Low-voltage distribution network collaborative management and control method based on cloud load library

Abstract

The invention discloses a low-voltage power distribution network collaborative management and control method based on a cloud load library. The method comprises the following steps: acquiring historical data of a low-voltage power distribution network; performing memory processing on the data and establishing a cloud load library; constructing an automatic matching model of the low-voltage distribution network load and the cloud load library model; the distributed photovoltaic system, the energy storage system and the charging pile single equipment are subjected to cluster regulation and control; and constructing a low-voltage distribution network cooperative control model to realize source-load-storage resource optimization control. By mining the historical data of massive multi-element loads, the accessed unknown type of power distribution network load can be identified, the perception capability of the controllable resources of the low-voltage power distribution network is improved, the control capability of the controllable resources of the power distribution network is improved, and on-site consumption of new energy is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of energy coordinated control, and in particular to a low-voltage distribution network coordinated control method based on a cloud load library. Background Art

[0002] Traditional distribution networks use indicators such as overload rate and line loss rate as control benchmarks. For example, the replacement and maintenance of transformers are closely related to the load rate. The load rate refers to the ratio of the transformer's annual maximum load to its rated capacity. If the load rate exceeds a certain threshold, it is considered that the substation is heavily overloaded and the transformer needs to be expanded. In addition, indicators such as voltage offset and three-phase imbalance can also be used as control standards. The advantage of this collaborative control method is that its principle is simple and the steps are short, which is suitable for unified planning and management of grassroots personnel of power companies. However, with the steady progress of the construction of new distribution networks, distributed power sources and electric vehicles have ushered in explosive growth, and various types of equipment are constantly connected to low-voltage distribution networks, which has caused the distribution network to have problems such as increased load volatility, increased bidirectional power flow and peak-to-valley differences, making it difficult for traditional control methods to adapt to the complex and changing new needs of new distribution networks.

[0003] Data show that with the continuous advancement of the construction of a new power system, the load on the distribution network has undergone tremendous changes. The user load has developed in a complex and interactive direction, changing from a traditional "power consumer" to a "power producer and consumer". Although the level of automation and digitalization of the distribution network continues to improve, the perception and control capabilities of the controllable devices of the diversified distribution network are still insufficient. The cloud load library can identify the unknown types of connected distribution network loads by mining the historical data of massive and diversified loads, and improve the perception of the controllable resources of the low-voltage distribution network. With the support of the load library, it is practical to build a collaborative control model for massive and diverse low-voltage distribution networks. Then, the cloud-edge collaborative architecture is used to solve the model and regulate the controllable devices of the distribution network, so as to improve the control capabilities of the controllable resources of the distribution network, promote the local consumption of new energy, reduce carbon emissions, reduce the operating losses of the low-voltage distribution network, and promote the construction of a new power system with new energy as the main body.

[0004] Chinese patent document CN115102953A discloses a "distribution network cloud-edge collaborative management and control system and method". The system includes: autonomous control terminals, aggregated collaborative devices and central cloud connected in a tree network; the central cloud is a primary node, the aggregated collaborative device is a secondary node, and the autonomous control terminal is a terminal node; the central cloud is connected to multiple aggregated collaborative devices, and each aggregated collaborative device is connected to multiple autonomous control terminals. The above technical solution has limited perception and control capabilities of controllable resources in the low-voltage distribution network. Summary of the invention

[0005] The present invention mainly solves the technical problem that the original technical solution has limited perception and control capabilities of controllable resources of low-voltage distribution networks, and provides a collaborative control method for low-voltage distribution networks based on a cloud load library. First, the dimensions of data such as loads, photovoltaic outputs, and electric vehicle charging power of each user in the low-voltage distribution network are reduced, and then the K-means algorithm is used to cluster the reduced-dimensional data; secondly, an automatic matching method for low-voltage distribution network loads and cloud load library models is proposed, and a matching model is constructed using a Softmax classifier; finally, a controllable device response mode of a distribution network collaborative control strategy is proposed, and a low-voltage distribution network collaborative control model supported by a load library is established, and finally a low-voltage distribution network collaborative control method is obtained. By mining the historical data of massive multi-dimensional loads, the unknown types of connected distribution network loads can be identified, the perception capability of controllable resources of the low-voltage distribution network can be improved, the control capability of controllable resources of the distribution network can be improved, and the on-site consumption of new energy can be promoted.

[0006] The above technical problem of the present invention is mainly solved by the following technical solution: The present invention comprises the following steps:

[0007] S1 obtains historical data of low-voltage distribution network;

[0008] S2 processes the data and builds a cloud load library;

[0009] S3 builds an automatic matching model between low-voltage distribution network load and cloud load library model;

[0010] S4 regulates distributed photovoltaic, energy storage systems and charging piles in clusters;

[0011] S5 builds a low-voltage distribution network collaborative control model to achieve optimal control of source, load and storage resources.

[0012] Firstly, the dimensionality of the load, photovoltaic output and electric vehicle charging power of each user in the low-voltage distribution network is reduced, and then the K-means algorithm is used to cluster the reduced data; secondly, an automatic matching method between the low-voltage distribution network load and the cloud load library model is proposed, and the matching model is constructed using the Softmax classifier; finally, the controllable device response mode of the distribution network collaborative control strategy is proposed, and a low-voltage distribution network collaborative control model supported by the load library is established, and finally a low-voltage distribution network collaborative management and control method is obtained.

