Distributed photovoltaic power fluctuation stabilizing control method based on density peak clustering algorithm

Abstract

The invention discloses a distributed photovoltaic power fluctuation stabilization control method based on a comparative density peak value clustering algorithm, which constructs a resource clustering control index system according to the power fluctuation stabilization demand of a power distribution network, and also provides a comparative density peak value clustering algorithm for automatically determining a clustering center. Aggregating the distributed resources into a plurality of clusters; and constructing a power distribution network power fluctuation stabilization control model which takes economy optimization as a target and takes a power fluctuation rate as a coupling constraint, thereby achieving the purposes of optimizing the regulation and control quantity of a resource polymer and solving the problem that distributed photovoltaic power fluctuation is difficult to stabilize.

Description

Technical Field

[0001] The present invention relates to the field of power grid control technology, and in particular to a distributed photovoltaic power fluctuation smoothing control method based on a comparative density peak clustering algorithm; this scheme is mainly used to optimize the resource aggregate regulation quantity, solve the problem that the distributed photovoltaic power generation power in the microgrid of the substation is large and difficult to smooth, and improve the power quality level in the substation. Background Art

[0002] With the rapid growth of distributed photovoltaic installed capacity, its rapid volatility has significantly increased the difficulty of smoothing the power fluctuation of the transmission and distribution network interconnection line. In order to solve the problem of power fluctuation in the distribution network, the regulation of distributed resources and flexible loads has become an increasing research focus. However, the characteristics of large overall number and small single capacity make it difficult to coordinate and regulate. At present, there are three main methods for controlling the power quality of distributed power sources:

[0003] (1) An intelligent control system and method for interactive reactive power support of distributed power sources (202210683576.7): This invention relates to the control technology of the operation of new energy microgrids, and aims to provide an intelligent control system and method for interactive reactive power support of distributed power sources. This invention can achieve the coordination of DG active output and reactive output by using an optimization algorithm under multi-objective constraints, and achieve the maximum active output of DG under the premise of ensuring voltage stability and power factor meeting the requirements, thereby improving economic benefits.

[0004] (2) A distribution network grid connection point voltage control method considering large-scale distributed power sources (202310096838.4): The method comprises: obtaining the voltage value of the common grid connection point, the active power and reactive power of each distributed power source flowing into the common grid connection point; when the difference between the voltage value of the common grid connection point and the reference voltage is greater than a first threshold, calculating the reactive output limit of each distributed power source connected to the common grid connection point; obtaining the total reactive control amount according to the difference between the voltage value of the common grid connection point and the reference voltage; adjusting the reactive control instructions of each power supply branch according to the total reactive control amount and the reactive output limit of each distributed power source, so as to control each distributed power source to respond to the reactive output according to the adjusted reactive control instructions.

[0005] (3) A method and system for power quality control of a distributed photovoltaic power station (202311091631.4): The method comprises: classifying and integrating the application data information of the photovoltaic power station, constructing a photovoltaic power station control strategy database, and then clustering and dividing the photovoltaic power station control strategy database according to a set of power quality influencing factors, and using a deep learning network structure to train and fuse the divided photovoltaic power station application factor classification data sets respectively to generate a photovoltaic power station adaptive controller set; performing similarity matching based on the attribute information of the photovoltaic power station and the photovoltaic power station adaptive controller set, and adaptively controlling the target distributed photovoltaic power station based on the matched target adaptive controller.

[0006] Among them, the contents disclosed in the above documents have the following limitations or defects:

[0007] (1) An intelligent control system and method for interactive reactive support of distributed power sources (202210683576.7) mentions that the intelligent control system involves a large amount of data transmission and calculation. If there is a lack of appropriate security measures, it may face the risk of cyber attacks and data leakage, affecting the security and reliability of the power grid. In addition, existing microgrid systems may use different technologies and standards. When the invention is actually deployed, additional modifications and debugging may be required to make it compatible with existing systems, which may increase the difficulty and cost of implementation.

[0008] (2) A distribution network connection point voltage control method considering large-scale distributed power sources (202310096838.4) mentions the need to obtain multiple parameters (voltage value, active power, reactive power, etc.) and perform multiple steps of calculation to generate control instructions. This method may have poor voltage control effects due to long calculation and response time in the face of rapid changes in the power grid. And this method relies on accurate modeling of distributed power sources and power grid parameters. If the actual situation deviates from the model, it may lead to inaccurate control instructions, which in turn affects the effect of voltage regulation.

[0009] (3) A method and system for controlling the power quality of a distributed photovoltaic power station (202311091631.4): Although the invention is intended to improve real-time controllability, the operation and data processing of deep learning models may require a large amount of computing resources. If the real-time requirements are high, the system may require high-performance computing equipment, which may increase costs. The training of deep learning networks requires a large amount of data and computing resources. If the training data is insufficient or unrepresentative, it may lead to insufficient generalization of the model, that is, the performance of the model in actual applications may not be as good as its performance on the training data.

[0010] In order to solve the problem that the power fluctuation of distributed photovoltaic power generation in the microgrid is large and difficult to smooth, the present invention proposes a distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm. Summary of the invention

[0011] In view of this, the purpose of the present invention is to propose a distributed photovoltaic power fluctuation smoothing control method based on a comparative density peak clustering algorithm that is reliable in implementation, flexible in application and capable of solving the problem of large power fluctuations and difficulty in smoothing distributed photovoltaic power generation in microgrids in substations.

[0012] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0013] A distributed photovoltaic power fluctuation smoothing control method based on a comparison density peak clustering algorithm, comprising:

[0014] Determine the control indicators for distributed photovoltaics;

[0015] Establishing a clustering algorithm model for clustering control indicators and determining the comparative density peak of cluster centers;

[0016] Based on the clustering algorithm model, an aggregation index model is constructed to analyze the comprehensive control performance of clustered aggregates;

[0017] Combined with the characteristics of the controlled object, a distributed photovoltaic power fluctuation smoothing control framework model is constructed to judge the regulation demand and resource grouping. The priority order of resource regulation is determined by combining the aggregation index model. Then, according to the result of the regulation demand judgment and the priority order of resource regulation, the regulation amount is sent to several distributed resources in each resource aggregate. The output size of each resource aggregate is used as the decision variable. While taking into account the effectiveness of power fluctuation smoothing and the regulation economy, the control framework model is solved by the particle swarm algorithm, and then the smoothing control is performed based on the solution result.

