Micro-energy network group division method based on multi-objective optimization

Abstract

The invention relates to a micro-energy network group division method based on multi-objective optimization, which considers the multi-objective optimization requirement in the micro-energy network group division process, introduces the idea of Pareto optimization into a generative adversarial network for different optimization objectives such as energy transmission cost, carbon emission, energy storage efficiency and sub-network internal balance degree, and improves the optimization efficiency of the micro-energy network group. And the targets can be mutually balanced and compromised in the optimization process. In the generation process of the generator, the weight of each target is dynamically adjusted, and the discriminator scores the generation scheme through the Euclidean distance, feeds back the advantages and disadvantages of the scheme and forms loop iteration, so that the generated division scheme gradually approaches the Pareto optimal leading edge, thereby improving the efficiency of multi-target optimization and the comprehensive performance of the division result. Through the dynamic updating of the Pareto leading edge and the adaptive learning mechanism of the GAN, a high-quality division scheme meeting the multi-target compromise requirement is automatically generated, and the rationality and flexibility of micro-energy network group division are remarkably improved.

Description

Technical Field

[0001] The invention relates to a micro energy grid group division method based on multi-objective optimization, and belongs to the field of micro grid group division. Background Art

[0002] With the widespread application of renewable energy, micro-energy grids are introduced as local energy management and supply units with the ability to operate independently. Due to the dispersed distribution and small scale of micro-energy grids, it is necessary to aggregate adjacent or interdependent micro-grids into micro-grid groups to achieve power coordination, resource sharing and energy scheduling within the region. By dividing micro-energy grid groups, cross-regional power transmission losses can be reduced, the stability and self-healing ability of the system can be enhanced, and energy utilization can be improved. In the context of large-scale renewable energy access, the reasonable division of micro-energy grid groups can help cope with load fluctuations, improve the economy and flexibility of the system, and effectively cope with the complex operating environment of the power system.

[0003] At present, there are still some deficiencies in the partitioning methods of micro-energy grid groups in terms of multi-objective optimization, flexibility and accuracy. First, the multi-objective conflicts in energy transmission are not fully considered. The trade-offs between different objectives such as energy transmission cost, energy storage efficiency, and carbon emissions may lead to deviations in the solution set and affect the overall optimization effect. Second, there is a lack of dynamic adjustment of the weights of different optimization objectives, and it is impossible to respond to the current changes in system demand in real time, and thus it is impossible to respond quickly to the partitioning results. Third, traditional evaluation methods often ignore the diversity of solution sets and the balance of multiple objectives when evaluating partitioning results, and are prone to falling into local optimality, making it difficult to ensure that the partitioning scheme has global optimal characteristics in the multi-objective space. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of poor multi-objective balance in the prior art and to provide a micro energy grid group division method with good multi-objective balance.

[0005] In order to achieve the above object, the technical solution proposed by the present invention is: a micro energy grid group division method based on multi-objective optimization, comprising the following steps:

[0006] A micro-energy grid cluster model is constructed with the objective functions of minimizing energy transmission costs, carbon emissions, and internal imbalance of subgrids, as well as maximizing the utilization efficiency of energy storage devices;

[0007] Based on the micro-energy grid cluster model and the Pareto frontier solution set, a generative adversarial network including a generator and a discriminator is constructed;

[0008] The generator uses a dynamic weight mechanism to adjust the weight of the objective function, construct a weighted objective function, and generate a partitioning scheme;

[0009] The discriminator calculates the distance between the partition scheme and the Pareto frontier solution, determines whether the partition scheme is a non-dominated solution, scores the partition scheme, and feeds back the pros and cons of the partition scheme to the generator;

[0010] Iterate the generator and discriminator, and dynamically update the Pareto frontier solution set based on the non-dominated solution during the iteration process; when the preset convergence condition is reached, the optimal Pareto frontier solution set is output;

[0011] The optimal Pareto front solution set is evaluated, and the optimal partitioning scheme is selected based on the comprehensive performance.

[0012] The further design of the above technical solution is as follows: constructing a micro energy grid cluster model includes the following steps:

[0013] S11. Each micro energy grid is defined as a node. The characteristics of node i are expressed as: the power generation of node i at time t P gen，i (t); the state E of the energy storage device of node i at time t stor，i (t); Power demand P of node i load，i (t); Carbon emissions E of node i at time t carbon,i (t);

[0014] S12. Define the energy transfer between different nodes as an edge. The characteristic of edge (i, j) is expressed as: the energy P transmitted from node i to node j trans，ij (t); Energy loss L when transmitting from node i to node j ij ; The cost of transmitting energy from node i to node j is C ij (t);

[0015] S13. Define the objective function, including energy transmission cost, carbon emission, utilization efficiency of energy storage equipment and internal balance of subnet, which are expressed as:

[0016]

[0017] In the formula, f1(x), f2(x), f3(x), and f4(x) are energy transmission cost, carbon emission, utilization efficiency of energy storage equipment, and internal imbalance of the subgrid, respectively; is a node set, each node i represents a micro energy network; ε is an edge set, each edge (i, j) represents the energy transmission path between node i and node j; C ij (t) is the cost of transmitting energy from node i to node j at time t; E i (t) is the carbon emission of node i at time t; η i (t) is the energy storage device utilization rate of node i at time t; B i (t) is the imbalance degree of node i at time t;

