scenario with the maximum operating cost of the DC microgrid is found.

[0171] Step S4, optimizing and solving the two-stage robust optimization scheduling model of the DC microgrid, and obtaining a robust optimization scheduling scheme of the DC microgrid that takes into account the uncertain fluctuation probability characteristics of the source and load.

[0172] The two-stage robust optimization scheduling model of the DC microgrid is a non-convex optimization problem, and it is difficult to directly obtain the optimal solution. The present invention uses a column constraint generation algorithm to decompose the two-stage robust optimization scheduling model of formula (14) into two solvable single-layer optimization problems. By alternately iterating the two decomposed problems, the optimal solution of the two-stage robust optimization scheduling model is finally obtained, that is, the working status and target power value of each device in the DC microgrid.

[0173] Specifically, the two-stage robust optimization scheduling model is decomposed using the C&CG algorithm (also known as the column constraint generation algorithm), so that the first stage of robust optimization is the outer min problem of the original model, and the second stage of robust optimization is the inner max-min problem of the model.

[0174] The first stage of the robust optimization problem makes decisions on the charging and discharging state of the energy storage system and the power purchase and sales state of the bidirectional grid-connected inverter in the DC microgrid. The first stage model of the robust optimization is shown in the formula:

[0175]

[0176] Where l represents the current iteration number of the column constraint generation algorithm; δ s represents the solution of the first stage optimization in the sth iteration; Indicates the value of the uncertain variable u obtained in the sth iteration.

[0177] The objective function of the first stage of robust optimization is In the formula, ε=[s M (t),s SOC (t)];

[0178] Constraints of the first stage of robust optimization: The constraint set of the first stage of robust optimization is the set of all time constraints of equations (6) to (13). The additional conditions for each iteration are In the first iteration, since the second stage has not yet been solved and the objective function is the economic optimization of the system energy scheduling, we first set b T δ≥0, that is, the cost is assumed to be greater than 0.

[0179] The second stage of robust optimization problem makes decisions on the output power of the generator set and the power purchase and sale value of the bidirectional grid-connected inverter according to the fluctuation of the output of distributed energy resources and the power fluctuation of user load in the DC microgrid. The second stage model of robust optimization is:

[0180]

[0181] The second stage of robust optimization is a max-min double-layer problem, which is still difficult to solve directly. The present invention converts the inner min problem in the double-layer model into a max problem, and merges the inner max problem and the outer max problem in the second stage into one max problem, that is, converts the double-layer problem into a single-layer optimization problem, as follows:

[0182] The objective function of the second stage of robust optimization is In the formula,

[0183] Constraints for the second stage of robust optimization:

[0184] The decision variables in the first stage of robust optimization are taken as constant matrices. The Lagrange multiplier matrix is ​​introduced into equations (6) to (13) respectively as equality constraints and inequality constraints. The inner layer problem transformed by KKT conditions is converted into equivalent constraints as follows:

[0185]

[0186] Where: L() is the Lagrangian function obtained by transformation; λ g is the Lagrange multiplier matrix of the inequality constraints; μ his the Lagrange multiplier matrix of the equality constraint; N g To robustly optimize the number of inequality constraints for the second-stage inner min problem; due to the KKT condition, λ will be increased g ≥0 and λ g T (Fε+Gδ-h)=0 as a supplementary constraint, and the nonlinear term λ in the supplementary constraint g T δ is difficult to solve directly, so the Big-M method is used to linearize the nonlinear term in equation (18) and obtain:

[0187]

[0188] In the formula, q g is the introduced 0-1 variable, i.e. q g ∈{0,1}; M is the Big-M coefficient. Theoretically, M is an infinite number, but in actual solution, it can be taken as a constant that is as large as possible.

[0189] The max-min double-layer problem in the second stage of robust optimization is converted into a single-layer max optimization problem that is easy to solve after KKT conditions and Big-M linearization, as shown in formula (20):

[0190]

[0191] S402, such as Figure 4 As shown in the figure, the confidence interval of the data under different confidence levels is first predicted by the typical data of uncertain variables in the power system, and the mean of the predicted data is used as the initial severe scenario, and the initial values ​​of the solution LB of the first stage of robust optimization and the solution UB of the second stage are set to be infinite. Then the first stage model is solved, and the solution results obtained in the first stage are passed to the second stage model to determine whether the solutions of the first stage model and the second stage model are less than the convergence value ξ. If they are less than the convergence value ξ, the final robust optimization result is output. Otherwise, the solution results obtained in the second stage are updated to the decision variables and constraints of the first stage problem, so that the two single-layer optimization problems are iterated alternately until the solutions of the first stage and the second stage converge quickly (this can improve the solution speed of the model), and then the optimal solution of the two-stage robust optimization scheduling model is obtained, that is, the working state and target power value of each device in the DC microgrid.