[0018] As a possible implementation method, further, the control indicators described in this solution include technical indicators and economic indicators;

[0019] Among them, the technical indicators include the control accuracy R corresponding to distributed photovoltaic n , one or more of the adjustable capacities;

[0020] The control accuracy R n Defined as:

[0021]

[0022] Where: P n,t , Respectively represent the actual output and expected output of adjustable resources; N t is the total number of time periods; E cap is the actual capacity of adjustable resources, where the adjustable resources include distributed photovoltaics and distributed energy storage;

[0023] The adjustable capacity includes an adjustable capacity α n,p With adjustable capacity α n,down , which are defined as:

[0024]

[0025] Where: P n,max , P n,min are the upper and lower bounds of adjustable resource operation respectively; P n,t The real-time operating power of adjustable resources;

[0026] The economic indicators include the unit control costs of distributed photovoltaic units, distributed energy storage units and interruptible flexible loads, which are defined as: and

[0027] As a preferred implementation option, preferably, the present solution establishes a clustering algorithm model for clustering control indicators and determining the comparative density peak of cluster centers, including:

[0028] The following two settings are proposed for the clustering center of the clustering algorithm model:

[0029] Set 1 that the local density of the cluster center is greater than the density of its surrounding neighboring points;

[0030] Set 2 that the distance between the cluster center and the data points with higher density is relatively far;

[0031] Based on setting 1 and setting 2, to determine the cluster center of the data set of the clustering algorithm model, two attributes need to be calculated for each data point, which are: the local density p of data point i i and the relative distance δ i ; Let the data set to be clustered S = {x1, x2, ..., x n},I S =1,2,…,n is the corresponding index set;

[0032] Among them, the local density p i The calculation methods include segment kernel algorithm and Gauss kernel algorithm, and the local density p i The truncated kernel algorithm is defined as follows:

[0033]

[0034] The function in the truncated kernel algorithm is defined as:

[0035]

[0036] In formula (3), i and j are the i-th data point and j-th data point respectively; r ij For data point x i and x j The Euclidean distance between c is the cutoff distance, which is the only parameter of the density peak clustering algorithm and actually plays the role of distance threshold;

[0037] Local density p i The Gaussian kernel algorithm is defined as follows:

[0038]

[0039] Among them, formula (3) is the calculation method of the local density of data i when the data set is large; when the data set is small, in order to reduce the impact of the selection of the truncation distance on the algorithm, the density peak clustering algorithm uses the Gaussian kernel function to estimate the local density of data i, as shown in formula (5);

[0040] Relative distance δ i The algorithm is defined as follows:

[0041]

[0042] According to the description of setting 2 of the density peak clustering algorithm, that is, the distance between the cluster center and the data points with higher density is relatively far, it can be seen that the delta of the cluster center point is i The value must be relatively large; the comparison quantity d is used to replace δ, and the relative distance is displayed in a quantitative comparison manner, thereby highlighting the cluster center, and τ is used to i As the shortest distance between data point i and its lower density point, it is specifically defined as follows:

[0043] d i =δ i -τ i (7)

[0044] Among them, d i is δ i With τ i The number of comparisons is used to assist in identifying potential cluster centers; when d i When the value is large, it indicates that the data point is very close to the data points in the low-density area and farther away from the data points with higher density, which meets the selection characteristics of the cluster center. i The value of is near zero, indicating that the data point has both denser and smaller data points, indicating that the data point is in a certain cluster; if d iIf it is much smaller than zero, the data point is closer to the data point with high density and farther from the data point with low density. Such points are cluster edge points.

[0045] Use Gaussian kernel function to estimate specific density value p i The probability density ρ y (p i ,y), which is defined as follows:

[0046]

[0047] Where: y represents the specific density value p i The possible distance value at the location, the value n is the total number of data points, the parameters a and b are the kernel width values, and the denominator is a normalized factor to ensure Kernel widths a and b are important parameters for probability density estimation and are expressed by p i and d i The standard deviation of is estimated, r p and r d Set to 0.5, the specific definition is as follows:

[0048]

[0049] According to the probability density estimate obtained by formula (9), then at a specific density value p i The expected value and variance of the minimum distance value y are estimated, which are defined as follows:

[0050]

[0051] By simplifying we can get:

[0052]

[0053] In formula (11), n ​​is the total number of data points, p i represents the local density of data point i, parameters a and b represent the kernel width, and d j Indicates the comparison quantity;

[0054] Completed at a specific density value p i After calculating the expected value and variance value of the minimum distance y, the following formula is used to automatically identify the cluster center:

[0055] TH d (p i )=μ y (p i )+3σ y (p i ) (12)

[0056] According to formula (12), we know that: y (pi ) represents the expected value of y, σ y (p i ) represents the standard deviation of y. If the minimum distance value d of the data points i >TH d (p i ), it will be automatically identified as the cluster center.

[0057] As a preferred implementation option, preferably, the input of the clustering algorithm model described in this solution is: data set S = {x1, x2, ..., x n},I S ={1,2,…,n} is the corresponding indicator set, and the output is: clustering result C = {c1,c2,…,c k}, k is the number of cluster centers; the method for identifying cluster centers by the clustering algorithm model includes the following:

[0058] Step 1: Input data S and calculate the distance matrix;

[0059] Step 2: According to formula (5), calculate the local density p of the data point i ;

[0060] Step 3 Calculate the data point δ i And the comparison value τ i , calculate the comparison quantity d according to formula (7) i value;

[0061] Step 4: The density value p at a specific location i Estimated probability density ρ y (p i ,y);

[0062] Step 5 At a specific density p i At , estimate the expected and variance values ​​of the minimum distance value y;

[0063] Step 6: Automatically identify cluster center points using the expectation and variance values ​​of the minimum distance value y at a specific density value;

[0064] Step 7 assigns the non-cluster center points to the clusters with higher density and closest distance;

[0065] Step 8 returns the clustering result C.

[0066] As a better implementation option, preferably, this scheme is based on a clustering algorithm model, and the aggregation index model for analyzing the comprehensive control performance of the aggregate after clustering includes:

[0067] The three indicators of adjustable capacity, control accuracy and unit control cost are aggregated, and the aggregation indicator model is defined as follows:

[0068]

[0069] Where: N k is the number of distributed units in aggregate k; N k,PV 、N k,ESS and N k,II are the number of photovoltaic units, energy storage units and flexible loads in aggregate k, respectively; l, p, q are the units of the three control resources in the aggregate, respectively; is the total controllable capacity of polymer k at time t; R k is the average control accuracy of the aggregate, and the average control accuracy of the three resources in the aggregate is calculated respectively and the average value is taken again; c k It is the aggregate unit control cost of the aggregate, representing the sum of the average control costs of various resources within the aggregate.