[0018] The optimization goals are to minimize energy transmission costs, carbon emissions, internal imbalance of subgrids and maximize the utilization efficiency of energy storage equipment;

[0019] S14. The supply and demand balance constraint and energy storage equipment capacity constraint of the objective function are respectively expressed as:

[0020] P gen，i (t)+P stor-out，i (t)+P in，i (t) = P load，i (t)+P stor-in，i (t)

[0021] 0≤E stor，i (t)≤E stor-max，i

[0022] Among them, P stor-out，i (t) represents the power (discharge power) released from the energy storage device by node i at time t; P in，i (t) represents the input power obtained by node i from other nodes at time t; P stor-in，i (t) represents the charging power of the energy storage device at node i at time t; E stor-max，i Indicates the maximum storage capacity of the energy storage device at node i.

[0023] The weighted objective function is:

[0024]

[0025] In the formula, f i (x) is the i-th objective function, which is energy transmission cost, carbon emission, energy storage efficiency or internal balance of subgrid; λ i (t) is the weight after dynamic adjustment at time step t.

[0026] The process of adjusting weights using the dynamic weight mechanism is as follows:

[0027] S21, when the generator starts training, the weights λ of all targets i (0) Set to equal;

[0028] S22. Define the dynamic weight adjustment rule as:

[0029]

[0030] Among them, fi(x) is the specific value of the solution x provided by the discriminator for the i-th target, and the weight λ i (t+1) is the weight of the next iteration;

[0031] S23, normalized dynamic weight, the normalization formula is:

[0032]

[0033] The Euclidean distance between the discriminator calculation generation solution and the Pareto front solution is:

[0034]

[0035] In the formula, f i (x) is the performance of solution x on the i-th objective; f i (p) is the performance of the Pareto frontier solution p on the i-th objective; n represents the number of objectives;

[0036] Select the solution x and all Pareto frontier solutions p∈P optimal The smallest Euclidean distance among them is taken as the distance of the solution.

[0037] The discriminator scores the partitioning scheme as follows:

[0038]

[0039] Where D(x) is the score of the discriminator on the generated solution x; d(x, P optimal ) is the distance between solution x and the Pareto optimal solution frontier.

[0040] The loss function of the discriminator is:

[0041]

[0042] Among them, D(z high,i ) represents the high quality score of the discriminator on the generated solution on the i-th target; D(z low,i ) represents the low quality score of the generated solution on the i-th target by the discriminator, and the discriminator feeds back to the generator the specific value f of the low quality solution x on each target i (x).

[0043] The dynamic updating of the Pareto frontier solution set comprises the following steps:

[0044] S31, the generator generates a new solution x;

[0045] S32, the discriminator receives the solution x generated by the generator and determines whether it dominates or is dominated by the existing Pareto frontier solution set;

[0046] S33. For solution x and frontier solution p, the condition that solution x dominates frontier solution p is:

[0047] and

[0048] If solution x dominates the solutions in the existing Pareto frontier, the dominated solutions are eliminated and solution x is added to the Pareto frontier solution set; if solution x does not dominate any solutions in the existing Pareto frontier, the discriminator regards x as a non-dominated solution and adds it to the Pareto frontier solution set.

[0049] The evaluation of the optimal Pareto frontier solution set includes:

[0050] S41. Preliminary screening based on hypervolume contribution:

[0051] r j =max(f j (x))+∈

[0052]

[0053] Where, ∈ is a positive number; V(x i ) is the solution x i In each target dimension, the j The difference between the two forms a hyperrectangular volume, V total Add the volumes of each solution together to get the total hypervolume;

[0054] Eliminate solutions whose marginal contribution is less than the threshold α;

[0055] S42, secondary screening based on diversity evaluation; use Euclidean distance to measure the differences between adjacent solutions. Suppose two solutions x i and x i+1 The target value of i )=[f1(x i ), f2(x i ), ..., f n (x i )] and f(x i+1 )=[f1(x i+1 ), f2(x i+1 ), ..., f n (x i+1 )], its Euclidean distance d(x i , x i+1 ) is defined as:

[0056]

[0057] Among them, d(x i , x i+1 ) is the solution x i and adjacent solutions x i+1 The Euclidean distance between

[0058] Eliminate solutions x whose distance to adjacent solutions is less than the threshold δ i ;

[0059] S53. Select the best solution based on comprehensive performance:

[0060]

[0061] In the formula, S(x i ) is the comprehensive score; ω1 and ω2 are weight coefficients;

[0062] Select the comprehensive score S(x i )The highest solution x * is the optimal solution.

