Power and capacity optimal configuration method and system for participation of energy storage in peak regulation of power grid

Abstract

The invention discloses a power and capacity optimal configuration method and system for participation of energy storage in power grid peak regulation, and relates to the technical field of energy storage capacity optimization, and the method comprises the steps: obtaining historical scene data of new energy power generation of a target power distribution network; inputting historical scene data into the generative adversarial network and clustering to obtain a high-quality typical scene of new energy power generation; constructing a time sequence model based on the time sequence load demand of the target power distribution network and the high-quality typical scene; obtaining a power response demand of the target power distribution network for the energy storage system based on the time sequence model; an energy storage battery capacity optimization configuration model with the minimum annual cost and the maximum absorption capacity is established based on the power response demand; and solving by adopting a non-dominated sorting genetic algorithm based on the energy storage battery capacity optimal configuration model to obtain an energy storage system optimal configuration scheme of the target power distribution network. And an energy storage configuration scheme with optimal economical efficiency and reliability in a typical region is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage capacity optimization, and more particularly to a power and capacity optimization configuration method and system for energy storage participating in grid peak shaving. BACKGROUND

[0002] At present, with the development of new power systems, multi-scale distributed power sources accessing distribution networks has become the norm. The intermittent output of wind power and photovoltaic power leads to an enlarged peak-valley difference of equivalent load, and the contradiction between power shortage in peak period and power surplus in valley period is prominent. The traditional peak shaving means has great limitations; the climbing rate of thermal power units is slow, the minimum technical output is limited, frequent start-stop increases equipment wear and tear and coal consumption, pumped storage is restricted by geographical conditions, and the construction period is as long as 5-8 years, and has a significant impact on the ecological environment. Energy storage systems become key regulation resources of new power systems due to their unique technical advantages: energy storage systems have millisecond-level response capability, can bidirectionally regulate power (charge / discharge), charge in the large new energy generation and low load valley period to absorb abandoned wind and light; and discharge in the load peak period to alleviate the power supply gap. The regulation accuracy reaches 98%, far exceeding traditional units, and can be deployed in flexible scenarios such as substations and user sides.

[0003] However, the current energy storage configuration faces the following technical challenges: 1. Difficulty in quantifying power response demand: the spatiotemporal uncertainty of new energy output and load leads to net load prediction error, and traditional statistical methods cannot accurately depict the power support direction (charge / discharge) and capacity threshold required by energy storage; 2. Insufficient coordination optimization of capacity configuration and operation strategy; 3. Single scenario limitation: early researches mainly focus on single target of peak shaving or frequency modulation, and underestimate the value of energy storage in multiple auxiliary services; 4. Economic constraints: the proportion of energy storage investment cost is more than 60%, and excessive capacity leads to prolonged payback period, and insufficient capacity cannot meet the peak shaving demand; 5. Lack of market mechanism and technical standards: the peak shaving compensation standards of different provinces are quite different, and there is a lack of unified performance evaluation index, which affects the enthusiasm of energy storage participating in the market.

[0004] Therefore, how to obtain an energy storage configuration scheme with optimal economy and reliability in a typical region is a problem to be solved by those skilled in the art. SUMMARY

[0005] Therefore, the present application provides a power and capacity optimization configuration method and system for energy storage participating in grid peak shaving, and obtains an energy storage configuration scheme with optimal economy and reliability in a typical region.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A power and capacity optimization configuration method for energy storage participating in grid peak shaving, comprising:

[0008] acquire historical scenario data of new energy power generation of a target power distribution network;

[0009] input the historical scenario data into a generative adversarial network and cluster to obtain high-quality typical scenarios of the new energy power generation;

[0010] construct a time series model based on time series load demand of the target power distribution network and the high-quality typical scenarios;

[0011] acquire power response demand of an energy storage system of the target power distribution network based on the time series model;

[0012] establish an energy storage battery capacity optimization configuration model with minimum annual cost and maximum consumption capacity based on the power response demand;

[0013] solve the energy storage battery capacity optimization configuration model based on a non-dominated sorting genetic algorithm to obtain an optimal configuration scheme of the energy storage system of the target power distribution network.

[0014] In one embodiment, the historical scenario data includes: illumination intensity, wind speed, wind direction, temperature, photovoltaic component efficiency, air density, cloud coverage, latitude and altitude, terrain complexity, and ground reflectivity.

[0015] In one embodiment, the method for obtaining the high-quality typical scenarios is:

[0016] input the historical scenario data into a generative adversarial network to obtain a typical scenario set of the new energy power generation;

[0017] randomly select K data points from the typical scenario set as initial clustering centers;

[0018] calculate the distance from each data point in the typical scenario set to each initial clustering center and assign it to the nearest initial clustering center to form K initial clusters;

[0019] recalculate the mean of each initial cluster as a new clustering center;

[0020] repeat the above process based on the new clustering center until the termination condition is met to obtain K final clusters;

[0021] select the scenario closest to the clustering center in each final cluster as a representative scenario;

[0022] all representative scenarios together constitute the high-quality typical scenarios.