[0070] As a preferred implementation option, preferably, the distributed photovoltaic power fluctuation smoothing control framework model described in this scheme takes the minute fluctuation rate of the power of the grid point under the main transformer of the distribution network as the control target; the implementation method of the distributed photovoltaic power fluctuation smoothing control framework model includes a first stage and a second stage;

[0071] Among them, the distributed photovoltaic power fluctuation smoothing control framework model performs the work of control demand judgment and resource grouping in the first stage, which includes: determining the reference value of the power of the downstream point according to the real-time data to generate the target value in the current control cycle; then, in the same control cycle, the fluctuation rate of the power of the downstream point is calculated according to the control step, and it is used as the coupling constraint of the minute-level AGC optimization control and the power fluctuation smoothing target, so as to calculate the control demand of the distribution network in the current period; in each control cycle, the power of the downstream point at the first moment is selected as the control reference value, and the resources are grouped at the same time, and the priority order of resource control is determined by the aggregation indicator model;

[0072] The second stage is to judge the control demand results in the first stage, and send the control quantity to several distributed resources in the resource aggregate according to the priority order of resource control. The output size of each resource aggregate is used as the decision variable, taking into account the effectiveness of power fluctuation smoothing and control economy, and using the particle swarm algorithm to solve the optimization control model. Among them, each time the optimization model is solved, it is judged whether the current control cycle is over according to the control step size and the number of cycle solutions, and then it is judged whether it is necessary to update the clustering index and perform resource clustering and optimization control in the next cycle.

[0073] As a preferred implementation option, preferably, the control method described in this solution also includes:

[0074] For the centralized optimization problem without resource clustering, the control costs of the three control resources, photovoltaic units, energy storage units and flexible loads, and the main grid power purchase cost are taken as components of the optimization target, which are defined as follows:

[0075]

[0076] Among them, in the formula: C PV (t), C ESS (t) and C IL (t) are the total control cost of distributed photovoltaic, total control cost of distributed energy storage and controllable cost of flexible load at the tth moment; Cgrid(t) is the main grid power purchase cost at the tth moment; The unit electricity purchase cost of the power grid at the current moment; is the power of the downstream network at the tth moment, that is, the required main network power, which can be solved by power flow calculation under the premise of determining the network topology and resource allocation.

[0077] As a preferred implementation option, preferably, this solution also includes:

[0078] According to the priority of resource regulation obtained by clustering, equation (14) is modified to construct an economic optimization objective with the aggregate regulation amount as the decision variable, which is defined as follows:

[0079]

[0080] Where: K is the number of aggregates obtained after clustering; C cluster,k (t) is the control cost of the kth aggregate at the tth moment, which can be quickly calculated according to the economic aggregate control instruction decomposition method. The main grid purchase electricity cost C grid (t)The calculation method remains unchanged.

[0081] As a preferred implementation option, preferably, this solution also includes:

[0082] In addition to considering the conventional system power flow constraints, the operation constraints of distributed control resources, active power balance constraints and power fluctuation rate constraints of the downstream network are also considered. The constraints of the aggregation index model are as follows:

[0083] (1) Distributed energy storage operation constraints

[0084] The distributed energy storage after aggregation has a strong control response capability and is a major control resource for smoothing the fluctuation of new energy. In the actual operation of the aggregate, it is generally necessary to meet the charging and discharging power constraints and capacity constraints:

[0085]

[0086] In addition, the capacity constraint should be calculated and analyzed in combination with the dynamic characteristics of the charging and discharging capacity of cluster energy storage:

[0087]

[0088] Where: ηin and ηout are the charging and discharging efficiencies of the energy storage unit respectively; assuming that all energy storage units in the system are of the same model, the cluster charging and discharging efficiencies are also the same;

[0089] (2) Distributed PV operation constraints

[0090] Distributed photovoltaics generally work near the maximum power point and only abandon light appropriately when the absorption is insufficient, that is, as a downward adjustment resource to participate in the power balance of the distribution network; in this process, distributed photovoltaics should meet the following constraints:

[0091]

[0092] Where: is the total amount of distributed photovoltaic regulation in aggregate k at time t; The maximum controllable amount of the lth distributed unit;

[0093] (3) Power balance constraints

[0094]

[0095] (4) Interruptible load operation constraints

[0096] Considering the aggregate output constraint of the interruptible load aggregate and the daily interruption number constraint,

[0097]

[0098] Where: They are the current cumulative interruption times and the maximum daily interruption times of the interruptible load unit respectively;

[0099] (5) Power fluctuation rate constraints at the downstream network

[0100] Power fluctuation rate R at lower grid point vol It represents the power change rate of the current moment relative to the previous moment. The power fluctuation rate constraint condition of the downstream network point is determined according to the distribution network structure and resource parameters. The fluctuation rate R vol Limit to the target volatility R B Within:

[0101]

[0102] In order to simplify the optimization control model and establish a direct connection between the power fluctuation rate constraint of the downstream grid point and the economic objective function, this paper adds it to the economic objective function in the form of a penalty function. In the optimization process, a larger value is set for the coefficient β that characterizes the intensity of the penalty function, and the economic objective function that considers the power fluctuation smoothing effect is obtained:

[0103]

[0104] Since the system power flow is a nonlinear constraint, the particle swarm optimization algorithm PSO is used for rapid solution to meet the needs of rapid smoothing control of distributed photovoltaic power fluctuations.

[0105] As a preferred implementation option, preferably, this solution also includes:

[0106] The total control amount of each resource aggregate can be obtained by solving the above optimization control model, and it needs to be further decomposed quickly. Compared with the conventional decomposition method of allocating according to capacity ratio, the present invention takes economic efficiency as the priority decision basis and constructs a distributed resource aggregate control instruction fast decomposition model to meet the needs of rapid smoothing of distributed photovoltaic power fluctuations. Its objective function and constraints are as follows:

[0107]

[0108] The essence of the optimization model is to sort the adjustable resources in the aggregate according to the economic indicators, establish a piecewise functional relationship between the total control cost of the aggregate and the total control quantity (15), and finally realize the rapid decomposition of the control instructions with the total control quantity of the aggregate as a hard constraint.

[0109] By adopting the above-mentioned technical scheme, the present invention has the following beneficial effects compared with the prior art: the scheme constructs a resource clustering control index system according to the power fluctuation smoothing demand of the distribution network, and at the same time, proposes a comparative density peak clustering algorithm that automatically determines the clustering center, so as to aggregate the distributed resources into several clusters; constructs a distribution network power fluctuation smoothing control model with economic optimization as the goal and power fluctuation rate as the coupling constraint, so as to optimize the resource aggregate control quantity and solve the problem of difficulty in smoothing distributed photovoltaic power fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0111] Figure 1It is a schematic diagram of the processing flow of the power fluctuation smoothing control framework in this scheme;

[0112] Figure 2 It is the IEEE-33 node distribution network with distributed control resources in this scheme;

[0113] Figure 3 This is the power comparison of the main transformer before and after the control of this solution;

[0114] Figure 4 It is the comparison of power fluctuation rate of the main transformer before and after the control of this scheme;

[0115] Figure 5 It is the relationship curve between the clustering silhouette coefficient and the number of clusters of this scheme;

[0116] Figure 6 This is a comparison chart of polymer output before and after control of this scheme;

[0117] Figure 7 It is the control quantity-control cost curve of the polymer before and after the economic optimization of this scheme. DETAILED DESCRIPTION

[0118] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only used to illustrate the present invention, but are not intended to limit the scope of the present invention. Similarly, the following examples are only partial embodiments of the present invention rather than all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0119] The distributed photovoltaic power fluctuation smoothing control method of this embodiment scheme is based on the comparison density peak clustering algorithm, which includes the following:

[0120] 1. Main indicators of this embodiment

[0121] Distributed resources in the distribution network have the characteristics of large overall number and small single capacity. If they are directly centrally controlled, there will be problems of large amount of calculation and high variable dimension. They need to be properly aggregated first to alleviate the problem of difficulty in solving. Therefore, clustering control indicators for multiple distributed resources and flexible loads are determined to aggregate and cluster the dispersed resources.

