A grid structure optimization method and device for high-penetration new energy distribution network

Abstract

The present invention discloses a grid structure optimization method and device for a high-penetration new energy distribution network, the method comprising: according to the grid structure of the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained by the power quality scoring function, the association relationship is trained using a teacher network containing a strong learner; the probability distribution of the trained teacher network is imparted to a student network containing a weak learner; a comprehensive energy complementary model is established to represent the structure of the high-penetration new energy distribution network, and a chaotic particle swarm algorithm is used to optimize the comprehensive energy complementary model with different parameters to obtain a plurality of different high-penetration new energy distribution network structures; the student network is used to evaluate and select the plurality of different high-penetration new energy distribution network structures. According to the present invention, the chaotic particle swarm-knowledge distillation combination is used to optimize the grid structure, improve the power distribution path, and improve the power quality of the distribution network system.

Description

Technical Field

[0001] The present invention relates to the field of power grid operation and maintenance technology, and in particular to a grid structure optimization method and device for a high-penetration new energy distribution network. Background Art

[0002] At present, the coupling degree of information-physical system of distribution network is constantly improving, and a large number of power electronic equipment are connected. The traditional distribution network planning method cannot achieve the problem of optimal resource allocation. The distribution network system under the high penetration rate of new energy access is gradually developing and integrating into a comprehensive energy distribution network. The comprehensive energy distribution network is a distribution network system that can integrate and manage distributed power generation resources in different energy forms, and is supplemented by energy management systems, monitoring systems and other safety automatic devices to achieve multi-energy complementary coordination.

[0003] In fact, the poor power quality in the distribution network includes indicators such as voltage deviation, harmonic voltage, voltage fluctuation, and three-phase imbalance. There are many factors that cause these power quality indicators, including unbalanced three-phase load, insufficient reactive power compensation capacity at the end of the system, and many harmonic sources in the system. However, these indicators are also largely due to the unreasonable grid structure of the distribution network itself. When optimizing the distribution network grid, existing applications always ignore the impact of power quality indicators, resulting in inadequate optimization of some key nodes and networks of the distribution network when optimizing the distribution network grid. Summary of the invention

[0004] The embodiments of the present invention provide a grid structure optimization method and device for a high-penetration new energy distribution network, which utilizes a chaotic particle swarm-knowledge distillation combination to optimize the grid structure, improve the power distribution path, improve the power quality of the distribution network system under high-penetration new energy access, reduce the network transmission loss of the distribution network, and improve the optimization reliability.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application provides a grid structure optimization method for a high-penetration new energy distribution network, comprising:

[0006] According to multiple power quality indicators, the power quality scoring function is obtained by using fuzzy analytic hierarchy process.

[0007] According to the grid structure of the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained by the power quality scoring function, the association relationship is trained using a teacher network containing a strong learner; the teacher network is used to perform knowledge distillation according to the association relationship between the power quality score and the grid structure to obtain the pros and cons distribution of different grid optimization schemes; the strong learner has a higher degree of learning of the training set and a higher degree of fitting of the empirical function than the weak learner;

[0008] imparting the trained probability distribution of the teacher network to the student network containing the weak learner;

[0009] A comprehensive energy complementation model is established to represent the structure of a high-penetration new energy distribution network, and a chaotic particle swarm algorithm is used to optimize the comprehensive energy complementation model with different parameters under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints, with the objective function of maximizing the power quality score function, to obtain multiple different high-penetration new energy distribution network structures;

[0010] The student network is used to evaluate the optimized multiple different high-penetration new energy distribution network structures, and the high-penetration new energy distribution network structure with the highest power quality score is used as the optimized grid structure.

[0011] In a possible implementation manner of the first aspect, the multiple power quality indicators include voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance; and the power quality score is obtained by weighted summing up the scores and weights corresponding to each indicator.

[0012] In a possible implementation manner of the first aspect, the scoring solution process corresponding to each indicator specifically includes:

[0013] Establishing a hierarchical model, constructing a fuzzy priority relationship matrix, and then solving the fuzzy priority relationship matrix to obtain a fuzzy consistency matrix;

[0014] The fuzzy consistent matrix is ​​converted into a reciprocal judgment matrix, and the relative weight of each index is calculated by using a ranking method;

[0015] According to the membership function and the evaluation value corresponding to the membership function, the score corresponding to each indicator is obtained.

