Robust evaluation method for distributed photovoltaic carrying capacity of power distribution network based on deep learning

Abstract

The invention relates to a photovoltaic bearing capacity assessment method, in particular to a power distribution network distributed photovoltaic bearing capacity robust assessment method based on deep learning, and the method comprises the steps: firstly carrying out the linear processing of an AC power flow equation; secondly, learning photovoltaic space-time probability characteristics from historical data of photovoltaic power generation by utilizing a generative adversarial network, designing a distributed robust fuzzy set based on photovoltaic power distribution induced by a generator and a Wasserstein distance, and constructing a distributed robust joint opportunity constraint photovoltaic bearing capacity evaluation model based on deep learning; and finally, solving by adopting a Bonferroni inequality and a conditional risk value, and converting the distributed robust joint opportunity constraint model into a deterministic linear continuous model for solving. The model meets the opportunity constraint of safe operation of a plurality of power systems at the same time, and complex and even nonlinear photovoltaic power space-time correlation is embedded, so that the conservative property of distributed robust optimization is further reduced.

Description

Technical Field

[0001] The present invention relates to a photovoltaic carrying capacity evaluation method, in particular to a robust evaluation method for distributed photovoltaic carrying capacity of a distribution network based on deep learning. Background Art

[0002] With the reduction of photovoltaic power generation costs and people's desire to reduce fossil fuel power generation, photovoltaic power generation has developed rapidly around the world. However, the vigorous development of distributed photovoltaics and its variable power generation characteristics may cause several operational problems in the distribution network (such as overvoltage or overcurrent violations). Therefore, before the actual deployment of distributed photovoltaic power generation, distribution network operators need to understand and determine the photovoltaic carrying capacity of the distribution network, that is, the maximum photovoltaic access capacity that the distribution network can carry without violating technical restrictions.

[0003] Since the fluctuation of photovoltaic output power has an important impact on the size of photovoltaic carrying capacity, the evaluation of photovoltaic carrying capacity under uncertainty models has always been the focus of research. Uncertainty models can be roughly divided into three directions, namely robust optimization (RO), stochastic optimization (SO) and distributed robust optimization (DRO). However, RO usually has conservative problems, while SO may produce overly ideal evaluation results and usually lead to heavy computational burdens.

[0004] DRO combines the advantages of RO and SO. DRO assumes that the true distribution is in a fuzzy set containing a series of potential distributions. The exact distribution of uncertain variables is unknown, but fuzzy sets are constructed based on limited data. Therefore, constructing fuzzy sets of uncertain variables is the basis of DRO. However, the traditional fuzzy sets constructed based on moment information cannot fully reflect the probability characteristics of random variables, which may lead to overly conservative decisions. Therefore, some scholars have studied the distance-based probability distribution fuzzy set construction method.

[0005] Although existing PV carrying capacity assessment studies have utilized the uncertainty of PV reflected in the data, these methods are generally based on a small number of typical scenarios. When the actual scenario deviates from the preset scenario, the assessment results will be biased. In addition, DRO usually establishes empirical distributions based on historical data. However, the empirical distribution model constructed based on data fitting is difficult to accurately capture the complex probabilistic characteristics and spatiotemporal correlation of PV output power. Generative adversarial networks are very effective in learning complex distributions of high-dimensional data samples. They do not need to assume any specific form of probability distribution. Therefore, the uncertainty information extracted by the generative adversarial network can be used to design the fuzzy set of PV power distribution using the distribution induced by the generator neural network.

[0006] In order to deal with the uncertainty of constraints, chance constraints are usually combined with DRO, such as distributed robust chance constraints based on Wasserstein distance. However, the existing distributed robust chance constraint photovoltaic carrying capacity assessment model mainly focuses on individual constraints, ignoring the spatiotemporal correlation of photovoltaic output power uncertainty embedded in many different constraints. The above problems limit the practical effect of DRO method in photovoltaic carrying capacity assessment problem. Summary of the invention

[0007] Aiming at the problem that the existing distributed robust chance-constrained photovoltaic carrying capacity assessment model ignores the spatiotemporal correlation of photovoltaic output power uncertainty, resulting in poor actual effect of photovoltaic carrying capacity assessment, the present invention proposes a robust assessment method for distributed photovoltaic carrying capacity of distribution network based on deep learning.