[0122] 1.1 Technical indicators

[0123] (1) Control accuracy

[0124] The operating characteristics and control methods of the adjustable resources of the power grid are different, and there are also individual differences between the same type of resources. In actual regulation applications, the accuracy of various resources will also be different. Therefore, this embodiment regards the regulation accuracy as one of the technical indicators of regulation. Regulation accuracy R n It can be expressed as

[0125]

[0126] Where: P n,t , Represent the actual output of adjustable resources and the expected output based on the regulation method; N t is the total number of time periods; E cap The actual capacity of the adjustable resource.

[0127] (2) Adjustable capacity

[0128] Adjustable capacity represents the adjustable resource control capability, which is distinguished according to the control scenario and control object. The adjustable capacity of distributed photovoltaic, energy storage and interruptible flexible load (IL) is calculated in the upward and downward control scenarios respectively. The upward control refers to increasing the output of source-side resources (reducing the load-side resource load). n,p With adjustable capacity α n,down It can be calculated uniformly by the following formula:

[0129]

[0130] Where: P n,max , P n,min are the upper and lower bounds of adjustable resource operation respectively; P n,t The real-time operating power of adjustable resources.

[0131] In order to ensure the photovoltaic absorption level, distributed photovoltaics should be operated in maximum power point tracking mode, so only its downward adjustment capability is considered. For interruptible loads, in order to ensure power supply reliability and power quality, its load level is maintained near the rated power for a long time, so only its upward adjustment capability is considered. Distributed energy storage has two operating states, charging and discharging, and can have both upward and downward adjustment capabilities. The upward adjustment resources considered in this embodiment include interruptible flexible loads and distributed energy storage in the discharging state, and the downward adjustment resources include distributed photovoltaics and distributed energy storage in the charging state.

[0132] 1.2 Economic indicators

[0133] The regulation cost of adjustable resources is affected by the construction investment and regulation quotation of different stakeholders, the amount of power regulation and the length of time involved in regulation. This embodiment takes the unit regulation cost of distributed photovoltaic units, distributed energy storage units and interruptible flexible load individuals as economic indicators and is defined as and

[0134] 2. Comparative density peak clustering algorithm to automatically determine cluster centers

[0135] This embodiment proposes a comparative density peak clustering algorithm for automatically determining cluster centers. The algorithm first uses a distance comparison quantity to replace the original distance parameter, and then automatically selects the cluster center through a two-dimensional interval estimation method, thereby automating the clustering process.

[0136] 2.1 Density peak clustering algorithm is a new type of clustering algorithm that can discover non-convex clusters. Its core idea is based on two important assumptions about the cluster center.

[0137] Assumption 1: The local density of the cluster center is greater than the density of its surrounding neighbors.

[0138] Assumption 2: The cluster center is relatively far away from data points with higher density than it.

[0139] According to these two settings, the cluster center should have both large local density and large relative distance. To determine the cluster center of a data set, for each data point, two attributes need to be calculated: the local density p of data point i and the i and the relative distance δ i . Let the data set to be clustered S = {x1, x2, ..., x m},I S =1,2,…,n is the corresponding indicator set.

[0140] Definition 1: Local density p i (Two calculation methods: truncated kernel algorithm and Gaussian kernel algorithm).

[0141] The truncated kernel algorithm is defined as:

[0142]

[0143] The function:

[0144]

[0145] In formula (3), i and j are the i-th data point and j-th data point respectively; r ij For data point x i and x jThe Euclidean distance between c It is the cutoff distance, which is the only parameter of the density peak clustering algorithm and actually plays the role of distance threshold.

[0146] The Gaussian kernel algorithm is defined as:

[0147]

[0148] Formula (3) is the calculation method of the local density of data i when the data set is large; when the data set is small, in order to reduce the impact of the selection of the truncation distance on the algorithm, the DPC algorithm uses the Gaussian kernel function to estimate the local density of data i, as shown in formula (5).

[0149] Definition 2: Relative distance δ i :

[0150]

[0151] According to the description of the second assumption of the density peak clustering algorithm, that is, the distance between the cluster center and the data points with higher density is relatively far, it can be seen that the delta of the cluster center point is i The value must be relatively large. This embodiment adopts the comparison quantity d to replace δ, and realizes the quantitative comparison method to display the relative distance, thereby highlighting the cluster center. i As the shortest distance between data point i and its lower density point, it is specifically defined as follows:

[0152] d i =δ i -τ i (7)

[0153] d i is δ i With τ i The number of comparisons helps to identify potential cluster centers. i When the value is large, it indicates that the data point is very close to the data points in the low-density area and farther away from the data points with higher density, which meets the selection characteristics of the cluster center. i The value of is near zero, indicating that the data point has both denser and less dense data points, indicating that the data point is in a certain cluster. i If it is much smaller than zero, the data point is closer to the data points with high density and farther away from the data points with low density. Such points are cluster edge points.

[0154] 2.2 This embodiment adopts a specific statistic to realize the automatic identification of cluster center points. The Gaussian kernel function is used to estimate the specific density value p i The probability density ρ y (p i ,y), which is defined as follows:

[0155]

[0156] Where: y represents the specific density value p i The value n is the total number of data points, and the parameters a and b are the kernel width values. The denominator is a normalization factor that ensures Kernel widths a and b are important parameters for probability density estimation and are expressed by p i and d i The standard deviation of is estimated, r p and r d Set to 0.5, the specific definition is as follows:

[0157]

[0158] According to the probability density estimate obtained by formula (9), then at a specific density value p i The expected value and variance of the minimum distance value y are estimated, which are defined as follows:

[0159]

[0160] By simplifying we can get:

[0161]

[0162] In formula (11), n ​​is the total number of data points, p i represents the local density of data point i, parameters a and b represent the kernel width, and d j Indicates the comparison quantity.