[0016] In a possible implementation of the first aspect, the grid structure according to the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained by the power quality scoring function, and the association relationship trained by using a teacher network containing a strong learner, specifically include:

[0017] Acquire historical data from the cyber-physical system of the distribution network under high penetration of new energy, wherein the historical data includes various grid structures that have been in operation and corresponding power quality score data;

[0018] A teacher network containing a strong learner is trained according to the historical data, and knowledge distillation is performed on the probability distribution in the teacher network; the probability distribution is related to the distribution of the power quality scoring function and the grid structure.

[0019] In a possible implementation manner of the first aspect, imparting the trained probability distribution of the teacher network to the student network containing the weak learner specifically includes:

[0020] Under the condition that the function value of the loss function of the preset knowledge distillation is minimized, the probability distribution of the trained teacher network is imparted to the student network containing the weak learner.

[0021] A second aspect of the embodiment of the present application provides a grid structure optimization device for a high-penetration new energy distribution network, comprising:

[0022] A quality scoring module is used to obtain a power quality scoring function based on multiple power quality indicators using a fuzzy hierarchical analysis method;

[0023] A network training module is used to train the association relationship using a teacher network containing a strong learner according to the grid structure of the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained through the power quality scoring function; the teacher network is used to perform knowledge distillation according to the association relationship between the power quality score and the grid structure to obtain the pros and cons distribution of different grid optimization schemes; the strong learner has a higher degree of learning of the training set and a higher degree of fitting of the empirical function than the weak learner;

[0024] A knowledge transfer module, used for transferring the trained probability distribution of the teacher network to the student network containing the weak learner;

[0025] The constraint optimization module is used to establish a comprehensive energy complementation model to represent the structure of a high-penetration new energy distribution network, and adopt a chaotic particle swarm algorithm to optimize the comprehensive energy complementation model with different parameters under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints, with the objective function of maximizing the power quality score function, to obtain multiple different high-penetration new energy distribution network structures;

[0026] The grid selection module is used to evaluate the multiple different high-penetration new energy distribution network structures using the student network, and take the high-penetration new energy distribution network structure with the highest power quality score as the optimized grid structure.

[0027] In a possible implementation of the second aspect, the multiple power quality indicators include voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance; and the power quality score is obtained by weighted summing up the scores and weights corresponding to each indicator.

[0028] In a possible implementation manner of the second aspect, the scoring solution process corresponding to each indicator specifically includes:

[0029] Establishing a hierarchical model, constructing a fuzzy priority relationship matrix, and then solving the fuzzy priority relationship matrix to obtain a fuzzy consistency matrix;

[0030] The fuzzy consistent matrix is ​​converted into a reciprocal judgment matrix, and the relative weight of each index is calculated by using a ranking method;

[0031] According to the membership function and the evaluation value corresponding to the membership function, the score corresponding to each indicator is obtained.

[0032] In a possible implementation manner of the second aspect, the network training module is specifically used to:

[0033] Acquire historical data from the cyber-physical system of the distribution network under high penetration of new energy, wherein the historical data includes various grid structures that have been in operation and corresponding power quality score data;

[0034] A teacher network containing a strong learner is trained according to the historical data, and knowledge distillation is performed on the probability distribution in the teacher network; the probability distribution is related to the distribution of the power quality scoring function and the grid structure.

[0035] In a possible implementation manner of the second aspect, the knowledge transfer module is specifically used to:

[0036] Under the condition that the function value of the loss function of the preset knowledge distillation is minimized, the probability distribution of the trained teacher network is imparted to the student network containing the weak learner.

[0037] Compared with the prior art, the grid structure optimization method and device of a high-penetration new energy distribution network provided by the embodiment of the present invention utilizes a chaotic particle swarm-knowledge distillation combination to optimize the grid structure, improve the power distribution path, and improve the power quality of the distribution network system under high-penetration new energy access. Since the optimization core of the student network comes from the probability distribution imparted by the teacher network, and the probability distribution of the teacher network is knowledge distilled, a higher response speed and estimation accuracy than ordinary traditional algorithms are obtained. Knowledge distillation is the compression of the relationship between the grid structure and the power quality score. The compressed probability distribution is imparted to the student network in the form of dark knowledge, so the iteration time is short, the execution efficiency is high, and a large amount of storage space is not required.