[0013] Preferably, the step S1 specifically includes obtaining historical data of the low-voltage distribution network through the State Grid data center and the procurement system, the historical data including load data of power users, output data of photovoltaic systems, charging and discharging data of energy storage systems and charging power data of charging piles; the field names of the data mainly include: distribution station area identification, area number, area name, transformer identification, public transformer identification, three-phase power, PT value, CT value, transformer timestamp, power user account number, user active value, user timestamp, photovoltaic installed capacity, photovoltaic real-time power, photovoltaic timestamp, energy storage capacity, energy storage real-time power, energy storage timestamp, charging pile rated power, charging pile real-time power, charging pile timestamp including each user's load, photovoltaic output, energy storage system and electric vehicle charging power.

[0014] Preferably, the step S2 specifically includes: using the seasonal machine-side Studentized deviation test algorithm to detect abnormal values ​​of the measured data for the acquired historical operation data of the low-voltage distribution network, and processing them as missing values, and then filling the missing values ​​based on the mean of the measured data at the same time to complete the data cleaning and correction work; using PCA principal component analysis to reduce the data dimension; using the K-Means algorithm as the clustering method, and using the elbow curve method to determine the number of clusters; calculating the distance between clusters based on the Euclidean distance, and using the SSE error square sum formula to judge the performance of the clustering, and completing the establishment of the cloud-based load library.

[0015] Preferably, the step S3 constructs an automatic matching model between the low-voltage distribution network load and the cloud load library model through a Softmax classifier, specifically including: given an n-dimensional vector, the Softmax function maps it to a probability distribution. The standard Softmax function is as follows:

[0016]

[0017] After the substation features are input into the Softmax classifier, the Softmax classifier outputs the probability that the substation belongs to any load storage substation type. If some of the probability values ​​are close, it means that the input substation belongs to multiple typical substations at the same time, that is, the classification effect is not ideal and there are problems with the recognition result. The credibility index is defined based on the information entropy to judge the quality of the matching effect. The larger the information entropy of the result, the larger the amount of information that is clarified, that is, the greater the uncertainty of the variable. The calculation formula of information entropy is as follows:

[0018] P(X=x i )=p i ,i=1,2,...,n

[0019]

[0020] In the formula, X is a discrete random variable with a finite number of occurrences, X is the sign that a station area belongs to any station area, and p i is the possibility that a certain area belongs to the i-th type of area in the load pool, and n is the number of typical areas;

[0021] The calculation formula of the credibility index is as follows:

[0022]

[0023] Preferably, the step S4 specifically includes:

[0024] First, the cloud master station sends the photovoltaic alignment obtained by the collaborative control model to the fusion terminal. The specific form of the photovoltaic alignment is [1×96] data, that is, the photovoltaic expected power value at 96 time points on the second day of the substation:

[0025] P pv =[p pv,1 ,p pv,2 ,p pv,3 …p pv,96 ]

[0026] Then, the fusion terminal decomposes the received photovoltaic line Ppv into multiple single-device photovoltaic lines based on the specific situation of the distributed photovoltaic in this area, so as to realize the fixed-point control of the individual photovoltaic equipment in the entire cluster. The control power of the photovoltaic single device is calculated by the following formula:

[0027]

[0028] Where P pv_j,t is the control power of the jth photovoltaic device in a certain area at time t, P pv,t is the quasi-linear power of the photovoltaic cluster in the area at time t, is the rated capacity of the jth photovoltaic device in the substation, and m is the total number of distributed photovoltaic single devices in the substation. Similarly, the control method of single devices in the energy storage cluster and charging pile cluster refers to the photovoltaic cluster.

[0029] As a preferred embodiment, the step S5 specifically includes, considering the large degree of load fluctuation of distribution transformers and the problem of distributed new energy consumption that are common in low-voltage distribution substations, taking the operation efficiency of the distribution network and the stable operation of the transformer as the goal, converting the source-grid-load-storage configuration of each type of substation into constraints, taking the distributed photovoltaic, energy storage system and controllable charging piles in the substation as the optimization objects, building a collaborative control model for the optimized operation of the distribution network, and realizing the optimized operation of the substation source-load-storage resources under the background of cloud-edge collaboration. The specific modeling is as follows:

[0030] The objective function of the low-voltage distribution network collaborative control model based on the cloud load library is:

[0031]

[0032] Among them, f 1 Indicates that the load fluctuation of the distribution transformer is minimal, f 2 It means that the distributed photovoltaic power consumption in the area is the largest and the load fluctuation of the distribution transformer is the smallest. In the formula, P tran，t is the load of the transformer in period t; is the average daily load of the transformer; P pv，t It is divided into the output of distributed photovoltaic in time period t; T is the scheduling period, which is consistent with the sampling frequency of the measurement system and is set to 96, and Δt=15min is the time interval.

[0033] As a preferred embodiment, the constraint conditions include power balance constraint of distribution station area, photovoltaic output constraint, energy storage state constraint and vehicle charging pile constraint, etc.;

[0034] Assuming that the grid loss in the substation area does not change much before and after the control strategy, the power balance constraint can ignore the grid loss. The final substation distribution transformer load is the sum of the substation user load, substation photovoltaic processing, substation energy storage load and charging pile load. The power balance constraint of the distribution substation area is:

[0035] P tran,t =P con,t +P ess,t +P cp,t -P pv,t

[0036] Where: P tran，t , P con，t , P ess，t , P cp，t and P pv，t They are the loads of transformer, user, energy storage cluster, charging pile cluster and distributed photovoltaic cluster in period t, in KW. ess，t When it is positive, it indicates that the energy storage cluster is in a charging state as a whole, and when it is negative, it indicates that it is in a discharging state.