[0163] Completed at a specific density value p i After calculating the expected value and variance value of the minimum distance y, the following formula is used to automatically identify the cluster center:

[0164] TH d (p i )=μ y (p i )+3σ y (p i ) (12)

[0165] According to formula (12), we know that: y (p i ) represents the expected value of y, σ y (p i ) represents the standard deviation of y. If the minimum distance value d of the data points i >TH d (p i ), it will be automatically identified as the cluster center.

[0166] The specific implementation steps of the clustering algorithm of this embodiment are as follows:

[0167] Input: Dataset S = {x1, x2, ..., x n},I S ={1,2,…,n} is the corresponding indicator set;

[0168] Output: Clustering result C = {c1, c2, ..., c k}, k is the number of cluster centers.

[0169] Step 1: Input data S and calculate the distance matrix;

[0170] Step 2: According to formula (5), calculate the local density p of the data point i ;

[0171] Step 3 Calculate the data point δ i And the comparison value τ i , calculate the comparison quantity d according to formula (7) i value;

[0172] Step 4: The density value p at a specific location i Estimated probability density ρ y (p i ,y);

[0173] Step 5 At a specific density p i At , estimate the expected and variance values ​​of the minimum distance value y;

[0174] Step 6: Automatically identify cluster center points using the expectation and variance values ​​of the minimum distance value y at a specific density value;

[0175] Step 7 assigns the non-cluster center points to the clusters with higher density and closest distance;

[0176] Step 8 returns the clustering result C.

[0177] 2.3 In order to facilitate the analysis of the comprehensive control performance of the aggregate after clustering, and to improve the feasibility of multiple control resources participating in distribution network control while achieving dimensionality reduction of decision variables, an aggregation index model is established for the aggregate containing multiple control resources.

[0178] The three indicators of adjustable capacity, control accuracy and unit control cost are aggregated respectively:

[0179]

[0180] Where: N k is the number of distributed units in aggregate k; N k,PV 、N k,ESS and N k,IIare the number of photovoltaic units, energy storage units and flexible loads in aggregate k, respectively; l, p, q are the units of the three control resources in the aggregate, respectively; is the total controllable capacity of polymer k at time t; R k is the average control accuracy of the aggregate, and the average control accuracy of the three resources in the aggregate is calculated respectively and the average value is taken again; c k It is the aggregate unit control cost of the aggregate, representing the sum of the average control costs of various resources within the aggregate.

[0181] 3. Distributed photovoltaic power fluctuation control strategy

[0182] In order to take into account both the economic efficiency of regulation and the effectiveness of stabilization, this embodiment scheme realizes the rapid decomposition of aggregate regulation instructions through the principle of economic allocation, solves the problem of power fluctuation at the main transformer outlet caused by the randomness of distributed photovoltaics, and improves the source-load interaction capability of the distribution network.

[0183] 3.1 Since the capacity of distributed photovoltaic units is small and widely distributed, this embodiment takes the minute fluctuation rate of the power of the distribution network main transformer as the control target and constructs a power fluctuation smoothing control framework, such as Figure 1 shown.

[0184] In the first stage of the control framework, the work of determining the control demand and grouping resources must be completed. Specifically, the baseline value of the power of the downstream network must be determined based on real-time data to generate the target value in the current control cycle. Then, in the same control cycle, the fluctuation rate of the power of the downstream network is calculated according to the control step, and it is used as a coupling constraint for the minute-level AGC optimization control and the power fluctuation smoothing target, thereby calculating the control demand of the distribution network in the current period. In each control cycle, the power of the downstream network at the first moment is selected as the control baseline value, and resource grouping is performed at the same time. The priority order of resource regulation is determined by the aggregation model as a reliable basis for the next stage of optimization control.

[0185] In the second stage of the control framework, according to the results of the first stage control demand judgment, the control quantity is sent to several distributed resources in the resource aggregate according to the priority order of resource control. The output size of each resource aggregate is used as the decision variable, taking into account the effectiveness of power fluctuation smoothing and control economy, and using particle swarm optimization (PSO) to solve the optimization control model. Each time the optimization model is solved, the number of solutions N is calculated according to the control step size and cycle. c Determine whether the current control cycle has ended, and then determine whether it is necessary to update the clustering indicators and perform resource clustering and optimization control for the next cycle.

[0186] 3.2 For the centralized optimization problem without resource clustering, the control costs of the three control resources and the main grid power purchase cost are taken as components of the optimization objective, namely

[0187]

[0188] Where: C PV (t), C ESS (t) and C IL (t) are the total control cost of distributed photovoltaic, total control cost of distributed energy storage and controllable cost of flexible load at the tth moment; C grid (t) is the main grid electricity purchase cost at time t; The unit electricity purchase cost of the power grid at the current moment; is the power of the downstream network at the tth moment, that is, the required main network power, which can be solved by power flow calculation under the premise of determining the network topology and resource allocation.

[0189] The number of decision variables of the distributed resource clustering method proposed in this embodiment is greatly reduced. Formula (14) is modified according to the resource control priority obtained by clustering, and an economic optimization target with the aggregate control amount as the decision variable is constructed:

[0190]

[0191] Where: K is the number of aggregates obtained after clustering; C cluster,k (t) is the control cost of the kth aggregate at the tth moment, which can be quickly calculated based on the economic aggregate control instruction decomposition method. The main grid power purchase cost C grid (t)The calculation method remains unchanged.

[0192] 3.3 In addition to considering the conventional system power flow constraints, this embodiment also considers the operation constraints of distributed control resources, active power balance constraints and power fluctuation rate constraints of downstream nodes. The constraints of the aggregation model are as follows:

[0193] (1) Distributed energy storage operation constraints

[0194] The distributed energy storage after aggregation has a strong control response capability and is a major control resource for smoothing the fluctuation of new energy. In the actual operation of the aggregate, it is generally necessary to meet the charging and discharging power constraints and capacity constraints:

[0195]

[0196] In addition, the capacity constraint should be calculated and analyzed in combination with the dynamic characteristics of the charging and discharging capacity of cluster energy storage:

[0197]

[0198] Where: ηin, ηout are the charging and discharging efficiencies of the energy storage unit. Assuming that all energy storage units in the system are of the same model, the cluster charging and discharging efficiencies are also the same.

[0199] (2) Distributed PV operation constraints

[0200] Distributed photovoltaics generally work near the maximum power point and only abandon the light appropriately when the power consumption is insufficient, that is, they participate in the power balance of the distribution network as a downward adjustment resource. In this process, distributed photovoltaics should meet the following constraints:

[0201]

[0202] Where: is the total amount of distributed photovoltaic regulation in aggregate k at time t; The maximum controllable amount of the lth distributed unit.

[0203] (3) Power balance constraints

[0204]

[0205] (4) Interruptible load operation constraints

[0206] Considering the aggregate output constraint of the interruptible load aggregate and the daily interruption number constraint,

[0207]

[0208] Where: They are the current cumulative interruption times and the daily maximum interruption times of the interruptible load unit.