[0038] In addition, when using the chaotic particle swarm algorithm for grid optimization, the distribution network power score and other constraints are fully considered, and various power quality indicators are comprehensively considered to improve the reliability of grid optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of a grid structure optimization method of a high-penetration new energy distribution network provided by an embodiment of the present invention;

[0040] Figure 2 It is a flow chart of optimizing the grid structure of a distribution network in combination with a chaotic particle swarm algorithm in one embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of the process of a teacher network teaching a student network in one embodiment of the present invention. DETAILED DESCRIPTION

[0042] 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.

[0043] See also Figure 1 An embodiment of the present invention provides a grid structure optimization method for a high-penetration new energy distribution network, comprising:

[0044] S10. According to multiple power quality indicators, a power quality scoring function is obtained by using a fuzzy hierarchical analysis method.

[0045] S11. According to the grid structure of the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained by the power quality scoring function, the association relationship is trained using a teacher network containing a strong learner; the teacher network is used to perform knowledge distillation according to the association relationship between the power quality score and the grid structure to obtain the advantages and disadvantages distribution of different grid optimization schemes; the strong learner has a higher degree of learning of the training set and a higher degree of fitting of the empirical function than the weak learner.

[0046] S12, imparting the trained probability distribution of the teacher network to the student network containing the weak learner.

[0047] S13. Establish a comprehensive energy complementation model to represent the structure of a high-penetration new energy distribution network, and use a chaotic particle swarm algorithm to optimize the comprehensive energy complementation model with different parameters under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints, with the objective function of maximizing the power quality scoring function, to obtain multiple different high-penetration new energy distribution network structures.

[0048] S14. Evaluate the multiple different high-penetration new energy distribution network structures using the student network, and use the high-penetration new energy distribution network structure with the highest power quality score as the optimized grid structure.

[0049] In the embodiment of the present invention, various grid structures and power quality evaluation results of the distribution network that have been operated in the past are collected as sample test data, and the knowledge distillation algorithm is used to learn and train the compression model (about the grid structure and power quality scoring relationship model), and the teacher network with strong learners is trained according to the historical grid structure and power quality evaluation data. The teacher network can perform knowledge distillation based on the power quality evaluation result data to obtain the pros and cons distribution of different grid optimization schemes. After training the teacher network, the teacher network is integrated into the student network containing weak learners for training. See Figure 2 Finally, the student network can be combined with the chaotic particle swarm algorithm to solve and optimize the grid structure of the distribution network. It is worth noting that the chaotic particle swarm algorithm is only used as a solution method to maximize the value of the power quality score function under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints. The staff can combine other algorithms to solve according to actual needs.

[0050] The core ideas of the embodiments of the present invention are as follows:

[0051] (1) The power quality scoring function of the distribution network is established using the analytic hierarchy process:

[0052] A hierarchical model is established, a priority relationship matrix is ​​constructed, the fuzzy consistent matrix is ​​solved, the relative weights of factors are calculated, the membership function is constructed, and finally the power quality function is obtained.

[0053] In fact, the poor power quality in the distribution network includes indicators such as voltage deviation, harmonic voltage, voltage fluctuation, and three-phase imbalance. There are many bad factors for these power indicators, including three-phase load imbalance, insufficient reactive power compensation capacity at the end of the system, and many harmonic sources in the system. However, these indicators are also largely due to the unreasonable grid structure of the distribution network itself. If the grid structure of the distribution network can be optimized and improved, the voltage support of the terminal node can be increased, and the power supply path can be rationally distributed, then the power quality of the system can be improved.

[0054] (2) Distribution network reconstruction and optimization

[0055] A comprehensive energy complementarity model for high-penetration new energy distribution networks is established, and a mathematical model is used to express the multi-energy complementary equipment operated by multiple entities in high-penetration new energy distribution networks. At the same time, the results of the distribution network power quality assessment system are determined as the objective function to be maximized, and constraints such as branch capacity constraints, node voltage constraints, and network radial constraints are constructed.

[0056] The high-penetration new energy distribution network is centered on the electric energy network and is equipped with multiple energy complementary equipment including gas turbines, energy storage devices, distributed wind power generation, distributed photovoltaic cells and other equipment, which can realize the conversion of different forms of energy.