[0063] The beneficial effects of the present invention are:

[0064] First, the present invention introduces the definition of nodes and edges in the modeling process, describes the characteristics and energy flow relationship of each micro-energy network in a refined manner, and transforms the micro-energy network group division problem into a multi-objective optimization problem. Then, the Pareto optimal frontier method is integrated into the construction and training process of GAN. The generator flexibly adjusts the weight of the objective function through discriminator feedback and introduces a dynamic adjustment mechanism. The discriminator scores the generated solution based on the Euclidean distance and feeds back the pros and cons of the solution to the generator, forming a loop iteration, so that the system gradually approaches the Pareto optimal frontier solution set, achieves a good balance in multiple objectives such as energy transmission cost, carbon emissions and energy storage utilization, and introduces a non-dominated solution priority selection mechanism, which significantly improves the optimization efficiency. Finally, the Pareto optimal frontier solution set is multiple-screened through hypervolume contribution and diversity evaluation, the scale of the solution set is reduced, the final solution set is ensured to be more representative and diverse, and the optimal solution is selected based on comprehensive performance; thereby improving the efficiency of multi-objective optimization and the comprehensive performance of the division results. The present invention automatically generates a high-quality partitioning scheme that meets the multi-objective compromise requirements through the dynamic update of the Pareto frontier and the adaptive learning mechanism of GAN, which significantly improves the rationality and flexibility of the micro-energy grid group partitioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flow chart of the method of the present invention.

[0066] Figure 2 This is a flow chart of the objective function of the dynamic weight adjustment generator introduced in the present invention.

[0067] Figure 3 It is a flow chart of the design of the discriminator and the selection of non-dominated solutions of the present invention.

[0068] Figure 4 It is a dynamic update flow chart of the Pareto frontier of the present invention.

[0069] Figure 5It is the evaluation and optimization flow chart of the Pareto frontier of the present invention.

[0070] Figure 6 Schematic diagram of the division scheme of micro energy grid cluster. DETAILED DESCRIPTION

[0071] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] Embodiment 1

[0073] like Figure 1 As shown, this embodiment provides a micro energy grid group division method based on multi-objective optimization, including the following steps:

[0074] S1. The modeling of micro energy grid cluster includes the following steps:

[0075] S11. Define nodes. Each node represents a micro energy grid, and the node characteristics are used to describe its power generation capacity, energy storage status, load demand and carbon emissions. The characteristics of node i can be expressed as: Power generation capacity: P gen，i (t), the power generation of node i at time t; energy storage state: E stor，i (t), the state of the energy storage device at node i at time t; load demand: P load,i (t), electricity demand of node i; carbon emission: E carbon，i (t), the carbon emissions of node i at time t.

[0076] S12. Define edges. Edges represent energy transfer between different nodes. The characteristics of edge (i, j) can be expressed as: Energy transfer amount: P trans，ij (t), the energy transmitted from node i to node j; energy transmission loss: L ij , the energy loss when transmitting from node i to node j; energy transmission cost: C ij (t), the cost of transmitting energy from node i to node j.

[0077] S13. Define the objective function. In order to achieve the optimal partitioning of the system, the goal is to optimize the operation of the micro-energy grid group in multiple dimensions, including energy transmission cost, carbon emissions, utilization efficiency of energy storage equipment and internal balance of the subnet, which are expressed as:

[0078]

[0079] In the formula, f1(x), f2(x), f3(x), and f4(x) are energy transmission cost, carbon emission, utilization efficiency of energy storage equipment, and internal imbalance of subgrid, respectively. The purpose is to minimize energy transmission cost, carbon emission, internal imbalance of subgrid, and maximize utilization efficiency of energy storage equipment. is a node set, each node i represents a micro energy network; ε is an edge set, each edge (i, j) represents the energy transmission path between node i and node j; C ij (t) is the cost of transmitting energy from node i to node j at time t; E i (t) is the carbon emission of node i at time t; η i (t) is the energy storage device utilization rate of node i at time t; B i (t) is the imbalance degree of node i at time t, reflecting the difference between power generation and load demand.

[0080] S14, introduce constraints. In order to ensure the normal operation of the system, the objective function needs to follow the supply and demand balance constraint and the energy storage device capacity constraint to ensure that the power generation, energy storage and energy obtained from other nodes of each node meet its load demand and the charge and discharge of the energy storage device cannot exceed the maximum capacity of the device. The supply and demand balance constraint and the energy storage device capacity constraint are expressed as:

[0081] P gen，i (t)+P stor-out，i (t)+P in，i (t) = P 1oad，i (t)+P stor-in，i (t)

[0082] 0≤E stor，i (t)≤E stor-max，i .

[0083] S2. Construct a generative adversarial network (GAN) including a generator and a discriminator;

[0084] S21. Construct a generator that introduces dynamic weights to adjust the weights of the objective function, such as Figure 2 As shown, the following steps are included:

[0085] S211. Build a generator model. The generator G is a deep neural network whose input is a random noise vector z and whose output is a fake data sample that is as close to the real data as possible.