[0192] Furthermore, in step S401, the constraints of the second stage of robust optimization may be refined and supplemented as follows:

[0193] 1) Refine formula (18) to:

[0194]

[0195] Where: λ1,...,λ 13 is the Lagrange multiplier matrix of inequality constraints; μ1, μ2 are the Lagrange multiplier matrices of equality constraints; K SOC is the charging and discharging cost of the energy storage system; K M is the external AC grid electricity price;

[0196] 2) Refine formula (19) to:

[0197]

[0198] λ3≤Mq3,P G Δt-P G_CP ≥-M(1-q3) (24)

[0199]

[0200]

[0201] 3) Based on equations (21) to (34), the KKT conditional formula of the equality constraint and the multi-interval uncertainty set are supplemented to transform the second-stage max-min double-layer problem into a single-layer max optimization problem that is easy to solve. The supplemented KKT conditional formula of the equality constraint is as follows:

[0202]

[0203] In order to verify the superiority of the method of the present invention, the DC microgrid data of a certain enterprise in summer was taken as a practical application case, and a DC microgrid architecture model was built in MATLAB / Simulink.

[0204] Example 1

[0205] Based on the past data of user load, photovoltaic power generation and wind power generation, the sample data is predicted by GPR (Gaussian process regression). Taking into account the probability interval coverage and optimization economy, the confidence interval with a confidence level of 90% is selected as the worst scenario, and the simulation is performed with a confidence level of 20%. Through the GPR prediction algorithm, the prediction mean and confidence intervals with confidence levels of 90%, 70% and 50% can be obtained. The prediction intervals of user load, photovoltaic power generation and wind power generation at different confidence levels are as follows: Figure 5 , Figure 6 and Figure 7 shown.

[0206] Since the power generation output of photovoltaic equipment is greatly affected by day and night, its power generation is directly determined by the light intensity. The present invention divides the photovoltaic power generation equipment data into daytime period (5:00-20:00) and nighttime period (0:00-5:00 and 20:00-24:00) based on day and night. Figure 6When the DC microgrid is in the night time period, the output of the photovoltaic power generation equipment is regarded as a constant value of 0. In the process of photovoltaic power generation equipment output prediction, the GPR algorithm prediction is performed independently for the daytime period. Figure 5 , Figure 6 and Figure 7 It can be seen that as the confidence level of the interval increases, more and more historical data values ​​are included in the confidence interval, and the coverage of the confidence interval for the fluctuation of uncertain factors increases with the increase of confidence level.

[0207] based on Figure 5 , Figure 6 and Figure 7 The GPR prediction data is used to construct the multi-interval uncertainty set shown in formula (5), and the two-stage robust optimization scheduling model of the multi-interval uncertainty set is solved alternately and iteratively through formulas (14) and (17). Figure 8 , Fig. 9 and Fig.10 They are respectively the load forecast output, photovoltaic power generation equipment forecast output and wind turbine power generation equipment forecast output under two-stage robust optimization scheduling based on multi-interval uncertainty sets. Fig.11 and Fig.12 They are the output power of the controllable generator set and the interaction power between the energy storage charging and discharging power and the AC and DC grid.

[0208] Depend on Figure 8 , Fig. 9 and Fig.10 It can be obtained that the fluctuations of user load, photovoltaic output and wind power output appear 12 times, 6 times and 2 times on the boundaries of 50%, 70% and 90% confidence intervals respectively, the number of rising fluctuations is the same as the number of falling fluctuations, and the fluctuation of uncertain factors in the example results conforms to the constraints of the uncertainty set described in the present invention.

[0209] Fig.11 The value is positive when the energy storage system operates in discharge mode, and the value is negative when the energy storage system operates in discharge mode. Fig.12 The value is positive when the DC microgrid purchases electricity from the external AC grid, and it is negative when the DC microgrid sells electricity to the external AC grid. The two-stage robust optimization scheduling model based on multi-interval uncertainty sets is solved, and the operating cost of the DC microgrid is finally obtained to be 1399.1 yuan.