[0057] (a) Combined cooling, heating and power generation micro-gas turbine

[0058] The combined heat and power system with micro-gas turbines as the main energy supply equipment can improve the comprehensive utilization capacity of energy through the cascade utilization technology of energy, and improve the energy efficiency. The mathematical model of the combined heat and power type micro-gas turbine is shown as follows:

[0059]

[0060]

[0061]

[0062] Where Q MT (t), P MT (t), η e (t) are the residual heat, power output and operating efficiency of the micro-turbine during period t, η l is the heat dissipation rate, V MT is the amount of natural gas, C MT is the comprehensive fuel cost, LHV NG Indicates the lower calorific value of natural gas, P NG Represents the price of natural gas.

[0063] (b) Boiler

[0064] The mathematical model of the gas boiler can be expressed as shown in the following formula:

[0065] Q GB =R GB ×η GB

[0066]

[0067]

[0068] Where Q GB (t), R GB , η GB (t) are the boiler heat output, boiler rated thermal power and thermal efficiency in period t, V GB The amount of natural gas used by the boiler; C GB is the unit fuel cost coefficient.

[0069] (c) Energy storage device

[0070] The basic model of the battery is shown as follows:

[0071]

[0072] Where E ES(t), E ES (t-1) are the residual heat levels in period t and period t-1, respectively, P ES_ch , P ES_dis and η ch , η di s are the power and efficiency during period t respectively.

[0073] (d) Demand-side management devices

[0074] The integrated energy distribution network operator has signed a load shifting agreement with the user. When the distribution network power is in short supply, the user's load can be shifted and the user can be provided with necessary compensation. The shiftable load can only be shifted within a given period of time according to the agreement signed with the user, so there is a start-up time constraint as shown in the formula:

[0075] t dmin ≤t d ≤t dmax ;

[0076] Where t d is the starting time of the class D movable load; t dmin is the minimum translation time of the dth type of translatable load; t dmax It is the maximum translation time of the d-type translatable load.

[0077] The translation amount constraint of the translatable load is as follows:

[0078] x t→t ≥0

[0079]

[0080] Where x t is the number of users with load transferability at time t, x i→t is the number of users shifted from period i to period t.

[0081] In addition, the integrated energy distribution network is also connected to distributed wind turbines (WT), distributed photovoltaic cells (PV) and other equipment. The entire system exchanges power with the upper-level distribution network through a common coupling point.

[0082] Constraints generally include branch capacity constraints, node voltage constraints, and network radial constraints.

[0083] (a) Branch capacity constraints

[0084] The constraint is shown in the formula: l,max ≤P l ≤P l,max .

[0085] Where Pl represents the active power of the lth branch; P l,max Indicates the transmission upper limit of the lth branch.

[0086] (b) Node voltage constraint This constraint is shown in the equation, U i,min ≤U i ≤U i,max .

[0087] Where U i is the voltage amplitude of the ith node; U i,min and U i,max are the minimum and maximum voltage amplitudes of the ith node, respectively, as shown in the following formula:

[0088]

[0089]

[0090] Where n is the number of system nodes; P i and Q i is the active and reactive power of the ith node; G ij , B ij and δ ij are the conductance, susceptance and voltage phase difference of the branch connecting the i-th node and the j-th node respectively. i and U j are the voltage amplitudes of the i-th node and the j-th node respectively.

[0091] (c) Network topology constraints

[0092] The constraint is shown as follows: g∈G.

[0093] Where g is the system topology; G is the set of distribution network topologies that satisfy the network radial shape.

[0094] (3) Knowledge distillation algorithm and relational model solution

[0095] Collect various grid structures and power quality assessment results of distribution networks that have been in operation in the past as sample test data, import them into the knowledge distillation algorithm, and then train a teacher network with a strong learner based on the historical grid structure and power quality assessment data. The teacher network can perform knowledge distillation based on the power quality assessment result data to obtain the pros and cons distribution of different grid optimization schemes. After the teacher network training is completed, the teacher network is integrated into the student network containing weak learners for training to obtain the knowledge distillation algorithm training results between the distribution network grid structure and the power quality assessment results. At the same time, the chaotic particle swarm algorithm is used to optimize the grid structure of the distribution network, and finally the trained student network is called to evaluate the grid quality of the distribution network under different grid structures.