[0086] S212, define a dynamic weighted objective function. Each time a partitioning scheme is generated, the objective function of the generator is not a fixed linear weight, but is dynamically adjusted over time and during the optimization process. The objective function is in the form of:

[0087]

[0088] In the formula, f i (x) is the i-th objective function, representing the objectives of different optimization directions, such as energy transmission cost, carbon emissions, energy storage efficiency, and subnet internal balance; λ i (t) is the weight after dynamic adjustment at time step t;

[0089] S213, weight initialization. When the generator starts training, the weights λ of all targets i (0) Set to equal;

[0090] S214. Define a dynamic weight adjustment strategy. Each target weight λ i The update of (t) is based on the performance of the target in the current generated solution x, and the discriminator provides the specific value f of the current solution x on each target. i (x), the weight adjustment rule is:

[0091]

[0092] When f i (x) is large (i.e., the target performs poorly), and the weight λ i (t+1) is increased in the next step to make the generator pay more attention to this goal; when f i (x) is small (i.e. the objective has been optimized well), and the weight λ i (t+1) will be appropriately reduced, so that the generator pays less attention to this goal and avoids this goal occupying too many resources in subsequent optimization;

[0093] S215, normalize dynamic weights. In order to prevent the weights from getting out of control during the adjustment process, all weights will be normalized after each adjustment, so that the contribution of each target weight in the system always maintains a reasonable proportion. The normalization formula is:

[0094]

[0095] Through normalization, the system can ensure that the relative weights between all objectives are always within a controllable range, preventing a certain objective from being infinitely magnified or ignored during the optimization process.

[0096] S22. Construct a discriminator that introduces a priority selection mechanism for non-dominated solutions, such as Figure 3 As shown, the following steps are included:

[0097] S221. Construct a discriminator model. The discriminator D is also a deep neural network whose task is to evaluate the authenticity of data samples.

[0098] S222, calculate the distance d(x, P) between the generator generated solution x and the Pareto frontier solution optimal ). Evaluate the gap between solution x and the currently known Pareto frontier solution, and then judge the pros and cons of the solution in the multi-objective space. For the performance of solution x in the multi-objective space [f1(x), f2(x), ..., f n (x)], and the Pareto frontier solution p∈P optimalThe performance of [f1(p), f2(p), ..., f n (p)], the Euclidean distance between them is calculated as:

[0099]

[0100] In the formula, f i (x) is the performance of solution x on the ith objective; f i (p) is the performance of the Pareto frontier solution p on the i-th objective; n represents the number of objectives;

[0101] In order to obtain the solution x and the Pareto frontier solution set P optimal distance, select x and all Pareto frontier solutions p∈P optimal The smallest Euclidean distance is taken as the distance of the solution:

[0102]

[0103] S223. Define a priority selection mechanism for non-dominated solutions. In multi-objective optimization, a non-dominated solution refers to a solution that is not dominated by other solutions on multiple objectives. Specifically, each time the discriminator receives a solution generated by the generator, it evaluates whether the solution is a non-dominated solution. If the solution x generated by the generator is better than or equal to other generated solutions on all objectives and better than other solutions on at least one objective, the solution is determined by the discriminator to be a non-dominated solution. The discriminator scores whether the generated solution is a non-dominated solution according to the following formula:

[0104]

[0105] Where D(x) is the score of the discriminator on the generated solution x; d(x, P optimal ) is the distance between solution x and the Pareto optimal solution frontier. If the distance is small, it means that the solution is close to the Pareto frontier; P optimal is the set of Pareto frontier solutions recorded in history. Through this non-dominated solution priority selection mechanism, the discriminator can identify and prioritize those generation solutions that are close to the Pareto frontier. Non-dominated solutions have higher priority in the system optimization process, so the discriminator will give them higher scores;

[0106] S224. Define the optimization and feedback mechanism of the discriminator. The task of the discriminator is not only to evaluate whether the generated solution is close to the actual partition result, but also to guide the generator to continuously generate solutions that meet the requirements of multi-objective optimization. Set the threshold to 0.8. If D(x) is greater than or equal to 0.8, it is judged as a high-quality solution with a high-quality score. If it is less than 0.8, it is judged as a low-quality solution with a low-quality score. By introducing the priority scoring mechanism for non-dominated solutions, the loss function of the discriminator is designed as follows:

[0107]

[0108] D(z high，i ) indicates that the discriminator scores the high quality performance of the generated solution on the i-th target, when z high,i is a high-quality solution close to the Pareto frontier, the score D(z high,i ) is close to 1, the discriminator can better identify high-quality solutions; D(z low,i ) is the score for low-quality solutions. low,i is a low-quality solution, the score D(z low,i ) is close to 0, the discriminator can correctly identify low-quality solutions and feed them back to the generator for adjustment. When the discriminator score is low, the discriminator also provides the generator with the specific value f of the solution x on each target i (x) is used to measure the performance of the solution on each objective. The generator analyzes the specific value of each objective in detail and increases the attention to the objectives with poor performance by dynamically adjusting the strategy. Through the priority scoring mechanism for non-dominated solutions, the discriminator can give targeted feedback to help the generator gradually approach the Pareto frontier in the multi-objective optimization process.

[0109] S3, dynamic update of Pareto frontier, such as Figure 4 As shown, the following steps are included:

[0110] S31, the generator generates a new solution x. The task of the generator is to generate a new partitioning scheme x in the multi-objective space, and gradually adjust the generated scheme through the feedback of the discriminator, giving priority to generating the scheme that is closer to the Pareto frontier. The generator does not know whether the current solution is the optimal solution, it depends on the feedback of the discriminator;

[0111] S32, the discriminator determines the dominance relationship. The discriminator receives the solution x generated by the generator and determines whether it dominates or is dominated by the existing Pareto frontier solution set. The discriminator determines whether the new solution can be added to the Pareto frontier solution set based on the dominance relationship, and feeds back the pros and cons of the solution to the generator to guide the generator to adjust towards the optimal solution.