[0008] The technical solution adopted by the present invention is as follows: a method for robust evaluation of distributed photovoltaic carrying capacity of distribution network based on deep learning, comprising the following steps:

[0009] Step 1: Linearize the AC power flow equation; the voltage of the PQ node in the distribution network is expressed as follows: v i,t =v nom +(RP i,t +XQ i,t ) / |v nom |; Where: v nom is the nominal voltage, P i,t is the active power of each node at each time, Q i,t is the reactive power of each node at each moment, R and X are resistance matrix and reactance matrix respectively;

[0010] Step 2: Taking the maximum capacity of distributed photovoltaic access in the distribution network as the goal, considering the joint opportunity constraints consisting of power flow constraints, photovoltaic output constraints, voltage constraints, current constraints and transformer reverse overload rate constraints, a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model is constructed.

[0011] 1) The objective function is to maximize the total photovoltaic access capacity in the distribution network:

[0012]

[0013] In the formula, is the photovoltaic grid-connected capacity of photovoltaic grid-connected node i in the distribution network; Ω PV It is the collection of photovoltaic grid-connected nodes in the distribution network;

[0014] 2) Power flow constraints:

[0015]

[0016] Where: S i,t is the complex power of each node at each time, P i,t and Q i,t are the active power and reactive power of each node at each moment respectively; and are the active power and reactive power of each photovoltaic unit at each moment, and are respectively the active load power and reactive load power of each node at each moment; Ω n is the set of all nodes in the distribution network; R and X are resistance matrix and reactance matrix respectively, and Y is the block matrix of the distribution network node admittance matrix, V 1,t is the voltage at the equilibrium node;

[0017] 3) Photovoltaic output constraints

[0018]

[0019] Where: is the photovoltaic output coefficient caused by solar irradiance, is the upper limit of the power factor angle;

[0020] 4) Joint opportunity constraints

[0021]

[0022] Where: The joint opportunity constraint includes voltage constraint, current constraint and transformer reverse overload rate constraint; and are the upper and lower limits of voltage respectively, ε is the risk coefficient of inequality constraint exceeding the limit, and are the upper and lower limits of the current respectively; λ max is the maximum reverse load rate of the transformer, S T,max represents the maximum transmission capacity of transformer T, p represents the probability distribution, is a fuzzy set of probability distribution;

[0023] Step 3: To solve the uncertainty of distributed photovoltaic output, a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model based on deep learning is constructed. The probability distribution fuzzy set in the distributed robust joint opportunity constraint model is constructed using a deep learning-based method.

[0024] 1) Construct a generative adversarial network (GAN). The goal of the GAN is to model the probability distribution of historical data on photovoltaic output.

[0025] 2) Construct a fuzzy set based on deep learning and learn the best approximation P of the probability distribution P of historical data of photovoltaic output based on the generative adversarial network GAN G , and use fuzzy sets based on Wasserstein distance to hedge the uncertainty of the distribution, the probability distribution P of PV historical data and the best approximation P G The Wasserstein distance between is defined as:

[0026]

[0027] Where: Ξ is a random variable The support set of is the sample set of random variables; Π is and On Marginal P and P G The joint probability distribution of ||·|| is the l1 norm; therefore, the fuzzy set based on deep learning is defined as:

[0028]

[0029] Where: represents the set of all probability distributions in the support set Ξ; ρ is the radius of the Wasserstein ball;

[0030] After the distributed robust joint opportunity-constrained photovoltaic carrying capacity assessment model based on deep learning is constructed, it is solved to solve the photovoltaic carrying capacity of the distribution network.

[0031] The above-mentioned robust evaluation method for distributed photovoltaic carrying capacity of distribution network based on deep learning includes the following steps: The specific process of linearizing the AC power flow equation in step 1 is:

[0032] 1) The operation of the power grid in the sinusoidal steady state is described by Kirchhoff's current law as follows:

[0033]

[0034] Where: y1, and Y are the block matrices of the distribution network node admittance matrix; I 1,t is the current at the equilibrium node, i i,t =[I 2,t ,I 3,t ,...,I N,t ] T is the current of the PQ node, V 1,t is the voltage at the equilibrium node, v i,t =[V 2,t ,V 3,t ,...,V N,t ] Tis the voltage of the PQ node, N is the number of distribution network nodes, and i is the distribution network node number;

[0035] 2) The complex power injection at each node is as follows:

[0036]

[0037] Where: S i,t =[S 2,t ,S 3,t ,...,S N,t ] T is the complex power of each node at each time, and S i,t =P i,t +jQ i,t , j is the imaginary part, P i,t =[P 2,t ,P 3,t ,...,P N,t ] T is the active power of each node at each time, Q i,t =[Q 2,t ,Q 3,t ,...,Q N,t ] T is the reactive power of each node at each time; * indicates conjugation;