[0037] As a preferred method, according to the matched substation type, the photovoltaic access ratio is known, and combined with the rated capacity of the substation distribution transformer, the photovoltaic installed capacity can be calculated, thereby obtaining the upper and lower limit constraints of photovoltaic output. The upper and lower limit constraints of photovoltaic output are:

[0038] 0≤P pv,t <P tran,e ·C pv

[0039] Where: P pv,t is the power of the photovoltaic cluster in the area at time t, P tran，e is the rated capacity of the distribution transformer, in KW; C pvis the access ratio of photovoltaic systems in the substation area;

[0040] The photovoltaic output curve is affected by weather factors and naturally conforms to the Beta distribution. When the cooperative control model outputs the photovoltaic line, it is necessary to add output distribution constraints. The photovoltaic curve of the load storage area is used as the basis to constrain the photovoltaic shape. Considering that the output result of the cooperative control model is the photovoltaic power line of the next day, that is, the time interval is small and the meteorological factors do not change much, the photovoltaic output distribution constraints are as follows:

[0041]

[0042] Where: P′ pv_min,t is the photovoltaic baseline P′ pv,t The minimum value within one hour before and after time period t; P′ pv_max,t is the photovoltaic baseline P′ pv,t The maximum value within one hour before and after time period t.

[0043] As a preferred method, according to the matched substation type, the access ratio of its energy storage system is known, and combined with the rated capacity of the substation distribution transformer, the capacity of the substation energy storage system can be calculated, thereby obtaining the upper and lower limit constraints of the energy storage charging and discharging power. The energy storage charging and discharging power constraints are:

[0044] -P tran,e ·C ess_p <=P ess,t <P tran,e ·C ess_p

[0045] Where P ess,t is the power of the energy storage system at time t, P tran,e is the rated capacity of the distribution transformer, C ess_p is the access ratio of energy storage systems in the substation area;

[0046] The capacity constraint formula of the energy storage system is as follows:

[0047]

[0048] Where: SOC t is the state of charge of the energy storage cluster at time t; η ess is the energy storage charging and discharging efficiency, which is taken as 0.95 here; E ess is the configuration capacity of the energy storage cluster; SOC min , SOC max They are the upper and lower limits of the state of charge of the energy storage system, which are set to 5% and 90% respectively.

[0049] As a preferred method, time constraints are introduced into the optimization scheduling model, and u t and d tThey represent the power-on and power-off actions of the charging pile at time t, respectively. t represents the status of the charging pile cluster at time t, 1 for charging and 0 for idle, then u t and d t It is obtained by the following constraints:

[0050]

[0051] Set the minimum charging time T on , initial charging time X on，0 , minimum idle time T off , initial idle time X off，0 , the initial state of the charging pile cluster I 0 , then the minimum charging time constraint of the charging pile cluster is as follows:

[0052]

[0053] The minimum idle time constraints are as follows:

[0054]

[0055] The beneficial effects of the present invention are as follows: by mining the historical data of massive multi-dimensional loads, the unknown types of connected distribution network loads can be identified, the perception ability of the controllable resources of the low-voltage distribution network can be improved, the management and control capabilities of the controllable resources of the distribution network can be improved, and the on-site consumption of new energy can be promoted. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flow chart of the present invention.

[0057] Figure 2 It is an elbow curve diagram for determining the number of clusters of samples in a station area according to the present invention.

[0058] Figure 3 It is a schematic diagram of a photovoltaic cluster coordinated control response of the present invention.

[0059] Figure 4 It is a schematic diagram of a photovoltaic cluster coordinated control response of the present invention.

[0060] Figure 5 It is a schematic diagram of comparison before and after control of area No. 1 in an embodiment of the present invention.

[0061] Figure 6 It is a schematic diagram of comparison before and after control of area No. 2 in an embodiment of the present invention.

[0062] Figure 7 It is a schematic diagram of comparison before and after the control of the No. 3 station area in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0064] Embodiment: A low-voltage distribution network collaborative control method based on a cloud load library in this embodiment, such as Figure 1 As shown, the following steps are included:

[0065] S1. Obtain historical data of low-voltage distribution networks through the State Grid data center and the user procurement system, including the load of each user, photovoltaic output, energy storage system, and electric vehicle charging power. After data dimension reduction, cluster the data to reduce the massive low-voltage distribution network to a small number of typical low-voltage distribution networks, and complete the establishment of the cloud load library.