[0096] It should be noted that both the teacher network and the student network are an application of the knowledge distillation algorithm.

[0097] Compared with the prior art, the grid structure optimization method of a high-penetration new energy distribution network provided by the embodiment of the present invention uses a chaotic particle swarm-knowledge distillation combination to optimize the grid structure, improve the power distribution path, and improve the power quality of the distribution network system under high-penetration new energy access. Since the optimization core of the student network comes from the probability distribution imparted by the teacher network, and the probability distribution of the teacher network is knowledge distilled, a higher response speed and estimation accuracy than ordinary traditional algorithms are obtained. Knowledge distillation is the compression of the relationship between the grid structure and the power quality score. The compressed probability distribution is imparted to the student network in the form of dark knowledge, so the iteration time is short, the execution efficiency is high, and a large amount of storage space is not required.

[0098] In addition, when using the chaotic particle swarm algorithm for grid optimization, the distribution network power score and other constraints are fully considered, and various power quality indicators are comprehensively considered to improve the reliability of grid optimization.

[0099] Exemplarily, the multiple power quality indicators include voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance; the power quality score is obtained by weighted summing the scores and weights corresponding to each indicator.

[0100] Exemplarily, the scoring solution process corresponding to each indicator specifically includes:

[0101] Solving the constructed fuzzy priority relation matrix, we can get the fuzzy consistent matrix;

[0102] The fuzzy consistent matrix is ​​converted into a reciprocal judgment matrix, and the relative weight of each index is calculated by using a ranking method;

[0103] According to the membership function and the evaluation value corresponding to the membership function, the score corresponding to each indicator is obtained.

[0104] The steps of establishing a power quality scoring function using the hierarchical analysis method in the embodiment of the present invention are as follows:

[0105] a. Establish a hierarchical model:

[0106] Taking into account the four indicators of voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance of the distribution network, a power quality assessment system is established, and each factor is refined into factors at the criterion level.

[0107] b. Construct a priority relationship matrix:

[0108] Construct the fuzzy priority relation matrix R = (r ij ) n×nTo evaluate the importance relationship between various factors. The properties of this matrix are as follows:

[0109]

[0110] Where r ii is the importance relationship between the ith factor and itself; r ij and is the importance relationship between the i-th factor and the j-th factor; r ji is the importance relationship between the jth factor and the ith factor. The 0.1-0.9 degree method is used to define the meaning of the coefficient range, that is, if r ii =0.5 means that both are equally important; if r ij ∈[0.1,0.5], which means r j R i Important; if ij ∈(0.5,0.9), then it means r i R j Important. Because it involves four indicators, the maximum value of n is 4.

[0111] c. Solve to obtain the fuzzy consistent matrix:

[0112] For the fuzzy priority relation matrix R = (r ij ) n×n The sum of each row of Perform mathematical transformations of the following form: rij =(r i -r j ) / 2n+0.5, and get the fuzzy consistent matrix.

[0113] d. Calculation of relative weights of factors:

[0114] The relative weight of the factors is calculated by ranking method: R = (r ij ) n×n Converted to the reciprocal judgment matrix R = (e ij ) n×n , where e ij =r ij / r ji .

[0115] e. Construct membership function:

[0116] A triangle is selected to construct the membership function, the mathematical form of which is:

[0117]

[0118]

[0119]

[0120] It should be noted that what is described here is the membership degree of the four major indicators to the target decision-making. The setting of the three interval values a1, a2, and a3 is determined by the expert scoring method, with the given principle of subjective weight. The membership function represents the weight of each evaluation indicator for the power quality score.

[0121] f. At the same time, based on the membership function, obtain the final evaluation results of each index factor:

[0122] In the formula, F1, F2, and F3 are the evaluation values.

[0123] g. Comprehensive evaluation:

[0124] Evaluate the target evaluation layer of the power quality assessment system for the distribution network, realize the comprehensive evaluation of the power quality status, and obtain the evaluation result:

[0125]

[0126] In the formula, S is the score of the non-target layer index; n (n>1) is the number of index factor layers under the index S; S j is the score of the index j (1<j<n); W j is the weight of the index j. The target layer refers to the evaluation of the four major indicators, and the non-target layer index refers to the comprehensive score of the power quality status, that is, the power quality scoring function.