[0023] In one embodiment, the method for obtaining the typical scenario set is:

[0024] use a deep convolutional generative adversarial network or a Transformer-based GAN as a generator;

[0025] based on a feedforward neural network or a convolutional neural network as the discriminator;

[0026] based on a Wasserstein distance or a least square loss as the loss function;

[0027] based on the loss function, optimizing the model parameters of the generator and the model parameters of the discriminator through adversarial training until the false scene generated by the generator is close to the real scene in statistical characteristics, obtaining a trained generative adversarial network;

[0028] based on the historical scene data input into the trained generative adversarial network, obtaining that the new energy power time series scene collectively constitutes the typical scene set.

[0029] In one embodiment, the power response demand acquisition method is:

[0030] based on the time series model, using a Bayesian optimization algorithm for dynamic optimization to obtain an optimized time series model;

[0031] based on the optimized time series model, obtaining the power response demand of the target power distribution network to the energy storage system.

[0032] In one embodiment, obtaining the optimized time series model specifically includes:

[0033] defining a hyperparameter search space;

[0034] selecting a Gaussian process as a surrogate model and constructing a mapping relationship between the hyperparameter search space and the model validation loss as a first objective function;

[0035] assuming that the first objective function is subject to a prior distribution;

[0036] based on the expected improvement or improvement probability as a collection function;

[0037] each time a new sample point is used to test the objective function, and the prior distribution of the first objective function is updated based on the obtained information;

[0038] determine the next sampling point through the collection function;

[0039] cycle the above process until a preset evaluation number is reached, obtaining a posterior distribution of the first objective function;

[0040] selecting the maximum point reached by the posterior distribution as the optimal parameter combination, completing optimization, and obtaining the optimized time series model.

[0041] In one embodiment, the energy storage battery capacity optimization configuration model specifically includes:

[0042] minimize the total annualized life cycle cost of the energy storage system, maximize the new energy accommodation rate of the power distribution network and model constraints;

[0043] The minimization of the total annualized life cycle cost of the energy storage system and the maximization of the new energy accommodation rate of the power distribution network jointly constitute the objective function;

[0044] The total annualized life cycle cost C of the energy storage system AT Specifically:

[0045] C AT = (C inv *C RF +C om )+C loss *E loss_annualized ;

[0046] Wherein, C inv represents the initial investment cost of the unit installed capacity of the energy storage, C RF represents the capital recovery factor, C om represents the annual operation and maintenance cost of the unit capacity of the energy storage, C loss represents the energy loss cost of one cycle of the energy storage, E loss_annualized represents the annualized equivalent energy loss of the energy storage system.

[0047] The new energy accommodation rate N of the power distribution network is specifically:

[0048]

[0049] Wherein, E Gen represents all the electric energy generated by the new energy power source, E Cur represents the electric energy generated by the abandoned new energy power source.

[0050] In one embodiment, the model constraints include: power distribution network operation constraints, energy storage system operation constraints and energy storage capacity and power association constraints.

[0051] Wherein, the energy storage system operation constraints include: charge / discharge power limit, energy storage charge / discharge state exclusion constraint and energy storage energy dynamic constraint.

[0052] In one embodiment, the energy storage system optimal configuration scheme acquisition method is:

[0053] Randomly generate an initial population, wherein each individual represents a candidate energy storage configuration scheme;

[0054] Perform genetic operations based on the initial population to obtain a final population;

[0055] perform hierarchical sorting on the final population based on the objective function, to obtain a plurality of hierarchical global non-dominated solutions and corresponding non-dominated sorting;

[0056] calculate congestion degree based on the global non-dominated solution of each hierarchical;

[0057] select a new generation population based on the non-dominated sorting and the congestion degree;

[0058] perform local search based on the new generation population and handle the constraint condition by using a penalty function method, until the maximum number of iterations is reached or the solution set converges, and output the optimal configuration scheme of the energy storage system;

[0059] The optimal configuration scheme of the energy storage system includes: optimal energy storage battery power and optimal capacity configuration parameters.

[0060] A power and capacity optimization configuration system for energy storage participating in grid peak shaving, comprising: a data acquisition module, a data processing module, a time sequence model construction module, a power demand acquisition module, a configuration model construction module and a configuration scheme output module;

[0061] The data acquisition module is configured to acquire historical scenario data of new energy power generation of a target distribution network.

[0062] The data processing module is configured to input the historical scenario data into a generative adversarial network and cluster to obtain high-quality typical scenarios of the new energy power generation.

[0063] The time sequence model construction module is configured to construct a time sequence model based on time sequence load demand of the target distribution network and the high-quality typical scenarios.

[0064] The power demand acquisition module is configured to acquire power response demand of the target distribution network for an energy storage system based on the time sequence model.

[0065] The configuration model construction module is configured to establish an energy storage battery capacity optimization configuration model with minimum annual cost and maximum consumption capacity based on the power response demand.

[0066] The configuration scheme output module is configured to solve the energy storage battery capacity optimization configuration model by using a non-dominated sorting genetic algorithm, to obtain an optimal configuration scheme of an energy storage system of the target distribution network.