[0209] (5) Power fluctuation rate constraints at the downstream network

[0210] Power fluctuation rate R at lower grid point vol It represents the power change rate of the current moment relative to the previous moment. The power fluctuation rate constraint condition of the downstream network point is determined according to the distribution network structure and resource parameters. The fluctuation rate R vol Limit to the target volatility R B Within:

[0211]

[0212] In order to simplify the optimization control model and establish a direct connection between the power fluctuation rate constraint of the downstream grid point and the economic objective function, this paper adds it to the economic objective function in the form of a penalty function. In the optimization process, a larger value is set for the coefficient β that characterizes the intensity of the penalty function, and the economic objective function that considers the power fluctuation smoothing effect is obtained:

[0213]

[0214] Since the system power flow is a nonlinear constraint, this paper uses the intelligent optimization algorithm PSO for rapid solution, which can meet the needs of rapid smoothing control of distributed photovoltaic power fluctuations.

[0215] 3.4 The total control amount of each resource aggregate can be obtained by solving the above optimization control model, and it needs to be further decomposed quickly. Compared with the conventional decomposition method of allocating according to capacity ratio, this embodiment scheme takes economic efficiency as the priority decision basis and constructs a distributed resource aggregate control instruction fast decomposition model to meet the needs of rapid smoothing of distributed photovoltaic power fluctuations. Its objective function and constraints are as follows:

[0216]

[0217] The essence of the optimization model is to sort the adjustable resources in the aggregate according to the economic indicators, establish a piecewise functional relationship between the total control cost of the aggregate and the total control quantity (15), and finally realize the rapid decomposition of the control instructions with the total control quantity of the aggregate as a hard constraint.

[0218] Combination Figure 2 As shown, this embodiment is based on the standard IEEE-33 node distribution network system, where distributed photovoltaic access nodes 5, 8, 10, the scale is 100; distributed energy storage access 7, 13, 20, respectively, the scale is 20; interruptible flexible load access 26, 28, 32, respectively, the scale is 40, assuming that node 4 is the main transformer downlink point, establish as follows Figure 2 In the distribution network topology shown, the normal state of charge range of the distributed energy storage unit is selected as 0.25~0.95, the initial state of charge is 0.4, and the charging and discharging efficiency is 94%.

[0219] In the optimization control process, the control period T = 5min and the control step length T c =1min, perform resource aggregation at the initial moment of the control cycle, and use the power of the downstream network point at this time as the control reference value. Based on the resource clustering control index proposed by the present invention, combined with the actual production and living conditions of multiple types of resources, the data boundaries of three types of clustering indexes are determined respectively, as shown in Table 1.

[0220] Table 1 Distributed resource clustering indicators

[0221] Resource Type Adjustable capacity / kW Control accuracy / % Unit control cost / yuan Distributed Photovoltaic 0～25 92～98 0.40～0.80 Distributed Energy Storage 0～10 96～98 0.25～0.45 Interruptible flexible load 0～30 94～99 0.80～1.20

[0222] In this embodiment, one control cycle is selected for simulation analysis. The control cycle T = 10 min and the control step length T c =1min, so the number of control steps is N c=10. The grid power of the main transformer at the initial moment of the control cycle is used as the grid power leveling reference value in this control cycle. According to the photovoltaic prediction data, the grid power reference value P of the main transformer is calculated. ref =3235.98kW. The PSO algorithm is used to solve the optimization control model, and the power and power fluctuation rate of the main transformer are compared before and after the optimization control. Figure 3 , 4 As shown in the figure, the power of the network point at each moment in the control cycle is controlled at the power reference value P ref Nearby, the maximum fluctuation before and after control dropped from 201.03kW to 84.11kW; the minute fluctuation rate of the power at the main transformer outlet within the control period was lower than 1.9%, and there were 3 moments before control when it exceeded 1.9%. The absolute value of the maximum fluctuation rate after control was 1.898%, which was 3.223% lower than before control, achieving a good control effect.

[0223] For the distribution network with large-scale access to distributed photovoltaics, this embodiment adopts a comparative density peak clustering algorithm that automatically determines the cluster center to cluster distributed resources. First, at the beginning of the control cycle, the real-time information parameters of the adjustable resources are collected and the data set is normalized; then, the silhouette coefficient method is used to evaluate the resource aggregation effect under different cluster numbers. The silhouette coefficient curve is as follows: Figure 5 As shown, it can be seen that when the number of clusters K = 2, the silhouette coefficient reaches its maximum value, and the resource clustering effect is optimal at this time.

[0224] Resource clustering in the power fluctuation smoothing control problem can not only reduce the dimension of decision variables of the optimization problem and determine the priority of adjustable resources, but also minimize the regulation cost. Figure 6 The output curves of the two polymers before and after polymerization are shown. In this figure, it can be observed that in this control cycle, the regulation amount of polymer 1 is much greater than that of polymer 2. The output curve of polymer 2 basically overlaps before and after regulation. This is because the power fluctuation of the downstream grid point caused by photovoltaic fluctuation is small, and the total regulation demand does not exceed the regulation capacity of polymer 1. This indirectly shows that the regulation priority of polymer 1 is higher. Combined with Figure 4 According to the analysis, when the power of the downstream grid fluctuates downward, the output of aggregate 1 is reduced, which is equivalent to increasing the load power in the distribution network, thereby increasing the power purchased by the main grid. For example, at t = 2min, the output of aggregate 1 is reduced by 42.3852kW, while the reduction of aggregate 2 is only 0.0859kW, which limits the power fluctuation rate of the downstream grid to the allowable range.

[0225] To achieve power fluctuation smoothing, it is necessary to further decompose the aggregate control instructions. This embodiment optimizes the resource control amount based on the principle of economy. Figure 7is the control quantity-control cost curve of different polymers before and after economic optimization under two control scenarios. Figure 7 It can be seen that: regardless of whether economic optimization is performed or not, the total regulation cost of aggregate 1 is always lower than that of aggregate 2; whether the same capacity of the same aggregate is regulated in an upward or downward regulation scenario, the total regulation cost of the aggregate after economic optimization is lower. For example, when aggregate 1 is reduced by 300kW, the regulation cost required before economic optimization is 150.23 yuan, while the resource regulation cost after economic optimization drops to 107.24 yuan.

[0226] Table 2 shows the regulation amount of various resources in the aggregate. According to the control command economic decomposition strategy, energy storage is in a priority regulation position, and distributed photovoltaic and interruptible load are adjusted last in the down-regulation scenario and up-regulation scenario respectively. For example: at t = 3min, the up-regulation amount of 78.3528kW required by aggregate 1 exceeds the regulation capacity of 53.6427kW of distributed energy storage, so the interruptible load is required to supplement the remaining required regulation amount of 24.7101kW; in the optimization period of t = 4min, the down-regulation amount required by aggregate 1 is 102.2835kW, which exceeds the adjustable range of energy storage, so the distributed photovoltaic is down-regulated by 51.3728kW. In summary, resource aggregation and command decomposition realize the priority judgment of adjustable resources, which meets the resource aggregation goal.