[0210] The energy required by the DC microgrid user load is supplied by the controllable generator sets and distributed energy generation in the DC microgrid, the power purchased by the AC power grid and the discharge of the energy storage system. When the system operates in severe conditions, the energy storage system and the controllable generator sets are regulated according to the optimization plan, which can effectively cope with the fluctuations of uncertain factors and ensure the best economic benefits of the system. When there is redundant energy in the DC microgrid, the energy storage system charges and sells it to the AC power grid for consumption, achieving the effect of peak shaving and valley filling. When the energy storage system has not reached the maximum storage capacity and the power purchase price is low, it will operate in charging mode, such as Fig.11 As shown in the 0:00-3:00 period and the 5:00-7:00 period; when the transaction electricity price is at a higher price and the energy storage system has the minimum storage capacity, the energy storage operates in the discharge mode, such as Fig.11 When the price of electricity purchased by the DC microgrid from the AC grid is greater than the fuel cost of the controllable generator set, the controllable generator set generates electricity in the maximum power mode, as shown in the figure. Fig.11 As shown in the period of 9:00 to 22:00; on the contrary, the controllable generator set maintains a relatively small output power, such as Fig.13 The time period from 0:00 to 8:00 is shown.

[0211] In order to verify the accuracy and economy of the two-stage robust optimization scheduling model based on multi-interval uncertainty sets involved in the present invention, the embodiment of the present invention is as follows: Figure 5 , Figure 6 and Figure 7 The historical data of user load, photovoltaic and wind power output are used as samples to construct three groups of uncertainty sets, namely, uncertainty sets based on single probability interval, uncertainty sets based on residual estimation, and uncertainty sets based on polynomial fitting estimation. The dispatch results of the three groups of uncertainty sets after robust optimization are analyzed in detail, and examples are compared from the two aspects of the economy of optimized dispatch and the accuracy of uncertainty set characterization.

[0212] Example 2

[0213] A single interval uncertainty set is a numerical interval consisting of the range of possible values ​​of an uncertain variable. Its interval boundaries usually use a wider range to cover various scenarios in which the uncertain variable may occur.

[0214] In order to verify the accuracy and economy of the two-stage robust optimization scheduling model based on multi-interval uncertainty sets, the present invention conducts comparative analysis with single probability interval confidence as a variable under the condition that the fluctuation of single probability interval uncertainty sets and multi-interval uncertainty sets are controlled to be the same. Fig.13 , Fig.14 and Fig.15They are the robust optimization results of user load power, photovoltaic power generation output and wind power generation output when the probability boundary confidence is 90%. Under the single probability interval uncertainty set scheduling strategy, the total system operating cost is 2138.7 yuan.

[0215] like Fig.13 , Fig.14 and Fig.15 As shown in the figure, each device in the DC microgrid is affected by the time-of-use electricity price, and the fluctuation of the uncertain factors in the robust optimization scheduling of the single probability interval uncertainty set with a confidence level of 90% mainly occurs in a concentrated form in a specific period of time. Fig.13 As shown in Figure 1, the electricity purchase price is at its peak during the time periods of 10:00-14:00 and 16:00-20:00, and the power of user loads fluctuates in a concentrated upward trend. Fig.14 As shown in Figure 1, the electricity price is relatively high during the time periods of 10:00-14:00 and 16:00-19:00, and the output of photovoltaic power generation equipment shows concentrated downward fluctuations at this time. Fig.15 As shown, the electricity price is relatively high during the periods of 10:00-14:00 and 16:00-23:00, and wind power generation equipment experiences concentrated downward fluctuations during this period.

[0216] In order to eliminate the influence on the effectiveness of the comparative examples caused by different probability interval divisions of the comparative examples, under the same condition of uncertain fluctuation constraints, the present invention adds simulation calculations for single probability interval uncertainty sets with confidence levels of 70% and 50%, and obtains the total operating cost of the DC microgrid after robust optimization scheduling under single probability interval uncertainty sets at different confidence levels as shown in Table 1.

[0217] Table 1 Robust optimization running costs for different single probability interval uncertainty sets

[0218] Uncertain Set Operating cost (yuan) Multiple interval uncertain sets 1399.1 90% probability interval uncertainty set 2138.7 70% probability interval uncertainty set 1969.4 50% probability interval uncertainty set 1747.5

[0219] Compared with the two-stage robust optimization scheduling results of multi-interval uncertainty sets, the optimized scheduling results obtained by the robust optimization based on the single probability interval uncertainty set are relatively brief in describing the fluctuation of the uncertainty set. The fluctuation of uncertain factors is too concentrated and the fluctuation amplitude is large, which makes it difficult to accurately describe the fluctuation of real uncertain factors, resulting in overly conservative optimization results and poor overall economic benefits of the system. As shown in Table 1, the operating cost of the robust optimization scheduling model based on the single probability interval uncertainty set with confidence levels of 90%, 70% and 50% decreases as the confidence level decreases, but it is still 52.8%, 40.8% and 24.9% higher than the robust optimization model based on the multi-interval uncertainty set, respectively, reflecting the improvement of the economic efficiency of the DC microgrid by the multi-interval uncertainty set.

