Power distribution network robust optimization method based on Gaussian mixture model accurate uncertain boundary

Through Gaussian hybrid model and improved affine decision-making rules, an adaptive and robust optimization model for the distribution network is constructed, which solves the problem of insufficient optimization in the face of photovoltaic uncertainty and achieves more accurate uncertainty Description and lower scheduling costs reduce the conservatism of optimization results.

CN119070401APending Publication Date: 2024-12-03HARBIN ENG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411129388.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

When traditional distribution networks face the uncertainty of new energy such as photovoltaics, it is difficult to achieve safe and economic operation, and the optimization methods are insufficient, resulting in high conservatism of robust optimization and large modeling errors.

Method used

Gaussian hybrid model is used to cluster photovoltaic historical data, build uncertain ensembles, and combine improved affine decision rules to establish a multi-stage adaptive robust optimization scheduling model for active distribution networks to reduce the conservatism of robust optimization.

Benefits of technology

By accurately describing the uncertainty of photovoltaic output, reducing modeling errors, reducing the conservatism of optimization results, improving the lowness of the previous total scheduling cost, enhancing the feasibility of decision variables, and obtaining more optimized feasible solutions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119070401A_ABST
    Figure CN119070401A_ABST
Patent Text Reader

Abstract

The invention relates to a power distribution network robust optimization method based on an accurate uncertain boundary of a Gaussian mixture model, and belongs to the technical field of power distribution network operation optimization. Clustering photovoltaic historical data through a Gaussian mixture model to obtain distribution conditions of photovoltaic output under two illumination conditions; constructing a plurality of uncertain sets through parameters generated by the Gaussian mixture model to describe the uncertainty of photovoltaic output; establishing an active power distribution network comprehensive cost objective function and power distribution network constraint conditions; constructing a multi-stage adaptive robust optimization scheduling model of the active power distribution network, and solving adaptive robust optimization by using an improved affine decision rule; the result shows that under the same affine rule, compared with a classic interval set and a polyhedron set, the GMM interval set and the GMM polyhedron set are used for describing the photovoltaic uncertainty, the photovoltaic output uncertain boundary description can be more accurate, and therefore the conservative property of the optimization result is small; and under the same polyhedron set, the improved affine rule solving result is superior to that of the traditional affine rule.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a distribution network robust optimization method based on the precise uncertain boundary of a Gaussian mixture model, and belongs to the technical field of distribution network operation optimization. Background Art

[0002] With the large-scale development of new energy sources such as photovoltaics, their randomness makes the operation of power grids face major risks. As an important carrier for distributing electric energy, the traditional distribution network is gradually upgraded to an active distribution network with interactive supply and demand, making the role of the distribution network more flexible and extensive. How to give full play to the potential of the distribution network, overcome the uncertainty of new energy, and ensure the safe and economical operation of the distribution network has become a problem that needs to be solved urgently. In view of the shortcomings of the distribution network optimization problem, the present invention proposes a day-ahead optimization method for active distribution networks considering photovoltaic uncertainty. First, the photovoltaic historical data is clustered by Gaussian Mixture Model (GMM) to obtain the distribution of photovoltaic output under the conditions of sufficient and insufficient light. The parameters generated by the Gaussian mixture model are used to construct an uncertain set to describe the uncertainty of photovoltaic output. The information contained in the photovoltaic output historical data is brought into the day-ahead optimization scheduling stage to reduce the modeling error of the robust optimization uncertainty set. Finally, the improved affine decision rule is used to solve the adaptive robust optimization model, which reduces the conservatism of robust optimization. Summary of the invention

[0003] The present invention proposes a robust optimization method for distribution network based on Gaussian mixture model with precise uncertainty boundary to achieve the optimization of distribution network's day-ahead active power. Considering the uncertainty of photovoltaic output, the Gaussian mixture model is used to characterize the photovoltaic output range, and the improved affine rule is used to solve the uncertainty optimization model, thereby reducing the modeling error and the conservatism of the model.