[0066] S2. Build an automatic matching model between low-voltage distribution network load and cloud load library model through Softmax classifier, define credibility index and evaluate matching recognition effect;

[0067] S3. All distributed photovoltaic, energy storage systems and charging piles in a low-voltage distribution network are aggregated into three clusters through the State Grid intelligent fusion terminal, and controllable devices are regulated in clusters;

[0068] Establish a low-voltage distribution network collaborative control model supported by a load library. Relying on the cloud load library, build a low-voltage distribution network collaborative control model that can cover all the load models in the library. Consider the large degree of load fluctuation of distribution transformers and the problem of distributed new energy consumption in low-voltage distribution stations. With the goal of distribution network operation efficiency and transformer operation stability, transform the source, grid, load and storage configuration of each type of station into constraints. Take the distributed photovoltaic, energy storage system and controllable charging piles in the station as the optimization objects, build a collaborative control model for the optimized operation of the distribution network, realize the optimized operation of the source, load and storage resources in the station under the background of cloud-edge collaboration, and finally obtain a collaborative control method for low-voltage distribution networks. ;

[0069] The principle and development of step S1 are as follows:

[0070] First, the acquired historical data is cleaned and corrected, and the seasonal machine-side studentized deviation test (SH-ESD) algorithm is used for outlier detection, and the missing value filling method of the mean of the measured data at the same time (MVSD) is used. Specifically, the MVSD method is first used to fill the missing values, and then the SH-ESD algorithm is used for outlier detection, the outliers are redefined as missing values, and then filled by the MVSD method to complete the data cleaning and correction work;

[0071] PCA is used to reduce the dimension of user load, photovoltaic output, and electric vehicle charging power. The basic idea of ​​PCA is to reduce the data dimension by maximizing the variance of the data, that is, to find a new set of dimensions to maximize its variance, thereby converting the original data into low-dimensional data. In PCA, we first convert the original data set into a covariance matrix. The calculation formula of the covariance matrix is:

[0072]

[0073] Where ω is the unit vector in the projection direction; x i is the centralized data, which is an n-dimensional column vector; n is the data dimension.

[0074] Then calculate the eigenvalues ​​and eigenvectors of the covariance matrix, and arrange the eigenvectors from large to small according to the corresponding eigenvalues, select the first n eigenvectors to form a new feature space, and then project the original data into the new feature space to obtain low-dimensional data. The PCA method mainly decomposes the covariance matrix of the variables to obtain the principal components (i.e., eigenvectors) and corresponding weight values ​​(i.e., eigenvalues) of the data, which can reduce the data from n dimensions to d dimensions. The calculation formula is:

[0075]

[0076] In the formula, x i is the i-th eigenvector, x' i is the data after dimension reduction;

[0077] The core idea of ​​the K-Means algorithm is: first randomly select k initial cluster centers C from the data set i (i≤1≤k), calculate the remaining data objects and the cluster center C i Euclidean distance, find the cluster center C closest to the target data object i , and assign the data objects to the cluster centers C i In the corresponding cluster, the Euclidean distance formula is:

[0078]

[0079] In the formula, X is the data object; C i is the center of the i-th cluster; m is the dimension of the data object; (X j -C ij ) is X and C i The difference of the j-th attribute;

[0080] The SSE formula is used to help us judge the performance of clustering. In fact, it is the sum of the square values ​​of the distances from each point to the centroid of its cluster. The smaller the SSE value, the closer the data points are to their centroids, and the better the clustering effect. The SSE formula is:

[0081]

[0082] In the formula, the size of SSE indicates the quality of the clustering result; k is the number of clusters.

[0083] In order to determine the number of cluster centers, the elbow curve method commonly used in cluster analysis technology is used, that is, by setting several cluster numbers, the sum of the distances between clusters is calculated, such as Figure 2 As shown in the figure, the clustering elbow curve of the samples in the station area. By observing the elbow curve, the number of clusters at the bend is the more appropriate number of clusters.

[0084] The principle and development of step S2 are as follows:

[0085] Given an n-dimensional vector, the Softmax function maps it to a probability distribution. The standard Softmax function is defined by the following formula:

[0086]

[0087] In the formula, we can see that the Softmax function solves the problem of mapping from an original n-dimensional vector X to an n-dimensional probability distribution. Assuming that the number of categories is n, the model hopes to predict the probability distribution of a sample in n categories.

[0088] After the substation features are input into the Softmax classifier, the Softmax classifier will output the probability that the substation belongs to any load storage substation type. If some of the probability values ​​are close, it means that the input substation belongs to multiple typical substations at the same time, that is, the classification effect is not ideal and there are problems with the recognition result. In view of this situation, a credibility index is defined based on information entropy to judge the quality of the matching effect. Information entropy is a basic concept of information theory. The larger the information entropy of the result, the larger the amount of information that is clarified, that is, the greater the uncertainty of the variable. The calculation formula of information entropy is as follows:

[0089] P(X=x i )=p i , i=1,2,...,n,

[0090] Where X is a discrete random variable with a finite number (70), and its probability distribution is Equation 1. In this example, X is the sign that a station belongs to any station, where p iis the possibility that a certain area belongs to the i-th type of area in the load pool, and n is the number of typical areas, which is 70 here.

[0091] The calculation formula of the credibility index is as follows:

[0092]

[0093] The principle and development of step S3 are as follows:

[0094] Since the control capacity of distributed photovoltaic, energy storage system and charging pile in low-voltage distribution network is small, if the control scheme is solved directly for a single device, it will consume a lot of computing resources and increase communication pressure. Therefore, it is often necessary to aggregate the three individual devices into three clusters through the fusion terminal, and control them in clusters. Taking the photovoltaic cluster as an example, the cluster response diagram of the collaborative control scheme of this example is as follows: Figure 3 As shown;

[0095] First, the cloud master station sends the photovoltaic alignment obtained by the collaborative control model to the fusion terminal. The specific form of the photovoltaic alignment is [1×96] data, that is, the photovoltaic expected power value at 96 time points on the second day of the substation:

[0096] P pv =[p pv,1 ,p pv,2 ,p pv,3 …p pv,96 ]

[0097] Then, the fusion terminal decomposes the received photovoltaic line Ppv into multiple single-device photovoltaic lines based on the specific situation of the distributed photovoltaic in this area, so as to realize the fixed-point control of the individual photovoltaic equipment in the entire cluster. The control power of the photovoltaic single device is calculated by the following formula:

[0098]

[0099] Where P pv_j,t is the control power of the jth photovoltaic device in a certain area at time t, P pv,t is the quasi-linear power of the photovoltaic cluster in the area at time t, is the rated capacity of the jth photovoltaic device in the substation, and m is the total number of distributed photovoltaic single devices in the substation. Similarly, the control method of single devices in the energy storage cluster and charging pile cluster refers to the photovoltaic cluster.