[0220] Example 3

[0221] In order to verify the superiority of the method involved in the present invention, this embodiment 3 introduces a multi-interval uncertainty set construction method based on residual estimation for comparison with the method involved in the present invention. The multi-interval uncertainty set describes the uncertain factors with multiple uncertainty intervals, which can accurately and comprehensively describe the uncertain information. However, in the existing multi-interval uncertainty set research, the ratio of the maximum deviation between the predicted mean μ of the regression equation and the actual observed value of the power system equipment is often used as the boundary of the multi-interval uncertainty set, which is referred to as the residual estimation method. However, the uncertainty set constructed based on the residual estimation method fails to reflect the connection between the fluctuation boundary of the uncertainty factor and the number of fluctuations. Therefore, the multi-interval uncertainty set based on residual estimation will inevitably be restricted by subjective factors in the setting of parameters, and it is difficult to accurately quantify the true fluctuation of the uncertain variables. Ultimately, the deviation between the multi-interval uncertainty set based on residual estimation and the actual fluctuation of the uncertain variables will directly lead to differences between the dispatching results of the DC microgrid and the actual operating characteristics of the DC microgrid.

[0222] The multi-interval uncertainty set comparison model based on residual estimation set in the comparative example has the same number of fluctuations as the multi-interval uncertainty set described in the present invention, and the multi-interval uncertainty boundaries used for the comparison are 90% of the residual value, 70% of the residual value, and 50% of the residual value.

[0223] By solving the robust optimization model based on residual estimation multi-interval uncertainty set, the results of user load, wind power output and photovoltaic output fluctuation dispatch in DC microgrid are obtained as follows: Fig.16 , Fig.17 and Fig.18 As shown in the figure, the total output and total consumption of each device in the DC microgrid are as follows: Fig.19 and Fig. 20 As shown. The multi-interval uncertainty based on residual estimation does not consider the historical fluctuations of uncertain factors in the interval boundary setting, and uses the residuals between historical data and predicted data to estimate the fluctuations of uncertain variables. Although the fluctuation interval obtained by the multi-interval uncertainty set based on residual estimation at the moment when the residual is large is slightly narrower than the fluctuation interval obtained by the probability interval prediction, the fluctuation interval is still divided with the maximum deviation as a reference at other moments, resulting in a wider overall width of uncertainty, which is inconsistent with the actual operation scenario. Under the multi-interval uncertainty set scheduling strategy based on residual estimation, the total operating cost of the DC microgrid is 1,484.7 yuan, which is 6.1% higher than the method described in the present invention. The locally too wide interval range not only fails to improve the robustness of the DC microgrid, but will lead to a decrease in the economy of robust optimization scheduling.

[0224] Example 4

[0225] Polynomial fitting is a parameterized regression method that assumes that the relationship between the independent variable and the dependent variable can be described by a polynomial function. This embodiment 4 is based on the polynomial fitting method as a comparative example to intuitively quantify the superiority of the method involved in the present invention. The polynomial fitting method is restricted by the fitting polynomial parameters. When the fitting order is set to a low level, it may not be able to fit the data fluctuation state well; when the fitting order is set to a high level, overfitting is prone to occur, resulting in a decrease in the generalization ability of the model. This embodiment 4 takes the fourth-order polynomial as an example to carry out a comparative analysis of the robust optimization scheduling of DC microgrids.

[0226] By solving the robust optimization model of multi-interval uncertainty sets based on polynomial fitting, the results of user load, wind power output and photovoltaic output fluctuation dispatch in DC microgrid are obtained as follows: Fig.21 , Fig. 22 and Fig.23 As shown in the figure, the total output and total consumption of each device in the DC microgrid are as follows: Fig.24 and Fig.25 shown.

[0227] When faced with nonlinear and complex data patterns such as historical data of power system operation, the ability of polynomial fitting is limited. The multi-interval uncertainty set divided by it usually cannot provide direct uncertainty estimation. The polynomial fitting method tends to give a definite fitting curve rather than an overall probability distribution. The limitations of the polynomial function form may lead to the inability to accurately capture the complex relationships in the data, resulting in inaccurate fitting results. Under the multi-interval uncertainty set scheduling strategy based on polynomial fitting, the total operating cost of the DC microgrid is 1,479.5 yuan, which is 5.7% higher than the method involved in the present invention.