[0112] S33. Record and update the Pareto frontier solution set. For two solutions x and p, the condition for solution x to dominate solution p is:

[0113] and

[0114] If solution x dominates some solutions in the existing Pareto frontier, these dominated solutions will be eliminated and the new solution x will be added to the Pareto frontier solution set. If solution x does not dominate any existing solution and is not dominated by an existing solution, the discriminator will regard x as a non-dominated solution, add it to the Pareto frontier solution set, and expand the Pareto frontier. If the new solution x is dominated by an existing solution, it will not be added to the solution set and the solution set remains unchanged;

[0115] S34, the generator and the discriminator are iteratively updated, and when the preset convergence conditions are reached, the final Pareto optimal frontier solution set is output.

[0116] S4, evaluation and optimization of Pareto frontier, such as Figure 5 The following steps are shown:

[0117] S41. Preliminary screening based on hypervolume contribution. Use hypervolume contribution to measure the importance of each solution in the multi-objective space, and screen out solutions that contribute less to the overall solution set, remove solutions with smaller marginal contributions, and thus reduce the size of the solution set. Solutions with small hypervolume contributions often have difficulty in achieving a reasonable balance between multiple objectives. Assume that the solution set x = {x1, x2, ..., x k}, each solution x i The performance on target j is f j (x i ), and the reference point r = (r1, r2, ..., r n ) is the worst value of all targets. For target f j For example, the reference point r j The worst (largest) value in each target dimension of the solution set is usually taken, and a slight distance is added to ensure that the solution set is contained between the reference point and each solution.

[0118] r j =max(f j (x))+∈

[0119]

[0120] In the formula, ∈ is a small positive number; V(x i ) is the solution x i In each target dimension, the j The hyperrectangular volume formed by the difference between the two is calculated and the volume of each solution is added together to obtain the total hypervolume V total. Filter out solutions whose marginal contribution is less than the threshold α, eliminate them, and retain high-quality solution sets;

[0121] S42. Secondary screening based on diversity evaluation. Use diversity evaluation to ensure that the solution set is evenly distributed in the target space and to screen out solutions that are too close to other solutions. Avoid duplicate solutions in the solution set to make the final solution set more representative and diverse. The basis of diversity evaluation is the distance between solutions, and the Euclidean distance is used to measure the difference between adjacent solutions. Suppose two solutions x i and x i+1 The target value of i )=pf1(x i ), f2(x i ), ..., f n (x i )] and f(x i+1 )=[f1(x i+1 ), f2(x i+1 ), ..., f n (x i+1 )], its Euclidean distance d(x i , x i+1 ) is defined as:

[0122]

[0123] The larger the distance between solutions, the greater the difference in the distribution of the two solutions in the multi-objective space; the smaller the distance, the closer the two solutions are to each other, and there may be duplication. A distance threshold δ is set to measure diversity. If the solution x i If the distance to the adjacent solution is lower than the threshold δ, the solution is considered to be repeated or unrepresentative and is removed from the solution set. This ensures the diversity and balanced distribution of the solution set in the multi-objective space.

[0124] S43. Select the best solution based on comprehensive performance. According to the solution set screened by hypervolume contribution and diversity evaluation, select the solution with the best comprehensive performance. Perform a comprehensive score on each remaining solution based on hypervolume contribution and diversity distribution to ensure that the final selected solution performs balanced on multiple objectives.

[0125]

[0126] In the formula, S(x i ) is the comprehensive score; V(x i ) is the solution x i The marginal hypervolume contribution of , normalized as a proportion of the total hypervolume; d(x i , x i+1 ) is the solution x iThe distance between adjacent solutions is normalized as the ratio of the total solution set distance; ω1 and ω2 are weight coefficients, reflecting the importance of hypervolume contribution and diversity. The optimal solution finally selected is the one with the comprehensive score S(x i )The highest solution x * , the solution with the highest comprehensive score x * It has the most balanced performance in the multi-objective space, making the greatest contribution to the overall solution set while maintaining good diversity in distribution, so it can be regarded as the optimal solution.

[0127] Application Examples

[0128] A specific embodiment of the micro energy grid group division method of the present invention is as follows:

[0129] Figure 6 As shown in the figure, it is the micro energy grid group of this example, each dot represents a micro energy grid, and each micro energy grid has a certain independent energy supply and storage capacity. The irregular areas of different colors in the figure represent different micro energy grid groups, and the micro energy grids within these areas are connected by certain lines to form an energy network structure within the area.