[0067] Compared with the prior art, the power and capacity optimization configuration method and system for energy storage participating in power grid peak regulation provided by the application can generate a new energy generation high-quality typical scenario set by using a generative adversarial network (GAN) combined with k-means clustering, can dynamically optimize the hyperparameters of a time sequence model by using a Bayesian optimization algorithm, can accurately predict net load power and analyze energy storage power response demand, can realize a bidirectional regulation mechanism of low-valley charging (valley filling) and high-peak discharging (peak shaving), can establish an optimization model with multiple objectives of minimizing the whole life cycle cost (including investment, operation and maintenance and loss cost) of energy storage and maximizing the new energy consumption rate, can introduce an improved NSGA-II algorithm to solve the optimal energy storage power and capacity configuration, and can finally output an energy storage configuration scheme with optimal economy and reliability in a typical region by improving convergence accuracy and diversity through non-dominated sorting, congestion calculation and local search strategy. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.

[0069] Figure 1 The power and capacity optimization configuration method for energy storage participating in power grid peak regulation provided by the present application is provided.

[0070] Figure 2 The NSGA-II algorithm flowchart provided by the present application is provided.

[0071] Figure 3 The power and capacity optimization configuration system for energy storage participating in power grid peak regulation provided by the present application is provided. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0073] Embodiment 1

[0074] As shown in the following table, the power and capacity optimization configuration method for energy storage participating in power grid peak regulation disclosed by the present application comprises the following steps, and numbers S1-S6 are set for convenience of description, which are not used to limit the sequence of the steps of the present application: Figure 1 As shown in the following table, the power and capacity optimization configuration method for energy storage participating in power grid peak regulation disclosed by the present application comprises the following steps, and numbers S1-S6 are set for convenience of description, which are not used to limit the sequence of the steps of the present application:

[0075] The specific mechanism of energy storage participating in grid peak regulation is as follows:

[0076] Off-peak charging: during the night or off-peak load period, the energy storage system absorbs excess power from the grid and converts it into other forms of energy for storage; peak discharge: during the daytime or evening peak load period, the energy storage system releases the stored energy to supplement the power supply gap of the grid.

[0077] S1 obtains historical scenario data of new energy power generation of a target power distribution network.

[0078] Further, the historical scenario data includes: light intensity, wind speed, wind direction, temperature, photovoltaic component efficiency, air density, cloud coverage, latitude and altitude, terrain complexity, and surface reflectivity.

[0079] Further, the target power distribution network is an actual power distribution network, i.e., an actual power distribution network corresponding to the application scenario of the present patent.

[0080] S2 inputs the historical scenario data into a generative adversarial network and clusters to obtain high-quality typical scenarios of new energy power generation.

[0081] Further, the method for obtaining the high-quality typical scenarios is as follows:

[0082] The historical scenario data is input into a generative adversarial network to obtain a typical scenario set of new energy power generation;

[0083] K data points are randomly selected from the typical scenario set as initial cluster centers;

[0084] The Euclidean distance or Mahalanobis distance of each data point in the typical scenario set to each initial cluster center is calculated, and the data point is assigned to the nearest initial cluster center to form K initial clusters;

[0085] The mean of each initial cluster is recalculated as a new cluster center;

[0086] The above process is repeated based on the new cluster center until a termination condition is met, and K final clusters are obtained;

[0087] The nearest scenario to the cluster center is selected from each final cluster as a representative scenario;

[0088] All representative scenarios together constitute the high-quality typical scenarios.

[0089] Further, in the present embodiment, the termination condition is that when any of the following conditions is met, the process is stopped: the centroid position no longer changes, the data points are no longer reassigned, and the sum of squared errors (SSE) reaches a local minimum.

[0090] Further, the method for obtaining the typical scenario set is as follows:

[0091] Deep convolutional generative adversarial networks or Transformer-based GANs as generators;

[0092] Feedforward neural networks or convolutional neural networks as discriminators for distinguishing between real historical scenarios and fake scenarios generated by the generator;

[0093] Wasserstein distance or least squares loss as loss function;

[0094] Optimize the model parameters of the generator and the model parameters of the discriminator based on the loss function through adversarial training until the fake scenarios generated by the generator are close to the real scenarios in statistical characteristics, obtaining a trained generative adversarial network;

[0095] Based on the input of historical scenario data into the trained generative adversarial network, the new energy power time series scenario is obtained to form a typical scenario set.

[0096] Further, the generative adversarial network (Generative Adversarial Networks, GAN) is a revolutionary deep learning model composed of two deep neural network models, generator G and discriminator D. The generator processes the noise signal z obeying the probability distribution p z (z) to obtain generated data G(z) obeying the probability distribution p G (z), and the goal is to make the probability distribution p G (z) of the generated data approximate the probability distribution p data of the historical data; the discriminator judges the generated data of the generator and the historical data obeying the probability distribution p data (x) and outputs the probability D(G(z)) that the generated data G(z) obeys the real distribution p

[0097] Core principle: game of generation and confrontation;

[0098] Basic architecture: Generator: input random noise vector, generate new samples simulating real data distribution (such as images, audio), goal is to deceive the discriminator, make it unable to distinguish between generated data and real data.