[0004] The object of the present invention is achieved in that:

[0005] A robust optimization method for distribution network based on Gaussian mixture model with accurate uncertainty boundary is proposed. The specific steps are as follows:

[0006] 1. The historical photovoltaic data are clustered using a Gaussian mixture model to obtain the distribution of photovoltaic output under two lighting conditions.

[0007] 2. Multiple uncertainty sets are constructed through the parameters generated by the Gaussian mixture model to describe the uncertainty of photovoltaic output.

[0008] 3. Establish the comprehensive cost objective function of the active distribution network and the distribution network constraints.

[0009] 4. Construct a multi-stage adaptive robust optimization dispatching model for active distribution networks, which is divided into two stages: pre-dispatching and re-dispatching. In the pre-dispatching stage, the electric load, generator on / off state variables, and energy storage charging and discharging state variables are determined before the uncertainty is observed; and in the re-dispatching stage, the amount of electricity purchased and sold from the superior power grid, the amount of demand response, the amount of generator power generation, and the amount of energy storage charging and discharging are determined after the uncertainty is observed, and their existence and feasibility have been guaranteed by pre-dispatching.

[0010] 5. Use improved affine decision rule to solve adaptive robust optimization.

[0011] The present invention also includes:

[0012] 1. The process of establishing the Gaussian mixture model in step 1 is as follows:

[0013] The Gaussian mixture model is not limited to a specific probability density function form assumption, and any probability density distribution can be approximated by a linear combination of several Gaussian density functions, which can accurately characterize the probability density distribution of random variables. The Gaussian mixture probability density function is expressed as follows:

[0014]

[0015] The Gaussian mixture model uses the expectation-maximization (EM) algorithm to fit n mixed sub-Gaussian distributions to obtain the mean and standard deviation of each distribution. The specific steps of the EM algorithm are as follows.

[0016] 1) Randomly initialize model parameters ω k , μ k , σ k The initial value of .

[0017] 2) Step E: Calculate the conditional probability expectation of the joint distribution

[0018]

[0019] Where: x = (x (1) ,x (2) ,...x (m) ) is the observed data; p(x (i) ,ω (i) |θ) is the joint distribution;

[0020] p(ω (i) |x (i) ,θ) is the conditional distribution;

[0021] 3) Step M: Maximize L(θ) to get θ j+1 :

[0022]

[0023] 4) If θ j+1 If convergence has been achieved, the algorithm ends. Otherwise, the E and M steps are continued for iteration.

[0024] At this point, the parameters of each sub-Gaussian distribution model of the Gaussian mixture model can be obtained using the EM algorithm.

[0025] 2. In step 2, multiple uncertain sets are constructed, including:

[0026] The uncertainty set is created using the Laida criterion, that is, the probability that the value is distributed in (μ-3σ,μ+3σ) is 0.9973, so this interval can be regarded as the actual possible interval of the uncertainty quantity.

[0027] Considering that the historical values ​​of photovoltaic output at different times can be considered to obey a multi-peak normal population, the important parameters (mean and standard deviation) of photovoltaic output in each period within 24 hours are obtained by mining the historical photovoltaic output data using a Gaussian mixture model and integrating them into the classic uncertainty set. The actual photovoltaic output, mean, standard deviation, and offset are linked through the Laida criterion. For this purpose, the following interval and polyhedron uncertainty sets based on Gaussian mixture model parameters are established. As shown in equations (4) and (5):

[0028]

[0029] Where: U is the photovoltaic output uncertainty set; P t PV is the photovoltaic output value at time t; are the mean and standard deviation of photovoltaic output at time t; ΔP t PV is the output fluctuation of photovoltaic at time t; δ t is the fluctuation coefficient of photovoltaic output at time t. From formula (5), we can know that δ t ∈[-1,1]; Γ is the conservatism, but in reality it is unlikely that all predictions will reach the boundary at the same time. This is determined by the central limit theorem and has nothing to do with which probability distribution the individual follows. The conservatism of the present invention is defined as the sum of the absolute values ​​of the ratios of the actual photovoltaic output fluctuation to the maximum fluctuation upper limit within the scheduling period. The larger the Γ value, the more conservative the model; the smaller the Γ value, the more radical the model.