[0100] The principle and development of step S4 are as follows:

[0101] Relying on the cloud-based load library, a distribution network source-load-storage collaborative control model that can cover all load models in the library is constructed. Considering the large degree of load fluctuation of distribution transformers and the problem of distributed new energy consumption common in low-voltage distribution stations, the source-grid-load-storage configuration of each type of station is transformed into constraints with the goal of distribution network operation efficiency and transformer operation stability. The distributed photovoltaic, energy storage system and controllable charging piles in the station are taken as optimization objects, and a collaborative control model for the optimized operation of the distribution network is constructed to achieve the optimized operation of the source-load-storage resources in the station under the background of cloud-edge collaboration. The specific modeling is as follows:

[0102] The objective function of the low-voltage distribution network collaborative control model based on the cloud load library is:

[0103]

[0104] Among them, f 1 Indicates that the load fluctuation of the distribution transformer is minimal, f 2 It means that the distributed photovoltaic power consumption in the area is the largest and the load fluctuation of the distribution transformer is the smallest. tran，t is the load of the transformer in period t; is the average daily load of the transformer; P pv，t It is divided into the output of distributed photovoltaic in time period t; T is the scheduling period, which is consistent with the sampling frequency of the measurement system and is set to 96, and Δt=15min is the time interval.

[0105] Constraints include power balance constraints in distribution stations, photovoltaic output constraints, energy storage status constraints, and vehicle charging pile constraints.

[0106] Assuming that the grid loss in the substation area does not change much before and after the control strategy, the power balance constraint can ignore the grid loss. The final substation distribution transformer load is the sum of the substation user load, substation photovoltaic processing, substation energy storage load and charging pile load. The power balance constraint of the distribution substation area is:

[0107] P tran,t =P con,t +P ess,t +P cp,t -P pv,t

[0108] Where: P tran，t , P con，t , P ess，t , P cp，t and P pv，t are the loads of transformer, user, energy storage cluster, charging pile cluster and distributed photovoltaic cluster in period t, in KW. ess，t When it is positive, it indicates that the energy storage cluster is in a charging state as a whole, and when it is negative, it indicates that it is in a discharging state.

[0109] Furthermore, according to the matched substation type, the proportion of photovoltaic access is known, and combined with the rated capacity of the substation distribution transformer, the photovoltaic installed capacity can be calculated, thereby obtaining the upper and lower limits of photovoltaic output. The upper and lower limits of photovoltaic output are:

[0110] 0≤P pv,t <P tran,e ·C pv

[0111] Where: P pv,t is the power of the photovoltaic cluster in the area at time t, P tran，e is the rated capacity of the distribution transformer, in KW; C pv It is the access ratio of photovoltaic systems in the substation area.

[0112] Since the photovoltaic output curve is affected by weather factors, it naturally conforms to the Beta distribution, rather than an adjustable shape. Therefore, when the collaborative control model outputs the photovoltaic line, it is necessary to add output distribution constraints. This project uses the photovoltaic curve of the load storage area as the basis to constrain the photovoltaic shape. Considering that the output result of the collaborative control model is the photovoltaic power line of the next day, that is, the time interval is small and the meteorological factors do not change much, the photovoltaic output distribution constraints are as follows:

[0113]

[0114] Where: P′ pv_min,t is the photovoltaic baseline P′ pv,t The minimum value within one hour before and after time period t; P′ pv_max,t is the photovoltaic baseline P′ pv,t The maximum value within one hour before and after time period t.

[0115] Furthermore, according to the matched substation type, the access ratio of its energy storage system is known. Combined with the rated capacity of the substation distribution transformer, the capacity of the substation energy storage system can be calculated, thereby obtaining the upper and lower limit constraints of the energy storage charging and discharging power. The energy storage charging and discharging power constraints are:

[0116] -P tran,e ·C ess_p <=P ess,t <P tran,e ·C ess_p

[0117] Where P ess,t is the power of the energy storage system at time t, P tran,e is the rated capacity of the distribution transformer, C ess_p It is the access ratio of energy storage system in the substation area.

[0118] Furthermore, the capacity constraint formula of the energy storage system is as follows:

[0119]

[0120] Where: SOC t is the state of charge of the energy storage cluster at time t; η ess is the energy storage charging and discharging efficiency, which is taken as 0.95 here; E ess is the configuration capacity of the energy storage cluster; SOC min , SOC max They are the upper and lower limits of the state of charge of the energy storage system, which are set to 5% and 90% respectively.