[0038] 3) Assume v i,t =v nom +Δv i,t Linearize the above equation

[0039]

[0040] Where: v nom =|v nom |∠θ nom is the nominal voltage, θ nom is the phase angle of the nominal voltage, Δv i,t Indicates the deviation of the PQ node voltage from the nominal voltage;

[0041] 4) Set v nom The voltage value when the input current at both ends of the network is zero: Δv i,t The solution is as follows

[0042]

[0043] Where: Z = Y -1 =R+jX, Z is the node impedance matrix; R and X are the resistance matrix and reactance matrix respectively, then:

[0044]

[0045] Where: Denotes Δv i,t The real part of Denotes Δv i,t The imaginary part of

[0046] 5) In the distribution network, the phase angle is 0, and θ is set nom =0, the voltage is expressed as follows:

[0047] v i,t =v nom +(RP i,t +XQ i,t ) / |v nom |.

[0048] The above-mentioned robust evaluation method for distributed photovoltaic carrying capacity of distribution network based on deep learning includes the following steps: after constructing fuzzy sets based on deep learning, solving the distributed robust joint opportunity constraint photovoltaic carrying capacity evaluation model based on deep learning, the specific process is as follows:

[0049] 1) The compact form of the joint opportunity constraint is expressed as follows:

[0050]

[0051] Where: y t is a vector consisting of decision variables in the distribution network, a k and b k Yes t is a linear function of , K is the number of individual chance constraints in the joint chance constraint;

[0052] 2) Use the Bonferroni inequality to transform the joint chance constraint into a single chance constraint:

[0053]

[0054] Where: ε k where is the risk coefficient of individual opportunity constraint, ε k ≥0 and

[0055] 3) Use the worst-case conditional value-at-risk (CVaR) constraint to approximate a single worst-case opportunity constraint:

[0056]

[0057] Where: τ k is an auxiliary variable; (·) + =max{·,0}; further transformation yields:

[0058]

[0059] Where: k,t 、s k,n,t , γ k,n is the dual variable generated by the transformation process at time t, ||·|| ∞ is the infinite norm; so far, by using the Bonferroni equation and CVaR approximation, the distributed robust joint chance constraint equation based on Wasserstein distance is transformed into a linear constraint equation, and the original distributed robust joint constraint problem is transformed into a mixed integer linear programming problem for solution.

[0060] The method of the present invention first linearizes the AC power flow equation to solve the non-convex problem of traditional AC power flow, so that the distributed robust joint opportunity constraint photovoltaic carrying capacity assessment problem can be calculated using conditional risk value pairs; then, the generative adversarial network is used to learn the photovoltaic spatiotemporal probability characteristics from the historical data of photovoltaic power generation, and a distributed robust fuzzy set is designed based on the photovoltaic power distribution and Wasserstein distance induced by the generator, and a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model based on deep learning is constructed, considering the spatiotemporal correlation of photovoltaic power uncertainty embedded in many different constraints; finally, the Bonferroni inequality and conditional risk value are used to solve, and the distributed robust joint opportunity constraint model is converted into a deterministic linear continuous model for solution. The model simultaneously satisfies the opportunity constraints of safe operation of multiple power systems, and embeds complex and even nonlinear photovoltaic power spatiotemporal correlation, further reducing the conservatism of distributed robust optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Schematic diagram of the distributed robust joint opportunity-constrained photovoltaic carrying capacity assessment model based on deep learning described in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The robust evaluation method of distributed photovoltaic carrying capacity of distribution network based on deep learning includes the following steps:

[0063] Step 1: To solve the non-convex problem of the AC power flow equation, the AC power flow equation is linearized;

[0064] 1) The operation of the power grid in the sinusoidal steady state is described by Kirchhoff's current law as follows:

[0065]

[0066] Where: y1, and Y are the block matrices of the distribution network node admittance matrix; I 1,t is the current matrix of the equilibrium node, i i,t =[I 2,t ,I3,t ,...,I N,t ] T is the current matrix of the PQ node, V 1,t is the voltage matrix of the equilibrium node, v i,t =[V 2,t ,V 3,t ,...,V N,t ] T is the voltage matrix of the PQ node, N is the number of distribution network nodes, and i is the node number of the distribution network;

[0067] 2) The complex power injection at each node is as follows:

[0068]