[0127] Exemplarily, the grid structure according to the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure are obtained through the power quality scoring function, and the correlation relationship is trained by using a teacher network containing a strong learner, specifically including:

[0128] Obtain historical data from the cyber-physical system of the distribution network under high-penetration new energy access. The historical data includes various grid structures that have been operated and the corresponding power quality score data;

[0129] Train a teacher network containing a strong learner according to the historical data, and perform knowledge distillation on the probability distribution in the teacher network; the probability distribution is related to the distribution of the power quality scoring function and the grid structure.

[0130] Please refer to Figure 3 , train the teacher network according to the existing historical operation data, where the teacher network is constructed by a strong classifier. This reinforcement learner considers using 2 convolutional neural networks CNN and 2 fully connected neural networks, connected by the ReLU function in the middle. The trained teacher network is Teacher-Net.

[0131] Next, we define the student network Student-Net. The student network consists of three fully connected layer networks, which is a weaker learner than the teacher network.

[0132] Since the dark knowledge in the teacher network varies according to the value of temperature T, the deformed softmax formula is as follows:

[0133] The higher the temperature coefficient, the softer the distribution generated by the softmax function. In the training phase, the temperature coefficient T is first increased, and in the test phase, the temperature coefficient T is reduced to 1, so that the knowledge in the original model (in this embodiment, the original model consists of a power quality scoring function and multiple constraints) is extracted to achieve knowledge distillation. That is, the original data only contains the label of the grid type. In actual engineering, a certain measurement result may not necessarily be a certain power quality assessment result. A more accurate description should be to describe this power quality assessment result in a probability distribution. This evaluation method distribution also contains some prior knowledge. Extracting and applying it can improve the accuracy of grid optimization based on power quality assessment results to a certain extent. At the same time, because the model is smaller, it can be applied to online optimization and quickly achieve the optimal analysis of grid types.

[0134] Exemplarily, imparting the trained probability distribution of the teacher network to the student network containing the weak learner specifically includes:

[0135] Under the condition that the function value of the loss function of the preset knowledge distillation is minimized, the probability distribution of the trained teacher network is imparted to the student network containing the weak learner.

[0136] Use the dark knowledge (probability distribution) of the teacher network Teacher-Net to teach the student network Student-Net for training and prediction. In particular, it is necessary to define the loss function of knowledge distillation:

[0137]

[0138] In the formula, represents the predicted value, y represents the true value, and λ is the weighting factor. represents the loss of the student network for prediction under discrete labels, It represents the loss of prediction based on knowledge distillation taught by the teacher network to the student network. The ultimate goal of training is to minimize the above loss value.

[0139] After training, when the distribution network structure containing distributed power generation and the corresponding distribution network are considered, the grid can be optimized based on the student network that contains the knowledge imparted by the teacher. The student network is a weak learner that can efficiently complete online grid classification optimization.

[0140] The embodiment of the present invention aims at improving the power quality of the distribution network, and constructs a model by using the grid structure optimization theory and the knowledge distillation algorithm. The established model first proposes a distribution network power quality assessment model based on the fuzzy analytic hierarchy process, then establishes a system grid structure reconstruction model (a relationship model between power quality score and grid structure) with the goal of optimizing the power quality index of the distribution network, and finally uses the knowledge distillation algorithm to design a solution process for the established model, which can significantly reduce the voltage deviation percentage, three-phase imbalance and voltage deviation of the distribution network, thereby improving the power quality assessment index, and has a good improvement effect.

[0141] An embodiment of the present application provides a grid structure optimization device for a high-penetration new energy distribution network, including: a quality scoring module, a network training module, a knowledge transfer module, a constraint optimization module and a grid selection module.

[0142] The quality scoring module is used to obtain a power quality scoring function based on multiple power quality indicators using a fuzzy hierarchical analysis method.

[0143] A network training module is used to train the association relationship using a teacher network containing a strong learner based on the grid structure of historical operating data of a high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained through the power quality scoring function; the teacher network is used to perform knowledge distillation based on the association relationship between the power quality score and the grid structure to obtain the pros and cons distribution of different grid optimization schemes; the strong learner has a higher degree of learning of the training set and a higher degree of fitting of the empirical function than the weak learner.

[0144] The knowledge transfer module is used to transfer the probability distribution of the trained teacher network to the student network containing the weak learner.