[0121] Furthermore, according to the matched substation type, the access ratio of its electric vehicle charging piles is known. Combined with the rated capacity of the substation distribution transformer, the upper and lower limit constraints of the charging power of the electric vehicle charging piles can be calculated. The charging power constraint of the charging piles is:

[0122] 0≤P cp,t ≤P tran,e ·C cp_p

[0123] Where P cp,t is the power of the electric vehicle charging pile in the area at time t, P tran,e is the rated capacity of the distribution transformer, C cp_p It is the access ratio of energy storage system in the substation area.

[0124] Furthermore, considering that the load characteristics of charging piles are closely related to user travel habits, and the charging time of electric vehicles is continuous, we introduce time constraints into the optimization scheduling model by referring to the minimum operating time and minimum downtime constraints in the unit commitment problem. Let u t and d t They represent the power-on and power-off actions of the charging pile at time t, respectively. t represents the status of the charging pile cluster at time t, 1 for charging and 0 for idle, then u t and d t It is obtained by the following constraints:

[0125]

[0126] Set the minimum charging time T on , initial charging time X on，0 , minimum idle time T off , initial idle time X off，0 , the initial state of the charging pile cluster I 0 , then the minimum charging time constraint of the charging pile cluster is as follows:

[0127]

[0128] The minimum idle time constraints are as follows:

[0129]

[0130] The four input data for model solution are as follows:

[0131] Rated capacity P of low voltage distribution transformer tran,e ,This data can be directly collected by the distribution and transformation fusion terminal in the substation area, and then uploaded to the regional cloud master station together with the load characteristic data of the substation area for use in solving the collaborative control model;

[0132] The previous day's area user load data P con , this data can be directly uploaded by the substation area with a dimension of 96 for the previous day's load data, but considering that the substation distribution transformer fusion terminal will submit feature data with a dimension of 19, and using this feature data through inverse processing, an inverse processing curve close to the user's actual load curve can be obtained, such as Figure 4 As shown, the data can be obtained by processing the characteristic data uploaded by the station area;

[0133] Level access of solar storage and charging, C pv , C ess , C cp , these three data are obtained through the matching results of the load library;

[0134] Typical photovoltaic curve P′ in the area pv ,According to the matching result of the terminal load of the substation, the photovoltaic curve corresponding to the substation type in the cloud substation load library is used as the typical photovoltaic curve of the newly connected substation.

[0135] Model output: This model outputs three curves as control instructions, namely the photovoltaic operation criterion, the energy storage operation criterion and the charging pile operation criterion, and the data dimension is [3×96].

[0136] Simulation verification:

[0137] Relying on the cloud-based substation load library, a distribution network source-load-storage collaborative control model is constructed that can cover all substation load models in the library; considering the large degree of distribution transformer load fluctuations and distributed new energy consumption problems common in low-voltage distribution substations, with the distribution network operation efficiency and transformer operation stability as the goal, the source-grid-load-storage configuration of each type of substation is transformed into constraints, and the distributed photovoltaic, energy storage systems and controllable charging piles in the substation are taken as optimization objects, a collaborative control model for the optimized operation of the distribution network is constructed, and the optimized operation of the substation source-load-storage resources under the background of cloud-edge collaboration is realized.

[0138] The source-load-storage coordinated control model constructed here is essentially a nonlinear optimization model with low difficulty in solving. The Gurobi solver can quickly and accurately give the results. In order to show the source-load-storage coordinated control effect of the distribution station area under different source-load-storage configurations, three residential stations shown in Table 1 are selected for optimized operation control and the results are analyzed. In order to intuitively reflect the difference in effect before and after optimization, the load rate before and after optimization and the peak-to-valley difference rate of load fluctuation are calculated respectively.

[0139]

[0140] Table 1 Introduction to the selected areas

[0141] Among them, Area 1 includes 78 low-voltage users, with a total photovoltaic installed capacity of 80KW, energy storage of 45KW, 6 charging piles, an output power of 48KW, and a data collection frequency of 15min / time, with a total of 96 points per day. Figure 5 , 6 As shown in Figure 7, before optimization, the peak-to-valley difference of the area was 162.21KW, and the peak-to-valley difference rate was 98.03%. There was an imbalance in the power load. Specifically, the first peak was from 0:00 to 4:00 in the morning. Combined with the analysis of the energy storage curve and the charging pile curve, it was found that the area was in the peak period of the area energy storage system and the charging pile power during this period. In addition, the photovoltaic output was 0 at this time, which resulted in a high load on the distribution transformer during this period. The second peak was from 9:00 to 11:00 in the morning. The main reason for the peak load during this period was the influence of the power consumption characteristics of the users in the area. This may be the time when most residents prepare for breakfast and lunch. Moreover, since the photovoltaic output is small at this time, even if the energy storage system discharges for balance, it still leads to a peak load on the distribution transformer. The third period is from 5:00 to 9:00 in the afternoon. This period is generally the time when residents get off work to prepare dinner and have entertainment after dinner. The users themselves have a high power consumption, and the owners of trams need to use car charging piles to charge, which leads to a peak load during this period.