[0069] Where: S i,t =[S 2,t ,S 3,t ,...,S N,t ] T is the complex power of each node at each time, and S i,t =P i,t +jQ i,t , j is the imaginary part, P i,t =[P 2,t ,P 3,t ,...,P N,t ] T is the active power of each node at each time, Q i,t =[Q 2,t ,Q 3,t ,...,Q N,t ] T is the reactive power of each node at each time; * indicates conjugation;

[0070] 3) Assume v i,t =v nom +Δv i,t Linearize the above equation

[0071]

[0072] Where: v nom =|v nom |∠θ nom is the nominal voltage, θ nom is the phase angle of the nominal voltage, Δv i,t Indicates the deviation of the PQ node voltage from the nominal voltage;

[0073] 4) Set v nom The voltage value when the input current at both ends of the network is zero: Δv i,t The solution is as follows

[0074]

[0075] Where: Z = Y -1 =R+jX, Z is the node impedance matrix; R and X are the resistance matrix and reactance matrix respectively, then:

[0076]

[0077] Where: Denotes Δv i,t The real part of Denotes Δv i,t The imaginary part of

[0078] 5) In the distribution network, the phase angle is close to 0, and θ is set nom =0, the voltage amplitude is approximately expressed as follows:

[0079] v i,t =v nom +(RP i,t +XQ i,t ) / |v nom |

[0080] Step 2: Taking the maximum capacity of distributed photovoltaic access in the distribution network as the goal, considering the joint opportunity constraints consisting of power flow constraints, photovoltaic output constraints, voltage constraints, current constraints and transformer reverse overload rate constraints, a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model is constructed. This model simultaneously meets the opportunity constraints for the safe operation of multiple power systems.

[0081] 1) The objective function is to maximize the total photovoltaic capacity in the distribution network:

[0082]

[0083] In the formula, is the photovoltaic grid-connected capacity of node i; Ω PV It is a collection of photovoltaic grid-connected nodes;

[0084] 2) Power flow constraints:

[0085]

[0086] Where: S i,t is the complex power of each node at each time, P i,t and Q i,t are the active power and reactive power of each node at each moment respectively; and are the active power and reactive power of each photovoltaic unit at each moment, and are respectively the active load power and reactive load power of each node at each moment; Ωn The set of all nodes.

[0087] 3) Photovoltaic output constraints

[0088]

[0089] Where: is the photovoltaic output coefficient caused by solar irradiance, is the upper limit of the power factor angle.

[0090] 4) Joint opportunity constraints

[0091]

[0092] Where: The joint opportunity constraint includes voltage constraint, current constraint and transformer reverse overload rate constraint; and are the upper and lower limits of voltage respectively, ε is the risk coefficient of inequality constraint exceeding the limit, and are the upper and lower limits of the current respectively; λ max is the maximum reverse load rate of the transformer, which is 80%; S T,max represents the maximum transmission capacity of transformer T, p represents the probability distribution, is a fuzzy set of probability distributions.

[0093] Step 3: In order to solve the uncertainty of distributed photovoltaic output, a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model based on deep learning is constructed. The probability distribution fuzzy set in the distributed robust joint opportunity constraint model is constructed based on deep learning, which can embed complex and even nonlinear photovoltaic power spatiotemporal correlation, further reducing the conservatism of distributed robust optimization.

[0094] 1) Construct a generative adversarial network (GAN). The goal of GAN is to model the probability distribution of historical data of photovoltaic output. The generator G transforms the noise z~P z Mapped to the corresponding generated sample G(z). Then, the discriminator D transforms the real sample x~P x Different from the generated sample G(z), its objective function is as follows:

[0095]

[0096] Where: It is the mathematical expectation function under a certain probability distribution.

[0097] 2) Construct a fuzzy set based on deep learning. Based on the generative adversarial network GAN, learn the best approximation P of the probability distribution P of the historical data of photovoltaic output G, and uses fuzzy sets based on Wasserstein distance to hedge the uncertainty of the distribution, so that the shape and characteristics of the true distribution can be captured more accurately, thereby further reducing the conservatism of DRO. Probability distribution P of photovoltaic historical data and the best approximation P G The Wasserstein distance between is defined as:

[0098]

[0099] Where: Ξ∈R is a random variable The support set of is the sample set of random variables; Π is and On Marginal P and P G The joint probability distribution of ; ||·|| is the l1 norm. Therefore, the fuzzy set based on deep learning is defined as:

[0100]

[0101] Where: represents the set of all probability distributions in the support set Ξ; ρ is the radius of the Wasserstein ball.