[0145] The constraint optimization module is used to establish a comprehensive energy complementation model to represent the structure of a high-penetration new energy distribution network, and adopt a chaotic particle swarm algorithm to optimize the comprehensive energy complementation model with different parameters under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints, with the objective function of maximizing the power quality score function, to obtain multiple different high-penetration new energy distribution network structures;

[0146] The grid selection module is used to evaluate the multiple different high-penetration new energy distribution network structures using the student network, and take the high-penetration new energy distribution network structure with the highest power quality score as the optimized grid structure.

[0147] Exemplarily, the multiple power quality indicators include voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance; the power quality score is obtained by weighted summing the scores and weights corresponding to each indicator.

[0148] Exemplarily, the scoring solution process corresponding to each indicator specifically includes:

[0149] Establishing a hierarchical model, constructing a fuzzy priority relationship matrix, and then solving the fuzzy priority relationship matrix to obtain a fuzzy consistency matrix;

[0150] The fuzzy consistent matrix is ​​converted into a reciprocal judgment matrix, and the relative weight of each index is calculated by using a ranking method;

[0151] According to the membership function and the evaluation value corresponding to the membership function, the score corresponding to each indicator is obtained.

[0152] Exemplarily, the network training module is specifically used for:

[0153] Acquire historical data from the cyber-physical system of the distribution network under high penetration of new energy, wherein the historical data includes various grid structures that have been in operation and corresponding power quality score data;

[0154] A teacher network containing a strong learner is trained according to the historical data, and knowledge distillation is performed on the probability distribution in the teacher network; the probability distribution is related to the distribution of the power quality scoring function and the grid structure.

[0155] Exemplarily, the knowledge transfer module is specifically used for:

[0156] Under the condition that the function value of the loss function of the preset knowledge distillation is minimized, the probability distribution of the trained teacher network is imparted to the student network containing the weak learner.

[0157] Compared with the prior art, the embodiment of the present invention provides a grid structure optimization device for a high-penetration new energy distribution network.

[0158] The chaotic particle swarm-knowledge distillation combination is used to optimize the grid structure, improve the power distribution path, and improve the power quality of the distribution network system under high penetration of new energy. Since the optimization core of the student network comes from the probability distribution taught by the teacher network, and the probability distribution of the teacher network is distilled through knowledge, it has a higher response speed and estimation accuracy than ordinary traditional algorithms. Knowledge distillation is the compression of the relationship between the grid structure and the power quality score. The compressed probability distribution is taught to the student network in the form of dark knowledge, so the iteration time is short, the execution efficiency is high, and it does not require a lot of storage space.

[0159] In addition, when using the chaotic particle swarm algorithm for grid optimization, the distribution network power score and other constraints are fully considered, and various power quality indicators are comprehensively considered to improve the reliability of grid optimization.

[0160] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be described again here.

[0161] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A grid structure optimization method for a high-penetration new energy distribution network, characterized in that: include: According to multiple power quality indicators, the power quality scoring function is obtained by using fuzzy analytic hierarchy process. According to the grid structure of the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained by the power quality scoring function, the association relationship is trained using a teacher network containing a strong learner; the teacher network is used to perform knowledge distillation according to the association relationship between the power quality score and the grid structure to obtain the pros and cons distribution of different grid optimization schemes; the strong learner has a higher degree of learning of the training set and a higher degree of fitting of the empirical function than the weak learner; imparting the trained probability distribution of the teacher network to the student network containing the weak learner; A comprehensive energy complementation model is established to represent the structure of a high-penetration new energy distribution network, and a chaotic particle swarm algorithm is used to optimize the comprehensive energy complementation model with different parameters under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints, with the objective function of maximizing the power quality score function, to obtain multiple different high-penetration new energy distribution network structures; The student network is used to evaluate the multiple different high-penetration new energy distribution network structures, and the high-penetration new energy distribution network structure with the highest power quality score is used as the optimized grid structure.

2. The grid structure optimization method of the high-penetration new energy distribution network according to claim 1 is characterized in that: The multiple power quality indicators include voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance; the power quality score is obtained by weighted summing the scores and weights corresponding to each indicator.