[0130] The micro energy grid group is divided using the division method of the above embodiment, and the specific process is as follows:

[0131] Modeling of S1 micro energy grid:

[0132] Node Properties Matrix

[0133] NodeID <![CDATA[P gen (kW)]]> <![CDATA[P load (kW)]]> <![CDATA[E stor (kWh)]]> C(kg / kWh) 1 30 20 15 0.5 2 40 25 18 0.6 3 35 30 20 0.5 ... ... ... ... ... 73 60 50 30 0.5

[0134] Edge property matrix

[0135] Edge(i,j) <![CDATA[L ij (Loss Factor)]]> <![CDATA[C ij (Cost per kWh)]]> (1,2) 0.05 10 (1,3) 0.03 8 ... ... ... (72,73) 0.06 15

[0136] Initial partitioning scheme

[0137] NodeID Scheme1 Scheme2 Scheme3 Scheme4 Scheme5 1 1 1 2 1 3 2 1 2 3 1 1 3 2 3 1 4 3 ... ... ... ... ... ... 73 4 4 4 3 2

[0138] For example, in scheme 1, 1, 1, 2, ..., 4 indicates that the first node is assigned to microgrid group 1, the second node is assigned to microgrid group 1, the third node is assigned to microgrid group 2, ..., and the 73rd node is assigned to microgrid group 4.

[0139] S2. Construct a generative adversarial network including a generator and a discriminator;

[0140] S21. Construct a generator that introduces dynamic weights to adjust the weights of the objective function:

[0141] Objective function definition:

[0142] The objective function of the generator G is:

[0143] L Gen =λ1f1+λ2f2+λ3f3+λ4f4

[0144] f1: energy transmission cost, f2: carbon emission, f3: energy storage equipment utilization, f4: subgrid supply and demand imbalance.

[0145] Initial weights: λ1=0.25, λ2=0.25, λ3=0.25, λ4=0.25

[0146] After evaluation by the discriminator, the generator receives the following feedback:

[0147] Scheme 3 and Scheme 5 are eliminated by the discriminator, especially the transmission cost f1 and carbon emission f2 perform poorly. The generator increases the weight of f1 and f2 and adjusts to:

[0148] λ1=0.35, λ2=0.3, λ3=0.15, λ4=0.2

[0149] It is stated that transmission costs and carbon emissions are the focus of the next round of optimization.

[0150] S22. Construct a discriminator that introduces a priority selection mechanism for non-dominated solutions:

[0151] For two solutions x and p, the condition for solution x to dominate solution p is:

[0152] and The target values ​​of the initial five scenarios are as follows:

[0153]

[0154] Filter results:

[0155] Schemes 1, 2, and 4 are not dominated and enter the Pareto frontier.

[0156] Scheme 3 is dominated by Scheme 2, and Scheme 5 is dominated by Scheme 4, so it is eliminated.

[0157] Dynamic updates of S3 and Pareto frontier:

[0158] Pareto frontier solution set: Scheme 1, Scheme 2, Scheme 4.

[0159] After adjusting the dynamic weights, five new solutions are generated, and the target values ​​are as follows:

[0160]

[0161] Filter results:

[0162] Scheme 6, Scheme 8, and Scheme 9 are not dominated and enter the Pareto frontier.

[0163] Scheme 7 is dominated by Scheme 6, and Scheme 10 is dominated by Scheme 8, so it is eliminated.

[0164] Updated Pareto front solution sets: Scheme 1, Scheme 2, Scheme 4, Scheme 6, Scheme 8, Scheme 9.

[0165]

[0166]

[0167] S4. Evaluation and analysis of Pareto frontier:

[0168] Preliminary screening based on hypervolume

[0169] By calculating the hypervolume contribution V(x i ), filter out solutions with lower contributions and eliminate them.

[0170] Calculation steps:

[0171] Assume ∈ = 0.01 and calculate the reference point r according to the target value table j :

[0172] r1=max(f1)+∈=3500+0.01=3500.01

[0173] r2=max(f2)+∈=445+0.01=445.01

[0174] r3=max(f3)+∈=-0.770+0.01=-0.760

[0175] r4=max(f4)+∈=71+0.01=71.01

[0176] Calculate V(x) for each solution i ):

[0177] Scheme 1:

[0178] V(x1)=(3500.01-3370)·(445.01-445)·(-0.760-(-0.775))·(71.01-62)

[0179] V(x1)=130.01·0.01·0.015·9.01≈0.1758

[0180] Scheme 2:

[0181] V(x2)=(3500.01-3500)·(445.01-390)·(-0.760-(-0.780))·(71.01-50)

[0182] V(x2)=0.01·55.01·0.02·21.01≈0.2311

[0183] Scheme 4:

[0184] V(x4)=(3500.01-3317)·(445.01-387)·(-0.760-(-0.803))·(71.01-71)

[0185] V(x4)=183.01·58.01·0.043·0.01≈4.548

[0186] Scheme 6:

[0187] V(x6)=(3500.01-3300)·(445.01-400)·(-0.760-(-0.790))·(71.01-55)

[0188] V(x6)=200.01·45.01·0.03·16.01≈432.48

[0189] Scheme 8:

[0190] V(x8)=(3500.01-3320)·(445.01-395)·(-0.760-(-0.800))·(71.01-52)

[0191] V(x8)=180.01·50.01·0.04·19.01≈684.51

[0192] Scheme 9:

[0193] V(x9)=(3500.01-3290)·(445.01-385)·(-0.760-(-0.770))·(71.01-48)

[0194] V(x9)=210.01·60.01·0.01·23.01≈289.63

[0195] Total hypervolume contribution:

[0196] V total =V(x1)+V(x2)+V(x4)+V(x6)+V(x8)+V(x9)=1411.58

[0197] According to the formula:

[0198]

[0199] Set α = 0.001

[0200] Scheme1:

[0201] Scheme2:

[0202] Scheme4:

[0203] Scheme6:

[0204] Scheme8:

[0205] Scheme 9:

[0206] Eliminate Scheme 1 and Scheme 2, and retain Scheme 4, Scheme 6, Scheme 8, and Scheme 9.