[0227] Table 2 Aggregate distributed resource control quantity

[0228]

[0229] This embodiment compares the time for solving the optimization control model before and after clustering and the total economic cost of the optimization control, and the results are shown in Tables 3 and 4. Obviously, the optimization control after clustering has great advantages in both solution speed and control cost, because resource aggregation realizes the judgment of the control priority of massive distributed resources, reduces the variable dimension, and significantly improves the utilization efficiency of resources and the reliability and economy of control, which meets the ultimate goal of resource clustering.

[0230] Table 3 Comparison of optimization time before and after clustering

[0231]

[0232] Table 4 Comparison of control costs before and after clustering

[0233]

[0234] The above descriptions are only some embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A distributed photovoltaic power fluctuation smoothing control method based on comparative density peak clustering algorithm, characterized in that: It includes: Determine the control indicators for distributed photovoltaics; Establishing a clustering algorithm model for clustering control indicators and determining the comparative density peak of cluster centers; Based on the clustering algorithm model, an aggregation index model is constructed to analyze the comprehensive control performance of clustered aggregates; Combined with the characteristics of the controlled object, a distributed photovoltaic power fluctuation smoothing control framework model is constructed to judge the regulation demand and resource grouping. The priority order of resource regulation is determined by combining the aggregation index model. Then, according to the result of the regulation demand judgment and the priority order of resource regulation, the regulation amount is sent to several distributed resources in each resource aggregate. The output size of each resource aggregate is used as the decision variable. While taking into account the effectiveness of power fluctuation smoothing and the regulation economy, the control framework model is solved by the particle swarm algorithm, and then the smoothing control is performed based on the solution result.

2. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 1 is characterized in that: The control indicators include technical indicators and economic indicators; Among them, the technical indicators include the control accuracy R corresponding to distributed photovoltaic n , one or more of the adjustable capacities; The control accuracy R n Defined as: Where: P n,t , Respectively represent the actual output and expected output of adjustable resources; N t is the total number of time periods; E cap is the actual capacity of adjustable resources, where the adjustable resources include distributed photovoltaics and distributed energy storage; The adjustable capacity includes an adjustable capacity α n,p With adjustable capacity α n,down , which are defined as: Where: P n,max , P n,min are the upper and lower bounds of adjustable resource operation respectively; P n,t The real-time operating power of adjustable resources; The economic indicators include the unit control costs of distributed photovoltaic units, distributed energy storage units and interruptible flexible loads, which are defined as: and 3. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 2 is characterized in that: The clustering algorithm model for clustering the control indicators and determining the comparative density peak of the cluster center includes: The following two settings are proposed for the clustering center of the clustering algorithm model: Set 1 that the local density of the cluster center is greater than the density of its surrounding neighboring points; Set 2 that the distance between the cluster center and the data points with higher density is relatively far; Based on setting 1 and setting 2, to determine the cluster center of the data set of the clustering algorithm model, two attributes need to be calculated for each data point, which are: the local density p of data point i i and the relative distance δ i ; Let the data set to be clustered S = {x1, x2, ..., x n },I S =1,2,…,n is the corresponding index set; Among them, the local density p i The calculation methods include segment kernel algorithm and Gauss kernel algorithm, and the local density p i The truncated kernel algorithm is defined as follows: The function in the truncated kernel algorithm is defined as: In formula (3), i and j are the i-th data point and j-th data point respectively; r ij For data point x i and x j The Euclidean distance between c is the cutoff distance, which is the only parameter of the density peak clustering algorithm and actually plays the role of distance threshold; Local density p i The Gaussian kernel algorithm is defined as follows: Among them, formula (3) is the calculation method of the local density of data i when the data set is large; when the data set is small, in order to reduce the impact of the selection of the truncation distance on the algorithm, the density peak clustering algorithm uses the Gaussian kernel function to estimate the local density of data i, as shown in formula (5); Relative distance δ i The algorithm is defined as follows: According to the description of setting 2 of the density peak clustering algorithm, that is, the distance between the cluster center and the data points with higher density is relatively far, it can be seen that the delta of the cluster center point is i The value must be relatively large; the comparison quantity d is used to replace δ, and the relative distance is displayed in a quantitative comparison manner, thereby highlighting the cluster center, and τ is used to i As the shortest distance between data point i and its lower density point, it is specifically defined as follows: d i =d i -t i (7) Among them, d i is δ i With τ i The number of comparisons is used to assist in identifying potential cluster centers; when d i When the value is large, it indicates that the data point is very close to the data points in the low-density area and farther away from the data points with higher density, which meets the selection characteristics of the cluster center. i The value of is near zero, indicating that the data point has both denser and smaller data points, indicating that the data point is in a certain cluster; if d i If it is much smaller than zero, the data point is closer to the data point with high density and farther from the data point with low density. Such points are cluster edge points. Use Gaussian kernel function to estimate specific density value p i The probability density ρ y (p i ,y), which is defined as follows: Where: y represents the specific density value p i The possible distance value at the location, the value n is the total number of data points, the parameters a and b are the kernel width values, and the denominator is a normalized factor to ensure Kernel widths a and b are important parameters for probability density estimation and are expressed by p i and d i The standard deviation of is estimated, r p and r d Set to 0.5, the specific definition is as follows: According to the probability density estimate obtained by formula (9), then at a specific density value p i The expected value and variance of the minimum distance value y are estimated, which are defined as follows: By simplifying we can get: In formula (11), n ​​is the total number of data points, p i represents the local density of data point i, parameters a and b represent the kernel width, and d j Indicates the comparison quantity; Completed at a specific density value p i After calculating the expected value and variance value of the minimum distance y, the following formula is used to automatically identify the cluster center: TH d (p i )=μ y (p i )+3σ y (p i ) (12) According to formula (12), we know that: y (p i ) represents the expected value of y, σ y (p i ) represents the standard deviation of y. If the minimum distance value d of the data points i >TH d (p i ), it will be automatically identified as the cluster center.

4. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 3 is characterized in that: The input of the clustering algorithm model is: data set S = {x1, x2, ..., x n },I S ={1,2,…,n} is the corresponding indicator set, and the output is: clustering result C = {c1,c2,…,c k }, k is the number of cluster centers; the method for identifying cluster centers by the clustering algorithm model includes the following: Step 1: Input data S and calculate the distance matrix; Step 2: According to formula (5), calculate the local density p of the data point i ; Step 3 Calculate the data point δ i And the comparison value τ i , calculate the comparison quantity d according to formula (7) i value; Step 4: The density value p at a specific location i Estimated probability density ρ y (p i ,y); Step 5 At a specific density p i At , estimate the expected and variance values ​​of the minimum distance value y; Step 6: Automatically identify cluster center points using the expectation and variance values ​​of the minimum distance value y at a specific density value; Step 7 assigns the non-cluster center points to the clusters with higher density and closest distance; Step 8 returns the clustering result C.

5. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm as claimed in claim 3 or 4, characterized in that: Based on the clustering algorithm model, the aggregation index model for analyzing the comprehensive control performance of the aggregate after clustering is constructed, including: The three indicators of adjustable capacity, control accuracy and unit control cost are aggregated, and the aggregation indicator model is defined as follows: Where: N k is the number of distributed units in aggregate k; N k,PV 、N k,ESS and N k,II are the number of photovoltaic units, energy storage units and flexible loads in aggregate k, respectively; l, p, q are the units of the three control resources in the aggregate, respectively; is the total controllable capacity of polymer k at time t; R k is the average control accuracy of the aggregate, and the average control accuracy of the three resources in the aggregate is calculated respectively and the average value is taken again; c k It is the aggregate unit control cost of the aggregate, representing the sum of the average control costs of various resources within the aggregate.

6. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 5 is characterized in that: The distributed photovoltaic power fluctuation smoothing control framework model takes the minute fluctuation rate of the power of the distribution network main transformer as the control target; the implementation method of the distributed photovoltaic power fluctuation smoothing control framework model includes a first stage and a second stage; Among them, the distributed photovoltaic power fluctuation smoothing control framework model performs the work of control demand judgment and resource grouping in the first stage, which includes: determining the reference value of the power of the downstream point according to the real-time data to generate the target value in the current control cycle; then, in the same control cycle, the fluctuation rate of the power of the downstream point is calculated according to the control step, and it is used as the coupling constraint of the minute-level AGC optimization control and the power fluctuation smoothing target, so as to calculate the control demand of the distribution network in the current period; in each control cycle, the power of the downstream point at the first moment is selected as the control reference value, and the resources are grouped at the same time, and the priority order of resource control is determined by the aggregation indicator model; The second stage is to judge the control demand results in the first stage, and send the control quantity to several distributed resources in the resource aggregate according to the priority order of resource control. The output size of each resource aggregate is used as the decision variable, taking into account the effectiveness of power fluctuation smoothing and control economy, and using the particle swarm algorithm to solve the optimization control model. Among them, each time the optimization model is solved, it is judged whether the current control cycle is over according to the control step size and the number of cycle solutions, and then it is judged whether it is necessary to update the clustering index and perform resource clustering and optimization control in the next cycle.

7. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 6 is characterized in that: The control method further comprises: For the centralized optimization problem without resource clustering, the control costs of the three control resources, photovoltaic units, energy storage units and flexible loads, and the main grid power purchase cost are taken as components of the optimization target, which are defined as follows: Among them, in the formula: C PV (t), C ESS (t) and C IL (t) are the total control cost of distributed photovoltaic, total control cost of distributed energy storage and controllable cost of flexible load at the tth moment; C grid (t) is the main grid electricity purchase cost at time t; The unit electricity purchase cost of the power grid at the current moment; is the power of the downstream network at the tth moment, that is, the required main network power, which can be solved by power flow calculation under the premise of determining the network topology and resource allocation.

8. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 7 is characterized in that: It also includes: According to the priority of resource regulation obtained by clustering, equation (14) is modified to construct an economic optimization objective with the aggregate regulation amount as the decision variable, which is defined as follows: Where: K is the number of aggregates obtained after clustering; C cluster,k (t) is the control cost of the kth aggregate at the tth moment, which can be quickly calculated according to the economic aggregate control instruction decomposition method. The main grid purchase electricity cost C grid (t) The calculation method remains unchanged.

9. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 8 is characterized in that: It also includes: In addition to considering the conventional system power flow constraints, the operation constraints of distributed control resources, active power balance constraints and power fluctuation rate constraints of the downstream network are also considered. The constraints of the aggregation index model are as follows: (1) Distributed energy storage operation constraints The distributed energy storage after aggregation has a strong control response capability and is a major control resource for smoothing the fluctuation of new energy. In the actual operation of the aggregate, it is generally necessary to meet the charging and discharging power constraints and capacity constraints: In addition, the capacity constraint should be calculated and analyzed in combination with the dynamic characteristics of the charging and discharging capacity of cluster energy storage: Where: ηin and ηout are the charging and discharging efficiencies of the energy storage unit respectively; assuming that all energy storage units in the system are of the same model, the cluster charging and discharging efficiencies are also the same; (2) Distributed PV operation constraints Distributed photovoltaics generally work near the maximum power point and only abandon light appropriately when the absorption is insufficient, that is, as a downward adjustment resource to participate in the power balance of the distribution network; in this process, distributed photovoltaics should meet the following constraints: Where: is the total amount of distributed photovoltaic regulation in aggregate k at time t; The maximum controllable amount of the lth distributed unit; (3) Power balance constraints (4) Interruptible load operation constraints Considering the aggregate output constraint of the interruptible load aggregate and the daily interruption number constraint, Where: They are the current cumulative interruption times and the maximum daily interruption times of the interruptible load unit respectively; (5) Power fluctuation rate constraints at the downstream network Power fluctuation rate R at lower grid point vol It represents the power change rate of the current moment relative to the previous moment. The power fluctuation rate constraint condition of the downstream network point is determined according to the distribution network structure and resource parameters. The fluctuation rate R vol Limit to the target volatility R B Within: In order to simplify the optimization control model and establish a direct connection between the power fluctuation rate constraint of the downstream grid point and the economic objective function, this paper adds it to the economic objective function in the form of a penalty function. In the optimization process, a larger value is set for the coefficient β that characterizes the intensity of the penalty function, and the economic objective function that considers the power fluctuation smoothing effect is obtained: Since the system power flow is a nonlinear constraint, the particle swarm optimization algorithm PSO is used for rapid solution to meet the needs of rapid smoothing control of distributed photovoltaic power fluctuations.

10. The distributed photovoltaic power fluctuation smoothing control method based on the comparison density peak clustering algorithm according to claim 9, characterized in that: It also includes: The total control amount of each resource aggregate can be obtained by solving the above optimization control model, and it needs to be further decomposed quickly. Compared with the conventional decomposition method of allocating according to capacity ratio, the present invention takes economic efficiency as the priority decision basis and constructs a distributed resource aggregate control instruction fast decomposition model to meet the needs of rapid smoothing of distributed photovoltaic power fluctuations. Its objective function and constraints are as follows: The essence of the optimization model is to sort the adjustable resources in the aggregate according to the economic indicators, establish a piecewise functional relationship between the total control cost of the aggregate and the total control quantity (15), and finally realize the rapid decomposition of the control instructions with the total control quantity of the aggregate as a hard constraint.

Images