3. The grid structure optimization method of a high-penetration new energy distribution network as claimed in claim 2, characterized in that: The scoring solution process corresponding to each indicator specifically includes: Establishing a hierarchical model, constructing a fuzzy priority relationship matrix, and then solving the fuzzy priority relationship matrix to obtain a fuzzy consistency matrix; The fuzzy consistent matrix is ​​converted into a reciprocal judgment matrix, and the relative weight of each index is calculated by using a ranking method; According to the membership function and the evaluation value corresponding to the membership function, the score corresponding to each indicator is obtained.

4. The grid structure optimization method of a high-penetration new energy distribution network according to claim 1, characterized in that: The grid structure according to the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained by the power quality score function, and the teacher network containing a strong learner are used to train the association relationship, specifically including: Acquire historical data from the cyber-physical system of the distribution network under high penetration of new energy, wherein the historical data includes various grid structures that have been in operation and corresponding power quality score data; A teacher network containing a strong learner is trained according to the historical data, and knowledge distillation is performed on the probability distribution in the teacher network; the probability distribution is related to the distribution of power quality scores and grid structures.

5. The grid structure optimization method of a high-penetration new energy distribution network according to claim 1, characterized in that: The method of imparting the trained probability distribution of the teacher network to the student network containing the weak learner specifically includes: Under the condition that the function value of the loss function of the preset knowledge distillation is minimized, the probability distribution of the trained teacher network is imparted to the student network containing the weak learner.

6. A grid structure optimization device for a high-penetration new energy distribution network, characterized in that: include: A quality scoring module is used to obtain a power quality scoring function based on multiple power quality indicators using a fuzzy hierarchical analysis method; A network training module is used to train the association relationship using a teacher network containing a strong learner according to the grid structure of the historical operation data of the high-penetration new energy distribution network and the power quality score corresponding to the grid structure obtained through the power quality scoring function; the teacher network is used to perform knowledge distillation according to the association relationship between the power quality score and the grid structure to obtain the pros and cons distribution of different grid optimization schemes; the strong learner has a higher degree of learning of the training set and a higher degree of fitting of the empirical function than the weak learner; A knowledge transfer module, used for transferring the trained probability distribution of the teacher network to the student network containing the weak learner; The constraint optimization module is used to establish a comprehensive energy complementation model to represent the structure of a high-penetration new energy distribution network, and adopt a chaotic particle swarm algorithm to optimize the comprehensive energy complementation model with different parameters under the constraints of branch capacity constraints, node voltage constraints and network radiation constraints, with the objective function of maximizing the power quality score function, to obtain multiple different high-penetration new energy distribution network structures; The grid selection module is used to evaluate the multiple different high-penetration new energy distribution network structures using the student network, and take the high-penetration new energy distribution network structure with the highest power quality score as the optimized grid structure.

7. The grid structure optimization device for a high-penetration new energy distribution network according to claim 6, characterized in that: The multiple power quality indicators include voltage deviation, harmonic voltage, voltage fluctuation and three-phase imbalance; the power quality score is obtained by weighted summing the scores and weights corresponding to each indicator.

8. The grid structure optimization device for a high-penetration new energy distribution network as claimed in claim 7, characterized in that: The scoring solution process corresponding to each indicator specifically includes: Establishing a hierarchical model, constructing a fuzzy priority relationship matrix, and then solving the fuzzy priority relationship matrix to obtain a fuzzy consistency matrix; The fuzzy consistent matrix is ​​converted into a reciprocal judgment matrix, and the relative weight of each index is calculated by using a ranking method; According to the membership function and the evaluation value corresponding to the membership function, the score corresponding to each indicator is obtained.

9. The grid structure optimization device for a high-penetration new energy distribution network according to claim 6, characterized in that: The network training module is specifically used for: Acquire historical data from the cyber-physical system of the distribution network under high penetration of new energy, wherein the historical data includes various grid structures that have been in operation and corresponding power quality score data; A teacher network containing a strong learner is trained according to the historical data, and knowledge distillation is performed on the probability distribution in the teacher network; the probability distribution is related to the power quality scoring function and the distribution of the grid structure.

10. The grid structure optimization device for a high-penetration new energy distribution network according to claim 6, characterized in that: The knowledge transfer module is specifically used for: Under the condition that the function value of the loss function of the preset knowledge distillation is minimized, the probability distribution of the trained teacher network is imparted to the student network containing the weak learner.

Images