[0099] Discriminator (Discriminator): receives real data and generated data, outputs the probability that the sample is real, goal is to accurately identify true and false samples as much as possible;

[0100] Training process: alternating optimization, forming a "minimax game", the objective function of the generative adversarial network is:

[0101]

[0102] wherein, represents minimizing the generator loss to make the fake data generated by the generator G more realistic, represents maximizing the discriminator gain to make the discriminator D more accurate in distinguishing real and fake data, V(D, G) represents a value function quantifying the adversarial performance of the discriminator and the generator, E represents an expectation operator, x represents a real data sample, p data (x) represents a real data distribution, D(x) represents a discriminator output, z represents a random noise vector, p z (z) represents a noise distribution, G(z) represents a generator output, and D(G(z)) represents a prediction of the discriminator on fake data.

[0103] Nash equilibrium: in an ideal state, when the generated data distribution pg approximates the real distribution pdata, the discriminator cannot distinguish between real and fake (the output probability is always 0.5), and the model reaches a dynamic balance.

[0104] Further, in one possible implementation, the Wasserstein distance is used as the loss function:

[0105]

[0106] wherein, Π (p，q) represents the set of all possible joint distributions of the distributions p and q combined, for each possible joint distribution y, a sample x and y can be sampled from it (x, y) ~ y, and the distance |x-y| of the pair of samples can be calculated, so the expected value E x，y～γ [||x-y||] of the sample pair distance under the joint distribution y can be calculated, and the lower bound of this expected value taken in all possible joint distributions is the Wasserstein distance. Through adversarial training, the generator and discriminator parameters are optimized until the generated scene approaches the real scene in statistical characteristics.

[0107] Further, in one possible implementation, the least squares loss is used as the loss function:

[0108] L(Y, Y hat )=∑(y i -y hat_i ) 2 ;

[0109] wherein, Y represents a vector of real observations, Y hat represents a vector of predicted values of the model, and y iy represents the true value of the i-th sample hat_i y represents the predicted value of the i-th sample.

[0110] The objective of the least squares loss function is to minimize the difference between the predicted value and the true observed value. By minimizing the loss function, the best model parameters can be found, so that the model can more accurately fit the training data.

[0111] S3 constructs a time series model based on the time series load demand of the target power distribution network and high-quality typical scenarios.

[0112] The input of the time series model includes: time series load demand, time series new energy power generation corresponding to each scenario in the high-quality typical scenario set generated by S2, weather data, calendar features and price signal;

[0113] The output of the time series model is: predicted future time series of power distribution network net load power;

[0114] The net load power is defined as the difference between the total load power in the power distribution network region and the total new energy power generation in the region.

[0115] S4 obtains the power response demand of the target power distribution network to the energy storage system based on the time series model.

[0116] Further, the power response demand obtaining method is:

[0117] Based on the time series model, a Bayesian optimization algorithm is used for dynamic optimization to obtain an optimized time series model;

[0118] Based on the optimized time series model, the power response demand of the target power distribution network to the energy storage system is obtained.

[0119] Further, the hyperparameters in the time series model can be determined by trial and error method, but the trial and error method can only obtain a relatively optimal time series model, and there is a large randomness and contingency, which cannot fully utilize the performance of the time series model. In view of this, the present application optimizes the time series model by using Bayesian theory, and the problem of optimizing the hyperparameters of the time series model by using Bayesian theory can be defined as:

[0120]

[0121] Wherein, x is the hyperparameter of the time series model; f is the objective function for evaluating the performance of the model, and the objective function has a complex structure and no gradient information; represents the hyperparameter space in the time series model; d represents the dimension of the hyperparameters to be optimized in the time series model.

[0122] Further, the optimized time series model is obtained, which specifically includes:

[0123] Define the hyperparameter search space: learning rate, number of hidden layer units, number of layers, dropout ratio, batch size, and sequence length;

[0124] Select Gaussian process as the surrogate model and construct the mapping relationship between the hyperparameter search space and the model validation loss as the objective function;

[0125] Assume that the objective function is subject to a prior distribution;

[0126] Based on the expected improvement or improvement probability as the acquisition function;

[0127] Test the objective function with a new sample point each time, and update the prior distribution of the objective function based on the information obtained;

[0128] Determine the next sampling point through the acquisition function;

[0129] Cycle the above process until the preset evaluation times are reached, and obtain the posterior distribution of the objective function;

[0130] Select the maximum point reached by the posterior distribution as the optimal parameter combination, complete the optimization, and obtain the optimized time series model.

[0131] S5 establishes an energy storage battery capacity optimization configuration model with minimum annualized cost and maximum consumption capacity based on power response demand.