[0102] 3) After constructing fuzzy sets based on deep learning, the compact form of the joint chance constraint is expressed as follows:

[0103] Where: y t is a vector consisting of decision variables in the distribution network, a k and b k Yes t is a linear function of , and K is the number of individual chance constraints in the joint chance constraint.

[0104] Step 4: Solve the distributed robust joint opportunity-constrained photovoltaic carrying capacity assessment model based on deep learning.

[0105] 1) First, the joint chance constraint is transformed into a single chance constraint using the Bonferroni inequality:

[0106]

[0107] Where: ε k where is the risk coefficient of individual opportunity constraint, ε k ≥0 and

[0108] 2) Use the worst-case conditional value-at-risk (CVaR) constraint to approximate a single worst-case chance constraint:

[0109]

[0110] Where: τ k is an auxiliary variable; (·) + =max{·,0}; further transformation yields:

[0111]

[0112] Where: k,t 、s k,n,t , γ k,n is the dual variable generated by the transformation process at time t, ||·|| ∞ is the infinite norm. So far, by using the Bonferroni equation and CVaR approximation, the distributed robust joint constraint equation based on Wasserstein distance is transformed into a linear constraint equation, and the original distributed robust joint constraint problem is transformed into a mixed integer linear programming problem, which can be solved using commercial solvers such as Gurobi.

[0113] In summary, compared with the prior art, the beneficial effects of the present disclosure are:

[0114] This invention proposes a distributed robust joint opportunity-constrained photovoltaic carrying capacity assessment model based on deep learning. First, a linearized power flow model is adopted to solve the non-convexity of the AC power flow model, and it is used in the photovoltaic carrying capacity assessment model to speed up the calculation efficiency; then, a generative adversarial network is used to learn the spatiotemporal probability characteristics from the historical data of photovoltaic power generation, and a fuzzy set of photovoltaic power distribution based on deep learning and Wasserstein distance is proposed, and a distributed robust joint opportunity-constrained photovoltaic carrying capacity assessment model based on deep learning is constructed. The model considers the spatiotemporal correlation of photovoltaic power uncertainty embedded in many different constraints, so that multiple opportunity constraints are satisfied at the same time, reducing the conservatism of traditional distributed robust optimization, effectively responding to the challenges of renewable energy uncertainty to photovoltaic carrying capacity assessment, and improving the accuracy and effectiveness of the photovoltaic carrying capacity assessment results of the distribution network.

Claims

1. A robust evaluation method for distributed photovoltaic carrying capacity of distribution network based on deep learning, characterized by: The following steps are involved: Step 1: Linearize the AC power flow equation; the voltage of the PQ node in the distribution network is expressed as follows: v i,t =v nom +(RP i,t +XQ i,t ) / |v nom |; Where: v nom is the nominal voltage, P i,t is the active power of each node at each time, Q i,t is the reactive power of each node at each moment, R and X are resistance matrix and reactance matrix respectively; Step 2: Taking the maximum capacity of distributed photovoltaic access in the distribution network as the goal, considering the joint opportunity constraints consisting of power flow constraints, photovoltaic output constraints, voltage constraints, current constraints and transformer reverse overload rate constraints, a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model is constructed. 1) The objective function is to maximize the total photovoltaic access capacity in the distribution network: In the formula, is the photovoltaic grid-connected capacity of photovoltaic grid-connected node i in the distribution network; Ω PV It is the collection of photovoltaic grid-connected nodes in the distribution network; 2) Power flow constraints: Where: S i,t is the complex power of each node at each time, P i,t and Q i,t are the active power and reactive power of each node at each moment respectively; and are the active power and reactive power of each photovoltaic unit at each moment, and are respectively the active load power and reactive load power of each node at each moment; Ω n is the set of all nodes in the distribution network; R and X are resistance matrix and reactance matrix respectively, and Y is the block matrix of the distribution network node admittance matrix, V 1,t is the voltage at the equilibrium node; 3) Photovoltaic output constraints Where: is the photovoltaic output coefficient caused by solar irradiance, is the upper limit of the power factor angle; 4) Joint opportunity constraints Where: The joint opportunity constraint includes voltage constraint, current constraint and transformer reverse overload rate constraint; and are the upper and lower limits of voltage respectively, ε is the risk coefficient of inequality constraint exceeding the limit, and are the upper and lower limits of the current respectively; λ max is the maximum reverse load rate of the transformer, S T,max represents the maximum transmission capacity of transformer T, p represents the probability distribution, is a fuzzy set of probability distribution; Step 3: To solve the uncertainty of distributed photovoltaic output, a distributed robust joint opportunity constraint photovoltaic carrying capacity assessment model based on deep learning is constructed. The probability distribution fuzzy set in the distributed robust joint opportunity constraint model is constructed using a deep learning-based method. 1) Construct a generative adversarial network (GAN). The goal of the GAN is to model the probability distribution of historical data on photovoltaic output. 2) Construct a fuzzy set based on deep learning and learn the best approximation P of the probability distribution P of historical data of photovoltaic output based on the generative adversarial network GAN G , and use fuzzy sets based on Wasserstein distance to hedge the uncertainty of the distribution, the probability distribution P of PV historical data and the best approximation P G The Wasserstein distance between is defined as: Where: Ξ is a random variable The support set of is the sample set of random variables; Π is and On Marginal P and P G The joint probability distribution of ||·|| is the l1 norm; therefore, the fuzzy set based on deep learning is defined as: Where: represents the set of all probability distributions in the support set Ξ; ρ is the radius of the Wasserstein ball; After the distributed robust joint opportunity-constrained photovoltaic carrying capacity assessment model based on deep learning is constructed, it is solved to solve the photovoltaic carrying capacity of the distribution network.