[0228] The present invention introduces interval coverage (Coverage Probability, CP) and average interval width (AverageInterval Width, AIW) as data indicators for quantifying the accuracy of interval division, in order to analyze the accuracy differences of various multi-interval uncertainty set characterization methods. Interval coverage and average interval width are important indicators for evaluating the accuracy and reliability of confidence intervals. Interval coverage measures the probability of whether an interval can contain true parameters or unknown observations in multiple repeated sampling or model fitting. The average interval width is the ratio of the average value of the confidence interval width to the predicted value, which is used to describe the degree of focus of the interval on uncertain factors. Under the same confidence level, the smaller the average interval width value, the better its ability to focus on data. The comparative data of interval coverage and average interval width obtained by the method involved in the present invention and other fitting methods are shown in Table 2 and Table 3 respectively.

[0229] Table 2 Interval coverage of various types of uncertain sets

[0230]

[0231]

[0232] Table 3 Average width of intervals of various uncertainty sets

[0233]

[0234] Analysis of Tables 2 and 3 shows the common characteristics of various uncertainty intervals. The interval coverage and average width of various uncertainty intervals are positively correlated with the confidence level, which is consistent with the changing relationship between the confidence level and the confidence interval. From the interval coverage of various uncertainty intervals shown in Table 2, it can be seen that the CP value error of the residual estimation uncertainty interval is the largest, the CP value error obtained by the GPR multi-probability interval is second, and the CP value error of the polynomial fitting uncertainty interval is the lowest. From the average width of the intervals of various uncertainty intervals shown in Table 3, it can be seen that the AIW value of the polynomial fitting uncertainty interval is much larger than the GPR probability uncertainty interval and the residual estimation uncertainty interval. However, the residual estimation interval with the largest CP error has a lower AIW value. Comprehensively analyzing the CP and AIW indicators of various uncertainty interval division methods, the residual estimation interval and the polynomial fitting interval have large defects in the CP error and AIW value respectively. The comprehensive comparison of various multi-interval uncertainty sets is shown in Table 4.

[0235] Table 4 Comprehensive comparison of various types of multi-interval uncertainty sets

[0236]

[0237] As shown in Table 4, compared with the GPR multi-probability interval uncertainty partitioning method described in the present invention, the residual estimation interval method and the polynomial fitting interval method are unable to accurately characterize the fluctuation of various uncertain variables in the DC microgrid when facing the DC microgrid optimization scheduling scenario, resulting in a decrease in the economy of the system operation.

[0238] By comparing the robust optimization scheduling results of various uncertainty sets, it can be seen that under the condition that the total amount of uncertainty remains the same, the multi-interval uncertainty set described in the present invention can reduce the number of extreme situations at the interval boundary by more accurately describing the probability fluctuation distribution of the uncertainty factors in the operation and scheduling of the power system, and finally reduce the over-conservatism of each uncertain variable in the robust optimization scheduling of the DC microgrid. And under the premise of meeting the system robustness requirements, it can improve the operation economy of the DC microgrid.

[0239] The multi-interval uncertainty set described in the present invention not only deeply mines the probability distribution information contained in the historical data of uncertain factors, but also integrates the obtained probability intervals to construct a multi-probability interval uncertainty set that integrates interval and probability information. Compared with the existing description method of uncertainty sets of multi-interval boundaries, the present invention takes into account the fluctuation correlation of the divided intervals, combines the relationship between the source load fluctuation range and the fluctuation frequency, and can more accurately describe the fluctuation law of the source load.

[0240] The present invention associates multi-interval uncertainty sets with a two-stage robust optimization scheduling model for DC microgrids, improving the existing robust optimization scheduling method for DC microgrids. Compared with the robust optimization scheduling schemes based on single probability interval uncertainty sets and other types of multi-interval uncertainty sets, the method involved in the present invention achieves a higher interval coverage rate with a lower interval width by accurately characterizing the uncertain variables, so that the fluctuation range of distributed energy and user load in the DC microgrid is more focused, thereby improving the utilization rate of distributed energy, reducing load redundancy and improving the economy of the scheduling scheme.

[0241] The present invention achieves conservative adjustment of the robust optimization model by changing the confidence of the uncertainty set or the number of intervals of the uncertainty set. For example, when some non-sensitive loads do not have high requirements on the stability of the system energy, the fluctuation range of the uncertain variables can be reduced to achieve the purpose of improving the economic benefits of operation. On the contrary, for sensitive loads, the stability of the system is higher, the interval range can be divided wider, and more energy margin can be left to ensure stability. Therefore, the present invention can flexibly characterize the fluctuation boundary and fluctuation frequency of uncertain variables for different scenarios, and formulate flexible operation and control strategies.