[0030] 3. The comprehensive cost objective function and constraints of the active distribution network in step 3 include:

[0031] The goal of the day-ahead optimization operation is to minimize the sum of the electricity purchase cost, demand response cost, diesel generator and gas turbine power generation cost, as shown below:

[0032]

[0033] Where: T and N are the number of dispatch periods and the number of generators respectively. They represent the electricity purchase cost, demand response cost and generator cost during the dispatch period respectively; λ fixed , variable , on They are the unit price of electricity purchased by the distribution network from the upper power grid at time t, the demand response cost coefficient, the generator fixed cost coefficient, the generator variable cost coefficient, and the generator startup cost coefficient. t purch , P t tran , P t cut 、u tn , g tn 、v tn They represent the amount of electricity purchased, the amount of load that can be transferred, the amount of load that can be reduced, the on / off status of the generator, the amount of power generated by the generator, and the generator on / off flag at time t in the dispatching cycle. The binary variable of the on / off status, when the value is 1, it means the generator is on; 0 means it is off. When the on / off flag is 1, it means that the generator is on at that moment.

[0034] The constraints include:

[0035] (1) System power balance constraints

[0036]

[0037] Where: P t ch and P t dis Respectively represent the charging and discharging power of energy storage at time t; P t purch and P t sell They represent the power purchased and sold by the distribution network to the upper grid at time t; g t P represents the sum of the output power of the generators in the distribution network at time t; t pv , P t load They represent the photovoltaic output power and load power consumption at time t respectively.

[0038] (2) Generator operation constraints

[0039]

[0040] Where: w tn It is the generator shutdown flag. When the shutdown flag is 1, it means that the generator is shut down at that moment. The minimum capacity of the nth generator set, The maximum capacity of the nth generator set, The minimum startup time of the nth unit, The minimum downtime of the nth generator set, The ramp rate limit of the nth generator set, The ramp rate limit for the nth generator set.

[0041] (3) Energy storage operation constraints

[0042]

[0043] Where: are 0-1 variables, representing the energy storage charging and discharging states respectively; P ch , Indicates the upper and lower limits of energy storage charging power; P dis , Indicates the upper and lower limits of energy storage discharge power; Soc t Indicates the energy storage charge state; η ch , η dis Indicates the energy storage charging and discharging efficiency coefficient; E e is the rated capacity of energy storage; Soc , Indicates the upper and lower limits of the state of charge. To ensure that there is still room for energy storage at the beginning of the next dispatch cycle, the state of charge should be within the range of 0.4-0.6 after the end of the previous dispatch cycle.

[0044] (4) Demand response resource constraints

[0045]

[0046] Where: They are the upper limits of load that can be reduced and load that can be transferred in each period; is the total amount of transferable load demand; The total upper limit of the load that can be reduced.

[0047] 4. The multi-stage adaptive robust optimization dispatching model of active distribution network constructed in step 4 includes:

[0048] The objective function (6) is transformed into a multi-stage adaptive robust optimization model as follows:

[0049]

[0050] Where: a T x is the sum of the startup cost and fixed cost of the generator set in the pre-dispatch stage; b Ty is the sum of the power purchase cost, demand response cost, and variable cost of the generator set in the re-dispatch stage; U is the photovoltaic output uncertainty set; ξ is an uncertain variable, which is the photovoltaic power generation power in the distribution network; x is the decision variable vector in the pre-dispatch stage, including u tn 、v tn 、w tn , Soc t , and y is the decision variable vector of the rescheduling phase, including P t purch , P t sell , P t cut , P t tran , g tn , P t dis and P t ch So far, the decision variables in the pre-dispatch stage are only 0-1 variables and continuous variable state of charge, and the decision variables in the re-dispatch stage are only continuous variables.