[0207] Secondary screening based on diversity assessment

[0208] Calculate the distance d(x i ,x j ), taking Scheme 4 and Scheme 6, Scheme 6 and Scheme 8 as examples.

[0209] d(x4,x6):

[0210]

[0211] d(x6,x8):

[0212]

[0213] Construct a distance matrix:

[0214] Solution No. Scheme4 Scheme6 Scheme8 Scheme9 Scheme4 0 22.60 28.84 35.17 Scheme6 22.60 0 20.64 25.99 Scheme8 28.84 20.64 0 14.56 Scheme9 35.17 25.99 14.56 0

[0215] Set the threshold δ = 15, if the distance d(x i ,x j )<δ, the two solutions are considered too close and one needs to be eliminated.

[0216] Eliminate Scheme 8 (the distance to Scheme 9 is 14.56). Keep Scheme 4, Scheme 6, and Scheme 9.

[0217] Update V total

[0218] V total =V(x4)+V(x6)+V(x9)=4.548+432.48+289.63=726.66 Update distance matrix

[0219] Solution No. Scheme4 Scheme6 Scheme9 Scheme4 0 22.60 35.17 Scheme6 22.60 0 25.99 Scheme9 35.17 25.99 0

[0220] Update total distance

[0221] ∑ j d(x j ,x j+1 )=22.60+35.17+25.99=83.76

[0222] Calculate the comprehensive score one by one

[0223] Scheme 4:

[0224] Hypervolume contribution normalization:

[0225]

[0226] Diversity section:

[0227]

[0228] Overall rating:

[0229] S(x4)=0.5·0.0063+0.5·0.7303=0.0032+0.3652≈0.3684

[0230] Scheme 6:

[0231] Hypervolume contribution normalization:

[0232]

[0233] Diversity section:

[0234]

[0235] Overall rating:

[0236] S(x6)=0.5·0.5952+0.5·0.6897=0.2976+0.3449≈0.6425

[0237] Scheme 9:

[0238] Hypervolume contribution normalization:

[0239]

[0240] Diversity section:

[0241]

[0242] Overall rating:

[0243] S(x9)=0.5·0.3986+0.5·0.5803=0.1993+0.2901≈0.4894

[0244] Comprehensive score calculation results:

[0245] Solution No. Overall Rating Scheme4 0.3684 Scheme6 0.6425 Scheme9 0.4894

[0246] The optimal solution is Scheme 6, with an overall score of 0.6425.

[0247] Figure 6 The results shown are the division scheme of the micro energy grid group in this example, which can improve the energy utilization efficiency and energy supply stability of the system and realize energy sharing and allocation.

[0248] The technical solutions of the present invention are not limited to the above-mentioned embodiments, and any technical solutions obtained by equivalent replacement methods fall within the scope of protection required by the present invention.

Claims

1. A micro energy grid group division method based on multi-objective optimization, characterized in that: The steps include: A micro-energy grid cluster model is constructed with the objective functions of minimizing energy transmission costs, carbon emissions, and internal imbalance of subgrids, as well as maximizing the utilization efficiency of energy storage devices; Based on the micro-energy grid cluster model and the Pareto frontier solution set, a generative adversarial network including a generator and a discriminator is constructed; The generator uses a dynamic weight mechanism to adjust the weight of the objective function, construct a weighted objective function, and generate a partitioning scheme; The discriminator calculates the distance between the partition scheme and the Pareto frontier solution, determines whether the partition scheme is a non-dominated solution, scores the partition scheme, and feeds back the pros and cons of the partition scheme to the generator; Iterate the generator and discriminator, and dynamically update the Pareto frontier solution set based on the non-dominated solution during the iteration process; when the preset convergence condition is reached, the optimal Pareto frontier solution set is output; The optimal Pareto front solution set is evaluated, and the optimal partitioning scheme is selected based on the comprehensive performance.