[0132] Further, the energy storage battery capacity optimization configuration model specifically includes:

[0133] Minimizing the total annual cost of the energy storage system throughout its life cycle, maximizing the new energy consumption rate of the distribution network, and model constraints;

[0134] Minimizing the total annual cost of the energy storage system throughout its life cycle and maximizing the new energy consumption rate of the distribution network together constitute the total objective function;

[0135] The total annual cost of the energy storage system throughout its life cycle C AT Specifically,

[0136] C AT = (C inv *C RF +C om )+C loss *E loss_annualized ;

[0137] Wherein, C inv represents the initial investment cost of the energy storage unit capacity (yuan / kWh or yuan / kW), C RF represents the capital recovery factor, C om represents the annual operation and maintenance cost of the energy storage unit capacity (yuan / kWh / year), C lossE represents the energy loss cost per energy storage cycle (yuan / kWh). loss_annualized This represents the annualized equivalent energy loss (kWh) of the energy storage system.

[0138] The specific renewable energy absorption rate N of the distribution network is:

[0139]

[0140] Among them, E Gen E represents all electrical energy generated by new energy sources. Cur This refers to the electrical energy generated by discarded renewable energy sources.

[0141] Furthermore, the model constraints include: distribution network operation constraints, energy storage system operation constraints, and energy storage capacity and power correlation constraints;

[0142] Among them, the constraints on the operation of energy storage systems include: charging / discharging power limits, mutual exclusion constraints on energy storage charging and discharging states, and dynamic constraints on energy storage energy.

[0143] Furthermore, the operating constraints of the distribution network are:

[0144]

[0145]

[0146] Where j, k, and l all represent distribution network bus indices, J(k) and L(k) are the parent node and child node, respectively, and P jk Q jk r jk x jk and I jk Let these represent the active power, reactive power, line resistance, line inductance, and line current from bus j to bus k, respectively. V k Let represent the active power, active load, reactive load of the photovoltaic system at bus k, and the bus voltage, respectively, and t represent the corresponding time. V represents the upper limit of the line current between bus j and bus k. min and V max V represents the upper and lower limits of the distribution network voltage, respectively. j This represents the voltage at node j in the distribution network. Indicates the maximum output power of photovoltaic power. This indicates the photovoltaic output power after reduction.

[0147] Furthermore, the charging / discharging power is limited as follows:

[0148] 0≤P charge ≤η c *P ess ;

[0149] 0≤P dicharge ≤η c *P ess ;

[0150] SOC min *E ess ≤SOC t ≤SOC max *E ess ;

[0151] where P charge denotes the charging power, which is the power absorbed by the energy storage system from the grid or renewable energy sources at the charging state, η c denotes the charging efficiency, which ranges from 0≤η c ≤1, is the energy conversion efficiency during the charging process, P ess denotes the rated power of the energy storage system, which is the maximum charging / discharging power designed value that the energy storage system can withstand, P discharge denotes the discharging power at time t, which is the power output by the battery during discharging, SOC min denotes, SOC max denotes, SOC t denotes the state of charge of the battery at time t, denotes the remaining capacity of the battery at time t, which is usually expressed in percentage or absolute energy value, but the calculation result in this formula is a relative value, which needs to be multiplied by the rated capacity to obtain the actual energy, E ess denotes the rated capacity of the energy storage system, which is the total energy capacity when the battery is fully charged;

[0152] The energy storage charging and discharging state mutual constraint is:

[0153] A binary variable is used to ensure that the energy storage cannot be charged and discharged at the same time;

[0154] The energy storage energy dynamic constraint is:

[0155] SOC t = SOC t-1 +(η c *P charge *Δt) / E ess -(P discharge *Δt) / (η d *E ess );

[0156] where SOC t-1 is the state of charge of the battery at time t-1, i.e. the SOC value of the previous time step, Δt represents the time interval, i.e. the time length from time t-1 to time t, η d denotes the discharging efficiency (0<η d≤ 1) represents the efficiency of chemical energy conversion into electrical energy during discharging, considering the energy loss during discharging.

[0157] Further, the energy storage capacity and power correlation constraint is:

[0158] E ess ≥ P ess / C rate ;

[0159] Wherein, C rate represents the maximum charge-discharge rate allowed by the energy storage, P ess represents the rated power of the energy storage system (unit: kW), represents the maximum power that the battery can continuously output, C rate represents the charge-discharge rate (unit: 1 / hour), and represents an index of the battery charge-discharge speed.

[0160] S6 obtains the optimal configuration scheme of the energy storage system of the target power distribution network by using a non-dominated sorting genetic algorithm based on the energy storage battery capacity optimization configuration model.

[0161] Further, the optimal configuration scheme of the energy storage system is obtained by:

[0162] Randomly generating an initial population, wherein each individual represents a candidate energy storage configuration scheme;

[0163] Performing genetic operations based on the initial population to obtain a final population;

[0164] Performing hierarchical sorting on the final population based on the objective function to obtain a plurality of hierarchical global non-dominated solutions and corresponding non-dominated sorting;

[0165] Calculating the crowding degree based on the global non-dominated solution of each hierarchical level;

[0166] Selecting a new generation population based on the non-dominated sorting and the crowding degree;

[0167] Performing local search based on the new generation population and using a penalty function method to handle the constraint conditions, until the maximum number of iterations is reached or the solution set converges, and outputting the optimal configuration scheme of the energy storage system;

[0168] The optimal configuration scheme of the energy storage system includes: optimal energy storage battery power and optimal capacity configuration parameters.