[0051] 5. The adaptive robust optimization model of active distribution network is a two-stage optimization model. The solution can be simplified by establishing a linear affine relationship between the decision variable y in the re-dispatch stage and the uncertainty of photovoltaic power generation ξ. The traditional affine decision rule will lead to conservative optimization results. Therefore, the traditional affine rule is improved to solve the optimization problem. When there are uncertain variables at time t-1, the decision variable at time t forms an affine relationship with the uncertainties at time t and time t-1. Otherwise, the decision variable at time t only forms an affine relationship with the uncertainties at time t, thereby increasing the flexibility of the decision variables, as shown in formula (12).

[0052]

[0053] Where: y t (ξ t ) is the affine expression of the decision variable in the rescheduling phase with respect to the uncertainty; Y t are the constant term and the linear term coefficient of the affine decision rule respectively; therefore, the adaptive robust optimization model of the improved affine decision rule can be simplified as shown in formula (13):

[0054]

[0055] In the formula: A, B, C, D, E are all coefficient matrices, r, d, w are coefficient vectors; the first row of constraints represents constraints related only to variables in the pre-dispatch phase, including generator state constraints, generator minimum start and stop time constraints, energy storage charging and discharging state constraints, and energy storage charge state constraints; the second row of constraints represents the coupling constraints of variables in the pre-dispatch and re-dispatch phases, including power balance constraints; the third row of constraints represents constraints related only to variables in the re-dispatch phase, including upper and lower limits of generator power, upper and lower limits of energy storage power, upper and lower limits of electricity purchase and sale, and demand response constraints. Finally, the active distribution network optimization model based on the GMM polyhedral uncertain set is transformed into a deterministic mixed integer linear programming model through robust equivalence, which can be directly solved by calling the commercial solver Gurobi.

[0056] Beneficial effects:

[0057] 1. The present invention uses GMM interval sets and GMM polyhedron sets to describe the uncertainty of photovoltaics, making the description of the uncertainty boundary of photovoltaic output more accurate and having stronger coverage. Compared with the classic interval sets and polyhedron sets, the present invention has a low total day-ahead scheduling cost and a low conservativeness of the optimization results;

[0058] 2. By using the improved affine rule, the decision information is enriched, the feasible space of decision variables is increased, and a more optimized feasible solution is easily obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of adaptive optimization of the distribution network of the present invention;

[0060] Figure 2 This is the cost diagram of Example A under sufficient lighting conditions of the present invention;

[0061] Figure 3 This is the cost diagram of Example B under sufficient lighting conditions of the present invention;

[0062] Figure 4 This is the cost diagram of Example C under sufficient lighting conditions of the present invention;

[0063] Figure 5 This is the cost diagram of Example D under sufficient lighting conditions of the present invention;

[0064] Figure 6 This is the cost diagram of Example A under insufficient lighting conditions of the present invention;

[0065] Figure 7 This is the cost diagram of Example B under insufficient lighting conditions of the present invention;

[0066] Figure 8 This is the cost diagram of Example C under insufficient lighting conditions of the present invention;

[0067] Fig. 9This is the cost diagram of Example D under insufficient lighting conditions of the present invention. DETAILED DESCRIPTION

[0068] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0069] In the description of the present invention, it should be understood that the terms "upper", "middle", "outer", "inner" and the like indicating directions or positional relationships are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the components or elements referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0070] Example:

[0071] A robust optimization method for distribution networks based on Gaussian mixture models with accurate uncertainty boundaries, such as Figure 1 As shown, the following steps are included:

[0072] Step 1: Select the photovoltaic output data of a certain area from January 1, 2021 to January 1, 2023, calculate the results of the photovoltaic output distribution parameters of the area at different times according to the Gaussian mixture model, establish an adaptive robust optimization uncertainty set, and simulate and analyze the distribution network. The system consists of generator sets (gas turbines, diesel generators), photovoltaics, energy storage, and loads. The test cycle is 24 hours, and the day-ahead time interval is 1 hour. For each device in the distribution network, its specific parameters are shown in Appendix Table 1. Table 2 is the time-of-use electricity price selected by the present invention, and an adaptive robust optimization model of active power on the day-ahead is established.