2. The micro-energy grid group division method based on multi-objective optimization according to claim 1 is characterized in that: Building a micro energy grid model includes the following steps: S11. Each micro energy grid is defined as a node. The characteristics of node i are expressed as: the power generation of node i at time t P gen,i (t); the state E of the energy storage device of node i at time t stor,i (t); Power demand P of node i load,i (t); Carbon emissions E of node i at time t carbon,i (t); S12. Define the energy transfer between different nodes as an edge. The characteristic of edge (i, j) is expressed as: the energy P transmitted from node i to node j trans,ij (t); Energy loss L when transmitting from node i to node j ij ; The cost of transmitting energy from node i to node j is C ij (t); S13. Define the objective function, including energy transmission cost, carbon emission, utilization efficiency of energy storage equipment and internal balance of subnet, which are expressed as: In the formula, f1(x), f2(x), f3(x), and f4(x) are energy transmission cost, carbon emission, utilization efficiency of energy storage equipment, and internal imbalance of the subgrid, respectively; is a node set, each node i represents a micro energy network; ε is an edge set, each edge (i, j) represents the energy transmission path between node i and node j; C ij (t) is the cost of transmitting energy from node i to node j at time t; E i (t) is the carbon emission of node i at time t; η i (t) is the energy storage device utilization rate of node i at time t; B i (t) is the imbalance degree of node i at time t; The optimization goals are to minimize energy transmission costs, carbon emissions, internal imbalance of subgrids and maximize the utilization efficiency of energy storage equipment; S14. The supply and demand balance constraint and energy storage equipment capacity constraint of the objective function are respectively expressed as: P gen,i (t)+P stor-out,i (t)+P in,i (t)=P load,i (t)+P stor-in,i (t) 0≤E stor,i (t)≤E stor-max,i Among them, P stor-out,i (t) represents the power (discharge power) released from the energy storage device by node i at time t; P in,i (t) represents the input power obtained by node i from other nodes at time t; P stor-in,i (t) represents the charging power of the energy storage device at node i at time t; E stor-max,i Indicates the maximum storage capacity of the energy storage device at node i.

3. The micro-energy grid group division method based on multi-objective optimization according to claim 2 is characterized in that: The weighted objective function is: In the formula, f i (x) is the i-th objective function, which is energy transmission cost, carbon emission, energy storage efficiency or subnet internal balance; λ i (t) is the weight after dynamic adjustment at time step t.

4. The micro-energy grid group division method based on multi-objective optimization according to claim 2 is characterized in that: The process of adjusting weights using the dynamic weight mechanism is as follows: S21, when the generator starts training, the weights λ of all targets i (0) Set to equal; S22. Define the dynamic weight adjustment rule as: Among them, f i (x) is the specific value of the solution x provided by the discriminator for the i-th target, and the weight λ i (t+1) is the weight of the next iteration; S23, normalized dynamic weight, the normalization formula is:

5. The micro-energy grid group division method based on multi-objective optimization according to claim 4 is characterized in that: The Euclidean distance between the discriminator calculation generation solution and the Pareto front solution is: In the formula, f i (x) is the performance of solution x on the i-th objective; f i (p) is the performance of the Pareto frontier solution p on the i-th objective; n represents the number of objectives; Select solution x and all Pareto frontier solutions p∈P optimal The smallest Euclidean distance among them is taken as the distance of the solution.

6. The micro-energy grid group division method based on multi-objective optimization according to claim 5 is characterized by: The discriminator scores the partitioning scheme as follows: In the formula, D(x) is the score of the discriminator on the generated solution x; d(x,P optimal ) is the distance between solution x and the Pareto optimal solution frontier.

7. The micro-energy grid group division method based on multi-objective optimization according to claim 6 is characterized by: The loss function of the discriminator is: Among them, D(z high,i ) represents the high quality score of the discriminator on the generated solution on the i-th target; D(z low,i ) represents the low quality score of the generated solution on the i-th target by the discriminator, and the discriminator feeds back to the generator the specific value f of the low quality solution x on each target i (x).

8. The micro-energy grid group division method based on multi-objective optimization according to claim 7 is characterized in that: The dynamic updating of the Pareto frontier solution set comprises the following steps: S31, the generator generates a new solution x; S32, the discriminator receives the solution x generated by the generator and determines whether it dominates or is dominated by the existing Pareto frontier solution set; S33. For solution x and frontier solution p, the condition that solution c dominates frontier solution p is: and If solution x dominates the solutions in the existing Pareto frontier, the dominated solutions are eliminated and solution x is added to the Pareto frontier solution set; if solution x does not dominate any solutions in the existing Pareto frontier, the discriminator regards x as a non-dominated solution and adds it to the Pareto frontier solution set.

9. The micro-energy grid group division method based on multi-objective optimization according to claim 8 is characterized in that: The evaluation of the optimal Pareto frontier solution set includes: S41. Preliminary screening based on hypervolume contribution: r j =max(f j (x))+∈ Where ∈ is a positive number; V(x i ) is the solution x i In each target dimension, the j The difference between the two forms a hyperrectangular volume, V total Add the volumes of each solution together to get the total hypervolume; Eliminate solutions whose marginal contribution is less than the threshold α; S42, secondary screening based on diversity evaluation; use Euclidean distance to measure the differences between adjacent solutions. Suppose two solutions x i and x i+1 The target value of i )=pf1(x i ),f2(x i ),…,f n (x i )] and f(x i+1 )=[f1(x i+1 ),f2(x i+1 ),…,f n (x i+1 )], its Euclidean distance d(x i ,x i+1 ) is defined as: Among them, d(x i ,x i+1 ) is the solution x i and adjacent solutions x i+1 The Euclidean distance between Eliminate solutions x whose distance to adjacent solutions is less than the threshold δ i ; S53. Select the best solution based on comprehensive performance: In the formula, S(x i ) is the comprehensive score; ω1 and ω2 are weight coefficients; Select the comprehensive score S(x i )The highest solution x * is the optimal solution.

Images