[0169] Further, the non-dominated sorting hierarchical method specifically includes:

[0170] The objective of hierarchical sorting is to divide the population into a plurality of non-dominated levels (Frnts), wherein Rank1 is the optimal level, and the specific steps are as follows:

[0171] Initializing attributes:

[0172] Initialize for each individual xi:

[0173] DminatingSeti: Set of all individuals dominated by xi (initially empty);

[0174] DominatedCounti: Number of individuals dominating xi (initially 0);

[0175] Compare dominance relation pairwise:

[0176] Iterate over all pairs of individuals (xi, xj):

[0177] If xi < xj, add j to DminatingSeti, DominatedCountj <- DominatedCountj + 1;

[0178] If xj < xi: add i to DminatingSetj, DominatedCounti <- DominatedCounti + 1;

[0179] Determine Rank1 (first front):

[0180] All individuals with DominatedCounti = 0 are assigned to Rank1 (not dominated by any other solution);

[0181] Recursively generate subsequent levels:

[0182] Current level k = 1:

[0183] Iterate over each individual xi in Rank1:

[0184] For each individual xj in its dominating set DminatingSeti, perform DominatedCountj <- DominatedCountj - 1;

[0185] If DominatedCountj = 0, add xj to temporary set Q (i.e., candidate for next level);

[0186] Assign individuals in Q to Rank2;

[0187] Iteration: Repeat the above process until all individuals are assigned a level (k = 2, 3,...);

[0188] Key outputs:

[0189] Rank1: Global non-dominated solutions (Pareto front), representing the best trade-off solutions between cost and accommodation rate;

[0190] Rank2 and subsequent: Solutions dominated by higher levels, with decreasing quality.

[0191] Further, the role of the new generation population: generate high-quality solution set (especially Rank 1 individuals) through elitist strategy, provide high-quality initial points for local search; preserve population diversity, avoid local search into the same area.

[0192] Further, the execution logic of local search:

[0193] Input: select Rank 1 individuals (optimal front solution) from the new generation population.

[0194] Operation: for each Rank 1 individual x, generate a new solution x' in its neighborhood (such as small perturbation parameters).

[0195] Evaluation and comparison: calculate the objective function values of x and x':

[0196] If x' dominates x, replace the original solution;

[0197] If they do not dominate each other, keep both to maintain diversity.

[0198] Output: the optimized solution set is recombined into the population.

[0199] Synergistic value: improve convergence accuracy: fine search around elite solutions, approximate the true Paret front; avoid premature convergence: search only for Rank 1 solutions, do not destroy the overall diversity of the population.

[0200] Further, the constraint processing mechanism: use the penalty function method to handle the constraint conditions, which include the charge and discharge power limit, SOC constraint, grid voltage limit, etc. If the configuration scheme violates the constraint, its fitness value will be greatly reduced, so it will be eliminated in the sorting; termination condition and output, when the maximum number of iterations or the solution set converges, the algorithm is terminated; output the Pareto optimal solution set in the current population, including the configuration scheme set of the energy storage power Pess and the capacity Eess.

[0201] Further, the non-dominated sorting genetic algorithm of the embodiment adopts the NSGA-II algorithm, which is an iterative evolutionary algorithm, including two main loops:

[0202] Outer main loop: control the number of evolution generations G;

[0203] Inner chaotic search loop: adaptive chaotic refinement search is performed in each generation;

[0204] Its workflow is as shown in Figure 2 :

[0205] Initialization phase: select parameters: population size, maximum number of evolution generations and chaotic control parameters;

[0206] Outer evolution loop:

[0207] Loop condition: G < maximum evolution generation;

[0208] Initialize population: evolution generation G = 0, initialize population PG;

[0209] Chaos mapping and objective function calculation: map chaos variables to decision variable space, calculate the objective function value of each individual based on the energy storage capacity optimization configuration model;

[0210] Non-dominated sorting and fitness assignment: perform fast non-dominated sorting, assign virtual fitness according to population ranking;

[0211] Generate offspring through genetic operations: perform double crossover tournament selection, real number crossover and chaos mutation to generate offspring population Q G ;

[0212] Merge and select the next generation: merge parent P G and offspring Q G , and calculate the objective function value of each individual in the merged population R G , perform fast non-dominated sorting on R G , calculate the crowding distance of individuals in the non-dominated front and sort, select the best Np individuals in the non-dominated front to form the next generation population P G+1 ;

[0213] Chaos refinement search (inner loop), perform chaos search in each generation to enhance local search ability:

[0214] Set iteration number K = 0;

[0215] Loop condition: maximum chaos iteration number not reached;

[0216] Loop inside: adaptive chaos refinement search, map chaos variables to new decision variables, K = K + 1, judge whether the maximum chaos iteration number is reached → if not, continue the loop, after chaos search is completed, return to the main loop;

[0217] Update generation: G = G + 1;

[0218] Judge whether the maximum evolution generation is reached, if yes, end the algorithm; otherwise, return to the main loop.