[0242] The above embodiments are preferred implementation schemes of the present invention. In addition, the present invention may also be implemented in other ways. Any obvious replacement without departing from the concept of the present technical solution is within the protection scope of the present invention.

[0243] In order to make it easier for ordinary technicians in the field to understand the improvements of the present invention over the prior art, some drawings and descriptions of the present invention have been simplified, and for the sake of clarity, some other elements are omitted in this application document. Ordinary technicians in the field should realize that these omitted elements may also constitute the content of the present invention.

Claims

1. A two-stage robust optimization scheduling method for uncertain sets of source and load power with multiple intervals, characterized by: include: Step S1, constructing a multi-interval uncertainty set considering the uncertain fluctuation probability characteristics of source and load based on the historical data of the DC microgrid; Step S2, constructing a DC microgrid operation model including distributed renewable energy; Step S3, considering the uncertain factors of distributed renewable energy and user load power, taking the operation economy of the DC microgrid as the optimization goal, introducing the multi-interval uncertainty set obtained in step S1 into the DC microgrid operation model constructed in step S2, and constructing a two-stage robust optimization scheduling model for the DC microgrid; Step S4, performing optimization scheduling solution on the DC microgrid two-stage robust optimization scheduling model obtained in step S3, and obtaining a DC microgrid robust optimization scheduling scheme that takes into account the uncertain fluctuation probability characteristics of sources and loads.

2. The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets according to claim 1 is characterized by: Step S1 includes: S101, collecting the photovoltaic, wind power and user load power data of the previous day to form a training set and a test set; Construct a standard Gaussian process model, using the photovoltaic, wind power and user load power data in the training set as the input vector and output vector of the standard Gaussian process model: y=f(x)+ε (1) Where x=[x1,x2,...,x n ] T represents the input vector of the standard Gaussian process model, y=[y1,y2,...,y n ] T represents the output vector of the standard Gaussian process model, ε represents the amount of noise that independently satisfies the normal distribution, and its mean is 0 and its variance is σ 2 , denoted as ε~N(0,σ 2 ); The Gaussian process regression is used to build the photovoltaic, wind and user load power prediction model based on Bayes' theorem, and the photovoltaic, wind and user load day-ahead power data in the test set are used as the input vector and output vector of the prediction model: Among them, y * Represents the output vector of the prediction model, that is, the prediction data result; x * represents the input vector of the prediction model; K(x,x), K(x,x * )、K(x * ,x) and K(x * ,x * ) represents the covariance matrix; I represents the identity matrix; The prediction results satisfying the Gaussian probability distribution are obtained from the above prediction model: Among them, μ * Represents the mean of the predicted data results; Indicates the variance of the predicted data results; Based on the fact that the distributed energy output and user load in the actual DC microgrid are not negative, the confidence interval of the input vector in the prediction model is determined as: [L 1-α ,U 1-α ]=[max(0,μ * -z α / 2 s * ),m * +z α / 2 s * ] (4) Among them, U 1-α With L 1-α They represent the upper and lower limits of the confidence interval under the condition of confidence level 1-α; α represents the significance level; z α / 2 is the bilateral α quantile of the standard Gaussian distribution; S102. Based on the actual fluctuation characteristics of uncertain factors, a multi-interval uncertainty set is constructed. Then, based on the probability distribution information contained in the historical data of uncertain factors, the number of fluctuations of uncertain variables within a day and the fluctuation probability interval are constrained as follows: Among them, U φ represents the multi-interval probability uncertainty set of equipment φ, which includes wind turbines, photovoltaic power generation equipment and user load equipment; u φ,t represents the uncertain decision value of the device φ in the period t; N i Represents the number of intervals in a multi-interval uncertainty set; N t Indicates the total number of segments in time period t; represents the mean power of the equipment φ in the power system during period t; and They represent the upper and lower boundary values ​​of the confidence interval of device φ in time period i, respectively; and They respectively represent that the device φ is in the upper boundary fluctuation state and the lower boundary fluctuation state of the confidence interval i during period t; μ *φ,t Represents the mean value of the predicted data results of device φ in period t; 1-α i Indicates different confidence levels.