[0073] Table 1 Distribution network equipment parameters

[0074]

[0075]

[0076] Table 2 Time-of-use electricity prices

[0077]

[0078] Step 2: Input forecast load data and conduct 24-hour active power dispatch.

[0079] Step 3: Solve the adaptive robust optimization model to obtain the pre-dispatch decision variables and re-dispatch affine expressions, and decide on the start and stop of the unit and the energy storage charging and discharging plan.

[0080] Step 4: Observe the photovoltaic output data of the second day and substitute it into the redispatch affine expression to obtain the redispatch decision variables.

[0081] Step 5: Repeat step 4 until the loop condition is no longer met.

[0082] In order to verify the superiority of the precise uncertainty set based on the Gaussian mixture model of the present invention, it is compared with the traditional uncertainty set, as shown in Table 3.

[0083] Table 3 Impact of uncertainty sets on scheduling results under different lighting conditions (Γ=6)

[0084]

[0085]

[0086] Table 3 shows the scheduling results of four uncertainty sets under two kinds of light conditions. It can be seen that under sufficient light conditions, the fixed cost, startup cost and transferable load cost of gas turbines in all uncertainty sets remain unchanged, because the four sets all determine the same gas unit start-up and shutdown plan and transfer load plan, while the start-up time of the diesel unit in the classic uncertainty set is one hour earlier than that in the GMM uncertainty set, resulting in a higher fixed cost for the diesel unit. This shows that the scheduling flexibility of the classic uncertainty set is low, and only the diesel unit can be started to cope with the changes in photovoltaic output and load; under insufficient light conditions, the start-up and shutdown plans of each unit are the same, because the photovoltaic output range is reduced, which reduces the impact of the uncertainty set range on the start-up and shutdown plan and transfer load plan of the unit. In general, the interval sets of the two light conditions contain the entire range of photovoltaic output distribution, so the scheduling cost is the largest. The polyhedron set has a better optimization result than the interval set due to the addition of conservative adjustment; while the range of the GMM uncertainty set is more accurate, and the scheduling cost is smaller than that of the classic uncertainty set.

[0087] Table 4 compares the scheduling results of the improved affine rule of the present invention with those of the traditional affine rule. The scheduling results of the improved affine rule on the classic polyhedron set and the GMM polyhedron set are better than those of the traditional affine rule.

[0088] Table 4 Affine decision rule for scheduling results (Γ=6)

[0089]

[0090] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0091] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A robust optimization method for distribution network based on Gaussian mixture model with precise uncertainty boundary, characterized by: The steps include: Step 1: Cluster the photovoltaic historical data through the Gaussian mixture model to obtain the distribution of photovoltaic output under two lighting conditions; Step 2: Construct multiple uncertainty sets to describe the uncertainty of photovoltaic output through the parameters generated by the Gaussian mixture model; Step 3: Establish the comprehensive cost objective function of the active distribution network and the distribution network constraints; Step 4: Construct a multi-stage adaptive robust optimization dispatch model for active distribution networks, which is divided into two stages: pre-dispatching and re-dispatching. In the pre-dispatching stage, the electric load, generator on / off state variables, and energy storage charging and discharging state variables are determined before the uncertainty is observed; in the re-dispatching stage, the amount of electricity purchased and sold from the superior power grid, the amount of demand response, the amount of generator power generation, and the amount of energy storage charging and discharging are determined after the uncertainty is observed, and their existence and feasibility have been guaranteed by pre-dispatching. Step 5: Solve the adaptive robust optimization using the improved affine decision rule.