[0142] Solve the coordinated control model of the distribution network in this area and obtain the comparison lines of photovoltaic, energy storage and charging piles as follows: Figure 5 , 6As shown in Figure 7, the overall trend of the distribution transformer after model optimization tends to be stable, and the peak-to-valley difference is significantly reduced; observing the charging piles and energy storage curves as the optimization objects, it is found that the charging time of the charging piles is mainly distributed in two periods: 0:00-8:00 and 21:00-24:00. There is a little charging time during the noon period, which is in line with the work and rest rules of residents to a certain extent; the optimized energy storage system has more flexible charging and discharging conversion than before, which can well balance the load fluctuation of the distribution transformer caused by the load characteristics of the residents themselves. After optimization, the load rate of the substation area increased, the volatility decreased greatly, the peak-to-valley difference decreased, and the peak-to-valley difference rate decreased from 98.03% to 16.01%. That is, when the substation is equipped with a storage system with sufficient capacity and a certain number of controllable charging pile loads, the distribution network collaborative control model proposed in this example can achieve better results and effectively reduce the load fluctuation of the distribution transformer in the substation area.

[0143] By comparing the situations before and after the optimization of the above three substations, we can get the table shown in Table 2, which shows the differences in the coordinated control effects caused by the differences in the configuration of the source, load and storage resources in the substation. Among them, the source, load and storage configuration is the ratio of the incoming capacity to the distribution and transformation capacity of the substation, that is, the access ratio; in the column of coordinated control effect, the initial value is the actual value of the substation before the coordinated control, and the optimized value is the difference between a certain value before optimization and the value after optimization. For the initial value and optimized value of the control effect column, the best and worst values ​​in the three substations are selected for special marking.

[0144]

[0145]

[0146] Table 2 Coordinated control effects under different source-load-storage configurations

[0147] Based on the above results, it can be seen that by mining the historical data of massive multi-dimensional loads, it is possible to identify the unknown types of connected distribution network loads, improve the perception ability of controllable resources of the low-voltage distribution network, improve the management and control capabilities of controllable resources of the distribution network, and promote the local consumption of new energy.

Claims

1. A low-voltage distribution network collaborative control method based on a cloud load library, characterized in that: The following steps are involved: S1 obtains historical data of low-voltage distribution network; S2 processes the data and builds a cloud load library; S3 builds an automatic matching model between low-voltage distribution network load and cloud load library model; S4 regulates distributed photovoltaic, energy storage systems and charging piles in clusters; S5 builds a low-voltage distribution network collaborative control model to achieve optimal control of source, load and storage resources.

2. A low-voltage distribution network collaborative control method based on a cloud load library according to claim 1, characterized in that: The step S1 specifically includes obtaining historical data of the low-voltage distribution network through the State Grid data center and the user procurement system. The historical data includes load data of power users, output data of photovoltaic systems, charging and discharging data of energy storage systems, and charging power data of charging piles. The field names of the data mainly include: distribution station area identification, area number, area name, transformer identification, public transformer identification, three-phase power, PT value, CT value, transformer timestamp, power user account number, user active value, user timestamp, photovoltaic installed capacity, photovoltaic real-time power, photovoltaic timestamp, energy storage capacity, energy storage real-time power, energy storage timestamp, charging pile rated power, charging pile real-time power, charging pile timestamp including each user's load, photovoltaic output, energy storage system and electric vehicle charging power.

3. A low-voltage distribution network collaborative control method based on a cloud load library according to claim 1 or 2, characterized in that: The step S2 specifically includes, for the acquired historical operation data of the low-voltage distribution network, using the seasonal machine-side student deviation test algorithm to detect abnormal values ​​of the measured data, and processing them as missing values, and then filling the missing values ​​based on the mean of the measured data at the same time, to complete the data cleaning and correction work; PCA principal component analysis was used to reduce data dimension; K-Means algorithm was used as clustering method, and the elbow curve method was used to determine the number of clusters; The distance between clusters is calculated based on the Euclidean distance, and the SSE error square sum formula is used to judge the performance of clustering to complete the establishment of the cloud load library.

4. A method for collaborative control of low-voltage distribution network based on cloud load storage according to claim 1, characterized in that: The step S3 constructs an automatic matching model between the low-voltage distribution network load and the cloud load library model through a Softmax classifier, specifically including: given an n-dimensional vector, the Softmax function maps it to a probability distribution. The standard Softmax function is as follows: After the substation features are input into the Softmax classifier, the Softmax classifier outputs the probability that the substation belongs to any load storage substation type. If some of the probability values ​​are close, it means that the input substation belongs to multiple typical substations at the same time, that is, the classification effect is not ideal and there are problems with the recognition result. The credibility index is defined based on the information entropy to judge the quality of the matching effect. The larger the information entropy of the result, the larger the amount of information that is clarified, that is, the greater the uncertainty of the variable. The calculation formula of information entropy is as follows: P(X=x i )=p i ,i=1,2,...,n In the formula, X is a discrete random variable with a finite number of occurrences, X is the sign that a station area belongs to any station area, and p i is the possibility that a certain area belongs to the i-th type of area in the load pool, and n is the number of typical areas; The calculation formula of the credibility index is as follows:

5. A low-voltage distribution network collaborative control method based on a cloud load library according to claim 1, characterized in that: The step S4 specifically includes: First, the cloud master station sends the photovoltaic alignment obtained by the collaborative control model to the fusion terminal. The specific form of the photovoltaic alignment is [1×96] data, that is, the photovoltaic expected power value at 96 time points on the second day of the substation: P pv =[p pv,1 ,p pv,2 ,p pv,3 …p pv,96 ] Then, the fusion terminal decomposes the received photovoltaic line Ppv into multiple single-device photovoltaic lines based on the specific situation of the distributed photovoltaic in this area, so as to realize the fixed-point control of the individual photovoltaic equipment in the entire cluster. The control power of the photovoltaic single device is calculated by the following formula: Where P pv_j,t is the control power of the jth photovoltaic device in a certain area at time t, P pv,t is the quasi-linear power of the photovoltaic cluster in the area at time t, is the rated capacity of the jth photovoltaic device in the substation, and m is the total number of distributed photovoltaic single devices in the substation. Similarly, the control method of single devices in the energy storage cluster and charging pile cluster refers to the photovoltaic cluster.