3. The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets according to claim 2 is characterized by: In step S2, when constructing the DC microgrid operation model, a corresponding mathematical model is established for each component in the DC microgrid according to the topological structure of the DC microgrid and the electrical connection relationship of each device, including: 1) DC microgrid power balance mathematical model, its power balance constraint formula is as follows: in, and They represent the active power purchased and sold by the DC microgrid to the external AC grid during period t; P L (t) represents the active power consumed by the user load in the DC microgrid during period t; P PV (t) and P WT (t) represents the active power output of photovoltaic power generation and wind power generation in the DC microgrid during period t; P G (t) represents the active power generated by the controllable generator set in the DC microgrid during period t; and represents the discharge and charge of the energy storage system in the DC microgrid during period t; 2) Voltage drop and network loss mathematical model, with the following constraints: Where: P g,i is the sum of the power flowing into node i; P d,i is the load power of node i; E + ={(i,j)|i<j;i,j∈N} and E - ={(i,j)|i>j; i,j∈N} is a set of directed paths, with the positive direction being i→j; E + is the set of branches flowing into node i; E - is the branch set out of node i; P ij is the power flowing between nodes i and j; V i With V j are the voltages of node i and node j respectively; R ij is the line impedance between nodes i and j; I ij is the current between nodes i and j; P loss represents the line loss between nodes i and j; F represents the vector of nodes; L represents the line between nodes; 3) Mathematical model of controllable generator set, and its output power constraint: in, and They represent the maximum output power and minimum output power of the generator set during period t, and their values ​​must take into account the limitations of the rated power and minimum load rate of the generator set; P G (t) is the output value of the controllable generator at time t; P G_CP represents the ramp power of the controllable generator; the power generation cost C of the generator set in period t G (t) are as follows: C G (t)=aP G (t) (9) Where: a represents the fuel cost of the generator set; 4) Mathematical model of energy storage system, which operates with the following constraints: in, Indicates the maximum charging and discharging power per unit time of the energy storage system; s SOC (t) represents the charging and discharging state of the energy storage system in the DC microgrid during period t. SOC When (t) is 1, the energy storage system in the DC microgrid is in a charging state. SOC When (t) is 0, the energy storage system in the DC microgrid is in a discharging state; and Respectively represent the minimum remaining capacity and the maximum remaining capacity of the energy storage system; E SOC (0) represents the initial capacity of the energy storage system at time 0; η ch With η dis They represent the operating efficiency of the energy storage system in the charging state and the operating efficiency in the discharging state respectively; 5) Mathematical model of energy interaction between DC microgrid and AC grid. The interaction power of this model must satisfy: in, It represents the maximum power of power interaction between the DC microgrid and the external AC grid. Its value depends on the maximum interactive power of the bidirectional grid-connected inverter at the connection node between the DC microgrid and the external AC grid. M (t) represents the energy interaction state between the DC microgrid and the external AC grid during period t. M When the value of (t) is 1, the bidirectional grid-connected inverter between the DC microgrid and the external AC grid operates in the power purchasing state. M When the value of (t) is 0, the bidirectional grid-connected inverter between the DC microgrid and the external AC grid operates in the power selling state; the energy interaction cost C between the DC microgrid and the external AC grid in period t M (t) is: Among them, K M (t) is the electricity price of the external AC grid during period t; 6) The mathematical model of distributed renewable energy output, whose output numerical constraints are: Where: and Respectively represent the installed capacity of photovoltaic power generation equipment and wind power generation equipment, that is, their power upper limits; P PV (t) and P WT (t) represents the actual output of photovoltaic power generation device and wind power generation device during period t respectively; and They represent the rated output of the photovoltaic power generation device and the wind power generation device in time period t respectively; and They respectively represent the output deviation of the photovoltaic power generation device and the wind power generation device in time period t.

4. The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets according to claim 3 is characterized by: In step S3, the uncertain factors of photovoltaic power output, wind power output and user load power are considered, and the operation economy of the DC microgrid is taken as the optimization goal. The multi-interval uncertainty set obtained in step S1 is introduced into the DC microgrid operation model constructed in step S2 to construct a DC microgrid two-stage robust optimization scheduling model as follows: Where: b T δ represents the operating cost in robust optimization scheduling, b is the system constant matrix, ε and δ are the decision vectors of the first-stage scheduling and the second-stage scheduling respectively; U is a multi-probability interval uncertainty set, which contains the uncertainty description of photovoltaic output, wind power output and user load; u is an uncertain variable, including the output of photovoltaic power generation equipment, wind power generation equipment and user load in the DC power grid system; δ∈F(ε,u) represents the feasible domain of the decision vector δ when ε and u are known, and its expression is as follows: Among them, D, K, F, G and I u They correspond to the coefficient matrices of the constraint variables of the robust one-stage and two-stage optimization of the DC microgrid respectively; d, k and h are constant column vectors.