[0219] By means of chaotic variable mapping and chaotic mutation operation, the randomness, ergodicity and initial value sensitivity of chaotic characteristics are utilized to effectively avoid the algorithm from falling into local optimum, enhance population diversity and improve global optimization capability for complex multi-objective optimization problems; in combination with fast non-dominated sorting and crowded distance calculation, non-inferior front individuals can be accurately identified and their distribution uniformity can be maintained, so that the Pareto optimal solution set output by the algorithm has good convergence and diversity, meeting the demand of micro-grid multi-objective scheduling; the adaptive chaotic refinement search link is introduced to locally optimize the non-inferior front individuals in the later evolution stage, further improving the precision of key solutions and balancing the global exploration and local development capability; through double-branch league selection, real number crossover and other operations, in combination with the objective function calculation of the micro-grid power model, the complex scheduling problems with multiple constraints and multiple variables in the micro-grid can be dealt with specifically, and the engineering practicability of the optimization result is improved.

[0220] Embodiment 2

[0221] As shown in Figure 3 Based on the same inventive concept, the present application also provides a power and capacity optimization configuration system for energy storage participating in grid peak regulation, comprising a data acquisition module, a data processing module, a time sequence model construction module, a power demand acquisition module, a configuration model construction module and a configuration scheme output module.

[0222] The data acquisition module is used to acquire historical scenario data of new energy power generation of the target distribution network.

[0223] The data processing module is used to input the historical scenario data into a generative adversarial network and cluster to obtain high-quality typical scenarios of new energy power generation.

[0224] The time sequence model construction module is used to construct a time sequence model based on the time sequence load demand and high-quality typical scenarios of the target distribution network.

[0225] The power demand acquisition module is used to acquire the power response demand of the target distribution network for the energy storage system based on the time sequence model.

[0226] The configuration model construction module is used to establish an energy storage battery capacity optimization configuration model with minimum annual cost and maximum consumption capacity based on the power response demand.

[0227] The configuration scheme output module is used to solve the energy storage battery capacity optimization configuration model based on the non-dominated sorting genetic algorithm to obtain the optimal configuration scheme of the energy storage system of the target distribution network.

[0228] Further, in this embodiment, the function implementation methods of each functional module correspond to the above methods one by one, which will not be repeated here.

[0229] Embodiment 3

[0230] Based on the same inventive concept, the application further provides an electronic device, comprising a processor and a memory, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the power and capacity optimization configuration method of the energy storage participating in the grid peak shaving according to any one of the embodiments 1.

[0231] Based on the same inventive concept, the application further provides a computer device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete the communication among each other through the communication bus.

[0232] The memory is used for storing a computer program.

[0233] The processor is used for executing the program stored on the memory, and can implement the power and capacity optimization configuration method of the energy storage participating in the grid peak shaving according to any one of the embodiments 1.

[0234] The electronic device can comprise a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete the communication among each other through the communication bus. The processor can call the logical instructions in the memory to execute the power and capacity optimization configuration method of the energy storage participating in the grid peak shaving according to any one of the embodiments 1.

[0235] In addition, the logical instructions in the memory described above can be implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the application essentially or the part of the prior art that makes a contribution or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0236] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0237] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the power and capacity configuration of energy storage for grid peak shaving, characterized in that, include: Acquire historical scenario data of renewable energy generation in the target distribution network; Based on the historical scene data, the data is input into a generative adversarial network and clustered to obtain high-quality typical scenarios for new energy power generation; A time-series model is constructed based on the time-series load demand of the target distribution network and the high-quality typical scenario. Based on the time series model, the power response demand of the target distribution network to the energy storage system is obtained; Based on the power response requirements, an energy storage battery capacity optimization configuration model is established that minimizes annualized cost and maximizes absorption capacity. The optimal configuration scheme of the energy storage system for the target distribution network is obtained by using a non-dominated sorting genetic algorithm based on the energy storage battery capacity optimization configuration model.

2. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 1, characterized in that, The historical scene data includes: light intensity, wind speed, wind direction, temperature, photovoltaic module efficiency, air density, cloud cover, latitude and altitude, terrain complexity, and surface reflectivity.

3. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 1, characterized in that, The method for obtaining the high-quality typical scenarios is as follows: Based on the historical scene data, a set of typical scenarios for the new energy power generation is obtained by inputting the data into a generative adversarial network. K data points are randomly selected as initial cluster centers based on the aforementioned typical scenario set; Calculate the distance from each data point in the typical scenario set to each of the initial cluster centers and assign them to the nearest initial cluster center to form K initial clusters; The mean of each initial cluster is recalculated as the new cluster center; Based on the new cluster centers, repeat the above process until the termination condition is met to obtain K final clusters; The scene closest to the cluster center in each final cluster is selected as the representative scene. All representative scenarios together constitute the high-quality typical scenario.

4. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 3, characterized in that, The method for obtaining the typical scenario set is as follows: Generative adversarial networks based on deep convolutional generative adversarial networks or Transformer-based GANs can be used as generators. Based on feedforward neural networks or convolutional neural networks as discriminators; Based on Wasserstein distance or least squares loss as the loss function; Based on the loss function, the model parameters of the generator and the model parameters of the discriminator are optimized through adversarial training until the fake scenes generated by the generator are close to the real scenes in statistical characteristics, thus obtaining a trained generative adversarial network. The historical scene data is input into the trained generative adversarial network to obtain the time series scenes of new energy power generation, which together form the typical scene set.

5. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 1, characterized in that, The method for obtaining the power response requirement is as follows: Based on the aforementioned time series model, a Bayesian optimization algorithm is used for dynamic optimization to obtain the optimized time series model. The power response requirements of the target distribution network to the energy storage system are obtained based on the optimized time-series model.

6. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 5, characterized in that, The optimized time series model is obtained, specifically including: Define the hyperparameter search space; A Gaussian process is selected as the surrogate model, and the mapping relationship between the hyperparameter search space and the model validation loss is constructed as the first objective function. Assume that the first objective function follows a prior distribution; The acquisition function is based on the expected improvement or enhancement probability. Each time, a new sample point is used to test the objective function, and the prior distribution of the first objective function is updated based on the obtained information; The next sampling point is determined by the acquisition function. Repeat the above process until the preset number of evaluations is reached to obtain the posterior distribution of the first objective function; The maximum value of the posterior distribution is selected as the optimal parameter combination to complete the optimization and obtain the optimized time series model.

7. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 1, characterized in that, The energy storage battery capacity optimization configuration model specifically includes: Minimize the annualized total cost of the energy storage system over its entire life cycle, maximize the renewable energy absorption rate of the distribution network, and meet the model constraints. The objective function consists of minimizing the annualized total cost of the energy storage system over its entire life cycle and maximizing the renewable energy absorption rate of the distribution network. The annualized total cost C of the energy storage system over its entire life cycle AT Specifically: C AT =(C inv *C RF +C om )+C loss *E loss_annualized ; Among them, C inv C represents the initial investment cost per unit installed capacity of energy storage. RF C represents the capital recovery factor. om C represents the annual operating and maintenance cost per unit capacity of energy storage. loss E represents the energy loss cost per energy storage cycle. loss_annualized This represents the annualized equivalent energy loss of the energy storage system; The specific value of the renewable energy absorption rate N in the distribution network is: Among them, E Gen E represents all electrical energy generated by new energy sources. Cur This refers to the electrical energy generated by discarded renewable energy sources.

8. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 7, characterized in that, The model constraints include: distribution network operation constraints, energy storage system operation constraints, and energy storage capacity and power correlation constraints. The operating constraints of the energy storage system include: charging / discharging power limits, mutual exclusion constraints of energy storage charging and discharging states, and dynamic constraints of energy storage energy.

9. The method for optimizing the power and capacity configuration of energy storage participating in grid peak shaving according to claim 7, characterized in that, The method for obtaining the optimal configuration scheme of the energy storage system is as follows: An initial population is randomly generated, where each individual represents a candidate energy storage configuration. Genetic operations are performed on the initial population to obtain the final population; Based on the objective function, the final population is hierarchically sorted to obtain global non-dominated solutions and corresponding non-dominated sorts at multiple levels. The congestion degree is calculated based on the global non-dominated solution at each level; A new generation of population is selected based on the non-dominated ranking and the crowding degree; Local search is performed based on the new generation population and the penalty function method is used to handle the constraints until the maximum number of iterations is reached or the solution set converges, at which point the algorithm terminates and outputs the optimal configuration scheme of the energy storage system. The optimal configuration scheme for the energy storage system includes: optimal energy storage battery power and optimal capacity configuration parameters.

10. A power and capacity optimization configuration system for energy storage participating in grid peak shaving, used to execute the power and capacity optimization configuration method for energy storage participating in grid peak shaving as described in any one of claims 1-9, characterized in that, include: The module includes a data acquisition module, a data processing module, a time series model building module, a power demand acquisition module, a configuration model building module, and a configuration scheme output module. The data acquisition module is used to acquire historical scenario data of new energy power generation in the target distribution network; The data processing module is used to input the historical scene data into a generative adversarial network and cluster it to obtain high-quality typical scenarios of the new energy power generation. The time-series model construction module is used to construct a time-series model based on the time-series load demand of the target distribution network and the high-quality typical scenario. The power demand acquisition module is used to acquire the power response demand of the target distribution network to the energy storage system based on the time series model. The configuration model construction module is used to establish an energy storage battery capacity optimization configuration model with the lowest annualized cost and the greatest absorption capacity based on the power response requirements. The configuration scheme output module is used to solve the optimal configuration scheme of the energy storage system of the target distribution network by using a non-dominated sorting genetic algorithm based on the energy storage battery capacity optimization configuration model.

Images