6. A method for collaborative control of low-voltage distribution network based on cloud load storage according to claim 1, characterized in that: The step S5 specifically includes, taking into account the large degree of load fluctuation of distribution transformers and the problem of distributed new energy consumption that are common in low-voltage distribution substations, taking the operation efficiency of the distribution network and the stable operation of transformers as the goal, converting the source-grid-load-storage configuration of each type of substation into constraints, taking the distributed photovoltaic, energy storage system and controllable charging piles in the substation as the optimization objects, building a collaborative control model for the optimized operation of the distribution network, and realizing the optimized operation of the source-load-storage resources in the substation under the background of cloud-edge collaboration. The specific modeling is as follows: The objective function of the low-voltage distribution network collaborative control model based on the cloud load library is: Among them, f1 indicates that the load fluctuation of the distribution transformer is the smallest, and f2 indicates that the distribution transformer load fluctuation is the smallest when the distributed photovoltaic consumption in the area is the largest. tran，t is the load of the transformer in period t; is the average daily load of the transformer; P pv，t It is divided into the output of distributed photovoltaic in time period t; T is the scheduling period, which is consistent with the sampling frequency of the measurement system and is set to 96, and Δt=15min is the time interval.

7. A method for collaborative control of low-voltage distribution network based on cloud load storage according to claim 6, characterized in that: Constraints include power balance constraints in distribution stations, photovoltaic output constraints, energy storage status constraints, and vehicle charging pile constraints. Assuming that the grid loss in the substation area does not change much before and after the control strategy, the power balance constraint can ignore the grid loss. The final substation distribution transformer load is the sum of the substation user load, substation photovoltaic processing, substation energy storage load and charging pile load. The power balance constraint of the distribution substation area is: P tran,t =P con,t +P ess,t +P cp,t -P pv,t Where: P tran，t , P con，t , P ess，t , P cp，t and P pv，t They are the loads of transformer, user, energy storage cluster, charging pile cluster and distributed photovoltaic cluster in period t, in KW. ess，t When it is positive, it indicates that the energy storage cluster is in a charging state as a whole, and when it is negative, it indicates that it is in a discharging state.

8. A method for collaborative control of low-voltage distribution network based on cloud load storage according to claim 7, characterized in that: According to the matched substation type, the proportion of photovoltaic access is known. Combined with the rated capacity of the substation distribution transformer, the photovoltaic installed capacity can be calculated, thereby obtaining the upper and lower limits of photovoltaic output. The upper and lower limits of photovoltaic output are: 0≤P pv,t <P tran,e ·C pv Where: P pv,t is the power of the photovoltaic cluster in the area at time t, P tran，e is the rated capacity of the distribution transformer, in KW; C pv is the access ratio of photovoltaic systems in the substation area; The photovoltaic output curve is affected by weather factors and naturally conforms to the Beta distribution. When the cooperative control model outputs the photovoltaic line, it is necessary to add output distribution constraints. The photovoltaic curve of the load storage area is used as the basis to constrain the photovoltaic shape. Considering that the output result of the cooperative control model is the photovoltaic power line of the next day, that is, the time interval is small and the meteorological factors do not change much, the photovoltaic output distribution constraints are as follows: Where: P′ pv_min,t is the photovoltaic baseline P′ pv,t Minimum value within one hour before and after period t; P′ pv_max,t is the photovoltaic baseline P′ pv,t The maximum value within one hour before and after time period t.

9. A method for collaborative control of low-voltage distribution network based on cloud load storage according to claim 7, characterized in that: According to the matched substation type, the access ratio of its energy storage system is known. Combined with the rated capacity of the substation distribution transformer, the capacity of the substation energy storage system can be calculated, thereby obtaining the upper and lower limit constraints of the energy storage charging and discharging power. The energy storage charging and discharging power constraints are: -P tran,e ·C ess_p <=P ess,t <P tran,e ·C ess_p Where P ess,t is the power of the energy storage system in the substation at time t, P tran,e is the rated capacity of the distribution transformer, C ess_p is the access ratio of energy storage systems in the substation area; The capacity constraint formula of the energy storage system is as follows: Where: SOC t is the state of charge of the energy storage cluster at time t; η ess is the energy storage charging and discharging efficiency, which is taken as 0.95 here; E ess is the configuration capacity of the energy storage cluster; SOC min , SOC max They are the upper and lower limits of the state of charge of the energy storage system, which are set to 5% and 90% respectively.

10. A method for collaborative control of low-voltage distribution network based on cloud load storage according to claim 7, characterized in that: Introducing time constraints into the optimization scheduling model, let u t and d t They represent the power-on and power-off actions of the charging pile at time t, respectively. t represents the status of the charging pile cluster at time t, 1 for charging and 0 for idle, then u t and d t It is obtained by the following constraints: Set the minimum charging time T on , initial charging time X on，0 , minimum idle time T off , initial idle time X off，0 , the initial state of the charging pile cluster is I0, then the minimum charging time constraint of the charging pile cluster is as follows: The minimum idle time constraints are as follows:

Images