2. The method for robust optimization of distribution network based on Gaussian mixture model accurate uncertainty boundary according to claim 1, characterized in that: The process of establishing the Gaussian mixture model in step 1 is as follows: The Gaussian mixture model is not limited to a specific probability density function form assumption, and any probability density distribution can be approximated by a linear combination of several Gaussian density functions, which can accurately characterize the probability density distribution of random variables. The Gaussian mixture probability density function is expressed as follows: The Gaussian mixture model uses the expectation-maximization (EM) algorithm to fit n mixed sub-Gaussian distributions to obtain the mean and standard deviation of each distribution; The specific steps of the EM algorithm are as follows; 1) Randomly initialize model parameters ω k , μ k , σ k The initial value of 2) Step E: Calculate the conditional probability expectation of the joint distribution: Where: x = (x (1) ,x (2) ,...x (m) ) is the observed data; p(x (i) ,ω (i) |θ) is the joint distribution; p(ω (i) |x (i) ,θ) is the conditional distribution; 3) Step M: Maximize L(θ) to get θ j+1 : 4) If θ j+1 If convergence has been achieved, the algorithm ends; otherwise, the E and M steps are continued for iteration; At this point, the parameters of each sub-Gaussian distribution model of the Gaussian mixture model can be obtained using the EM algorithm.

3. The method for robust optimization of distribution network based on Gaussian mixture model accurate uncertainty boundary according to claim 2 is characterized in that: In step 2, multiple uncertain sets are constructed, including: The creation of the uncertainty set uses the Laida criterion, that is, the probability that the value is distributed in (μ-3σ,μ+3σ) is 0.9973, so this interval can be regarded as the actual possible interval of the uncertainty quantity; In view of the fact that the historical values ​​of photovoltaic output at different times can be considered to obey the multi-peak normal population, the important parameters (mean and standard deviation) of photovoltaic output in each period within 24 hours are obtained by mining the historical data of photovoltaic output using Gaussian mixture model and integrating them into the classic uncertainty set. The actual photovoltaic output, mean, standard deviation and offset are linked by the Laida criterion. For this purpose, the following interval and polyhedron uncertainty sets based on Gaussian mixture model parameters are established; As shown in formula (4) and (5): Where: U is the photovoltaic output uncertainty set; P t PV is the photovoltaic output value at time t; are the mean and standard deviation of photovoltaic output at time t; ΔP t PV is the output fluctuation of photovoltaic at time t; δ t is the fluctuation coefficient of photovoltaic output at time t. From formula (5), we can know that δ t ∈[-1,1]; Γ is the conservatism, but in reality it is unlikely that all predictions will reach the boundary at the same time. This is determined by the central limit theorem and has nothing to do with which probability distribution the individual follows. The conservatism of the present invention is defined as the sum of the absolute values ​​of the ratios of the actual photovoltaic output fluctuation to the maximum fluctuation upper limit within the scheduling period. The larger the Γ value, the more conservative the model; the smaller the Γ value, the more radical the model.

4. The method for robust optimization of distribution network based on Gaussian mixture model accurate uncertainty boundary according to claim 3 is characterized in that: The comprehensive cost objective function and constraints of the active distribution network in step 3 include: The goal of the day-ahead optimization operation is to minimize the sum of the electricity purchase cost, demand response cost, diesel generator and gas turbine power generation cost, as shown below: Where: T and N are the number of dispatching periods and the number of generators respectively; C t purch , Respectively represent the electricity purchase cost, demand response cost and generator cost during the dispatch period; λ t purch , λ fixed , variable , on are the unit electricity price of the power purchased by the distribution network from the upper power grid at time t, the demand response cost coefficient, the generator fixed cost coefficient, the generator variable cost coefficient, and the generator startup cost coefficient; P t purch , P t tran , P t cut 、u tn , g tn 、v tn They represent the amount of electricity purchased, the amount of load that can be transferred, the amount of load that can be reduced, the on / off status of the generator, the amount of power generated by the generator, and the generator on / off flag at time t in the dispatching cycle; the binary variable of the on / off status, when the value is 1, it means the generator is on; 0 means it is off; when the on / off flag is 1, it means the generator is on at that moment; The constraints include: (1) System power balance constraints Where: P t ch and P t dis Respectively represent the charging and discharging power of energy storage at time t; P t purch and P t sell They represent the power purchased and sold by the distribution network to the upper grid at time t; g t P represents the sum of the output power of the generators in the distribution network at time t; t pv , P t load They represent the photovoltaic output power and load power consumption at time t respectively; (2) Generator operation constraints Where: w tn It is the generator shutdown flag. When the shutdown flag is 1, it means that the generator is shut down at that moment. The minimum capacity of the nth generator set, The maximum capacity of the nth generator set, The minimum startup time of the nth unit, The minimum downtime of the nth generator set, The ramp rate limit of the nth generator set, The ramp rate limit of the nth generator set; (3) Energy storage operation constraints Where: are 0-1 variables, representing the energy storage charging and discharging states respectively; P ch , Indicates the upper and lower limits of energy storage charging power; P dis , Indicates the upper and lower limits of energy storage discharge power; Soc t Indicates the energy storage charge state; η ch , η dis Indicates the energy storage charging and discharging efficiency coefficient; E e is the rated capacity of energy storage; Soc , Indicates the upper and lower limits of the state of charge; to ensure that there is still room for energy storage at the beginning of the next dispatch cycle, the state of charge should be within the range of 0.4-0.6 after the end of the previous dispatch cycle; (4) Demand response resource constraints Where: They are the upper limits of load that can be reduced and load that can be transferred in each period; is the total amount of transferable load demand; The total upper limit of the load that can be reduced.