5. The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets according to claim 4 is characterized by: In step S4, when optimizing and solving the two-stage robust optimization scheduling model of the DC microgrid obtained in step S3, the column constraint generation algorithm is used to decompose the two-stage robust optimization scheduling model into two solvable single-layer optimization problems. By alternately iterating the two decomposed problems, the optimal solution of the two-stage robust optimization scheduling model is finally obtained, that is, the working status and target power value of each device in the DC microgrid.

6. The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets according to claim 5 is characterized by: Step S4 includes: S401, the two-stage robust optimization scheduling model is decomposed by using the column constraint generation algorithm, so that the first stage of robust optimization is the outer min problem of the original model, and the second stage of robust optimization is the inner max-min problem of the model; The first stage problem of robust optimization makes decisions on the charging and discharging state of the energy storage system and the power purchase and sale state of the bidirectional grid-connected inverter in the DC microgrid. The first stage model of robust optimization is shown in the following formula: Where l represents the current iteration number of the column constraint generation algorithm; δ s represents the solution of the first stage optimization in the sth iteration; Indicates the value of the uncertain variable u obtained in the sth iteration; The objective function of the first stage of robust optimization is In the formula, ε=[s M (t),s SOC (t)]; Constraints of the first stage of robust optimization: The constraint set of the first stage of robust optimization is the set of all time constraints of equations (6) to (13). The additional conditions for each iteration are In the first iteration, let b T δ≥0, that is, the cost is assumed to be greater than 0; The second stage problem of robust optimization is to make decisions on the output power of the generator set and the power purchase and sale value of the bidirectional grid-connected inverter according to the fluctuation of the output of distributed energy resources and the fluctuation of the user load power in the DC microgrid. The second stage model of robust optimization is: The objective function of the second stage of robust optimization is In the formula, Constraints for the second stage of robust optimization: The decision variables in the first stage of robust optimization are taken as constant matrices. The Lagrange multiplier matrix is ​​introduced into equations (6) to (13) respectively as equality constraints and inequality constraints. The inner layer problem transformed by KKT conditions is converted into equivalent constraints as follows: Where: L() is the Lagrangian function obtained by transformation; λ g is the Lagrange multiplier matrix of the inequality constraints; μ h is the Lagrange multiplier matrix of the equality constraint; N g The number of inequality constraints for the second-stage inner min problem for robust optimization; The Big-M method is used to linearize the nonlinear term in equation (18), and we get: In the formula, q g is the introduced 0-1 variable, i.e. q g ∈{0,1}; M is the Big-M coefficient; The max-min double-layer problem in the second stage of robust optimization is converted into a single-layer max optimization problem after KKT conditions and Big-M linearization, as shown in formula (20): S402, the result of the second-stage problem decomposed in S401 is updated to the decision variables and constraints of the first-stage problem, so that the two single-layer optimization problems are iterated alternately, the solution of the first-stage problem and the solution of the second stage are converged, and the optimal solution of the two-stage robust optimization scheduling model is obtained, that is, the working status and target power value of each device in the DC microgrid.

7. The two-stage robust optimization scheduling method for source-load power multi-interval uncertain sets according to claim 6 is characterized by: In step S401, the second stage constraints of robust optimization are refined and supplemented as follows: 1) Refine formula (18) to: Where: λ1,...,λ 13 is the Lagrange multiplier matrix of inequality constraints; μ1, μ2 are the Lagrange multiplier matrices of equality constraints; K SOC is the charging and discharging cost of the energy storage system; K M is the external AC grid electricity price; 2) Refine formula (19) to: 3) The KKT conditional expression and multi-interval uncertainty set of the supplementary equality constraint are as follows:

Citation Information

Patent Citations

  • Microgrid two-stage adaptive robust optimization scheduling method based on interval probability uncertainty set

    CN115688970A

  • Two-stage robust capacity optimization configuration method considering flexibility demand constraint

    CN117937428A

  • Micro-grid two-stage robust optimization scheduling method based on data-driven uncertainty set

    CN118336702A

Cited By

  • Economic dispatching optimization method and system for micro-grid of charging and battery swapping station

    CN118966720A

  • Robust optimal configuration method and device for low-carbon integrated energy system based on source load uncertainty

    CN119228584A

  • Low-carbon optimal scheduling method and system for commercial park microgrid

    CN121216407A

  • Multi-period alternating current power flow boundary depicting method adaptive to real-time electric energy market

    CN121602400A