2. The method for robust evaluation of distributed photovoltaic carrying capacity of distribution network based on deep learning according to claim 1 is characterized in that: The following steps are involved: The specific process of linearizing the AC power flow equation in step 1 is: 1) The operation of the power grid in the sinusoidal steady state is described by Kirchhoff's current law as follows: Where: y1, and Y are the block matrices of the distribution network node admittance matrix; I 1,t is the current at the equilibrium node, i i,t =[I 2,t ,I 3,t ,...,I N,t ] T is the current of the PQ node, V 1,t is the voltage at the equilibrium node, v i,t =[V 2,t ,V 3,t ,...,V N,t ] T is the voltage of the PQ node, N is the number of distribution network nodes, and i is the distribution network node number; 2) The complex power injection at each node is as follows: Where: S i,t =[S 2,t ,S 3,t ,...,S N,t ] T is the complex power of each node at each time, and S i,t =P i,t +jQ i,t , j is the imaginary part, P i,t =[P 2,t ,P 3,t ,...,P N,t ] T is the active power of each node at each time, Q i,t =[Q 2,t ,Q 3,t ,...,Q N,t ] T is the reactive power of each node at each time; * indicates conjugation; 3) Assume v i,t =v nom +Δv i,t Linearize the above equation Where: v nom =|v nom |∠θ nom is the nominal voltage, θ nom is the phase angle of the nominal voltage, Δv i,t Indicates the deviation of the PQ node voltage from the nominal voltage; 4) Set v nom The voltage value when the input current at both ends of the network is zero: Δv i,t The solution is as follows Where: Z = Y -1 =R+jX, Z is the node impedance matrix; R and X are the resistance matrix and reactance matrix respectively, then: Where: Denotes Δv i,t The real part of Denotes Δv i,t The imaginary part of 5) In the distribution network, the phase angle is 0, and θ is set nom =0, the voltage is expressed as follows: in i,t =in nom +(RP i,t +XQ i,t ) / |v nom |。 3. The method for robust evaluation of distributed photovoltaic carrying capacity of distribution network based on deep learning according to claim 2 is characterized in that: The following steps are involved: After constructing the fuzzy set based on deep learning, the distributed robust joint opportunity-constrained photovoltaic carrying capacity evaluation model based on deep learning is solved. The specific process is as follows: 1) The compact form of the joint opportunity constraint is expressed as follows: Where: y t is a vector consisting of decision variables in the distribution network, a k and b k Yes t is a linear function of , K is the number of individual chance constraints in the joint chance constraint; 2) Use the Bonferroni inequality to transform the joint chance constraint into a single chance constraint: Where: ε k where is the risk coefficient of individual opportunity constraint, ε k ≥0 and 3) Use the worst-case conditional value-at-risk (CVaR) constraint to approximate a single worst-case opportunity constraint: Where: τ k is an auxiliary variable; (·) + =max{,0}; further transformation yields: Where: k,t 、s k,n,t , γ k,n is the dual variable generated by the transformation process at time t, ||·|| ∞ is the infinite norm; so far, by using the Bonferroni equation and CVaR approximation, the distributed robust joint chance constraint equation based on Wasserstein distance is transformed into a linear constraint equation, and the original distributed robust joint constraint problem is transformed into a mixed integer linear programming problem for solution.

Images