5. The method for robust optimization of distribution network based on Gaussian mixture model accurate uncertainty boundary according to claim 4 is characterized in that: The multi-stage adaptive robust optimization dispatching model of the active distribution network constructed in step 4 includes: The objective function (6) is transformed into a multi-stage adaptive robust optimization model as follows: Where: a T x is the sum of the startup cost and fixed cost of the generator set in the pre-dispatch stage; b T y is the sum of the power purchase cost, demand response cost, and variable cost of the generator set in the re-dispatch stage; U is the photovoltaic output uncertainty set; ξ is an uncertain variable, which is the photovoltaic power generation power in the distribution network; x is the decision variable vector in the pre-dispatch stage, including u tn 、v tn 、w tn , Soc t , and y is the decision variable vector in the redispatching phase, including P t purch , P t sell , P t cut , P t tran , g tn , P t dis and P t ch So far, the decision variables in the pre-dispatch stage are only 0-1 variables and continuous variable state of charge, and the decision variables in the re-dispatch stage are only continuous variables.

6. The method for robust optimization of distribution network based on Gaussian mixture model accurate uncertainty boundary according to claim 5 is characterized in that: The adaptive robust optimization model of active distribution network is a two-stage optimization model. The solution can be simplified by establishing a linear affine relationship between the decision variable y in the re-dispatch stage and the uncertainty of photovoltaic power generation power ξ. The traditional affine decision rule will lead to conservative optimization results. Therefore, the traditional affine rule is improved to solve the optimization problem. When there are uncertain variables at time t-1, the decision variable at time t forms an affine relationship with the uncertainties at time t and time t-1. Otherwise, the decision variable at time t only forms an affine relationship with the uncertainty at time t, thereby increasing the flexibility of the decision variables, as shown in formula (12). Where: y t (ξ t ) is the affine expression of the decision variable in the rescheduling phase with respect to the uncertainty; Y t are the constant term and the linear term coefficient of the affine decision rule respectively; Therefore, the adaptive robust optimization model of the improved affine decision rule can be simplified as shown in formula (13): Wherein: A, B, C, D, E are coefficient matrices, r, d, w are coefficient vectors; the constraints in the first row represent constraints related only to variables in the pre-dispatch stage, including generator state constraints, generator minimum start and stop time constraints, energy storage charging and discharging state constraints, and energy storage charge state constraints; the constraints in the second row represent the coupling constraints of variables in the pre-dispatch and re-dispatch stages, including power balance constraints; the constraints in the third row represent constraints related only to variables in the re-dispatch stage, including upper and lower limit constraints on generator power, upper and lower limit constraints on energy storage power, upper and lower limit constraints on purchased and sold electricity, and demand response constraints; the active distribution network optimization model based on the GMM polyhedral uncertain set is transformed into a deterministic mixed integer linear programming model through robust equivalence, which can be directly solved by calling the commercial solver Gurobi.