Dynamic clustering multi-element load layering and grouping optimization regulation and control method and device
Through dynamic clustering and clustering control models, combined with the adjustable capacity and geographical location of multiple loads, the solution speed and accuracy problems during massive access to multiple loads are solved, and the efficient utilization of multiple load resources and grid economic optimization are achieved.
Patent Information
- Application Number
- CN202410919965.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-01
AI Technical Summary
The existing technology causes "dimensional disaster" to increase in variables and constraints during massive access to multiple loads, which affects the resolution speed and accuracy of optimization and regulation. The traditional clustering method ignores the dynamic changes in adjustable capacity, resulting in resource utilization. Inefficient.
Using a dynamic clustering method based on K-means, combining the adjustable capacity and geographical location of multi-loads, dynamic aggregate load aggregation model, and through the group regulation optimization model and the in-group power distribution control model, recursive clustering and regulation are established. Iterative relationships ensure the accuracy of regulation and efficient utilization of resources.
It improves the speed and accuracy of the regulation of massive multi-loads, reduces electricity costs, improves the economic benefits of power grid operators, and realizes the efficient utilization of multi-load resources.
Smart Images

Figure CN118889441A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of multiple loads and power systems, and in particular to a method and device for dynamically clustering multiple load stratification and grouping optimization control. Background Art
[0002] my country is building a new power system with new energy as the main body. It is estimated that by 2050, the total power generation of wind and solar power in my country will account for more than 60%. However, the high proportion of new energy access will lead to a series of problems such as difficulty in absorbing power during peak hours, difficulty in ensuring supply during valley hours, and difficulty in regulating during fluctuations. Therefore, it is urgent to tap more flexible load resources to participate in power grid regulation to ensure the real-time balance of power supply and demand. In recent years, with the development of industry and agriculture, the scale of various loads has expanded year by year. Taking the county distribution network as an example, there are many types of diversified characteristic loads, strong seasonal fluctuations, and relatively dispersed distribution, and their electricity consumption is highly flexible. At the same time, its diversified loads usually configure certain energy storage units as backup power sources to ensure the uninterrupted operation of loads when the power grid is out of power. However, with the improvement of the reliability of my country's power grid, the backup energy storage of diversified loads is in a floating charge state most of the time, resulting in idle resources with considerable adjustable potential. Therefore, how to deeply explore and integrate the flexibility resources of the diversified loads and their backup energy storage in the county distribution network (for the sake of ease of description, the term "diverse loads" will be used below) to reduce the electricity costs of the diversified loads while improving the absorption capacity of distributed power sources has become the focus of academic research.
[0003] At present, most studies focus on centralized control of a single load. Although this can meet the operating constraints of each multi-load, it will inevitably introduce large-scale variables and constraints when a large number of multi-loads are connected, leading to the "curse of dimensionality" problem, which seriously affects the speed of solving the optimization control problem. To solve this problem, multiple multi-loads with similar characteristics can be aggregated to form an equivalent aggregate to characterize the operating characteristics of the multi-load group, thereby reducing the scale of variables in the power grid optimization control process and improving the solution speed. At present, the academic community has made some progress in the research on the optimization and control of distributed resource aggregates.
[0004] However, most of the current research on traditional clustering only conducts clustering based on fixed features such as energy consumption coefficient and load curve deviation. Although it improves the speed of solving the control problem, it ignores the changing characteristics of the controllable capacity of distributed resources during the implementation of the control scheme, resulting in uneven individual controllable capacity within the cluster and difficulty in distinguishing the aggregated controllable capacity between clusters, which affects the accurate formulation of charging and discharging decisions of each aggregate, and even causes low-capacity resources to discharge and high-capacity resources to charge, making it difficult to efficiently utilize the control potential. Therefore, it is urgent to design a dynamic clustering algorithm that considers the dynamic change characteristics of the controllable potential of distributed resources.
[0005] Compared with other distributed resources, the multi-load in the county distribution network is strongly affected by seasonal fluctuations and user behavior, has a relatively dispersed spatial distribution, and has strong electricity flexibility, so its adjustable potential changes in time and space are more complex. Existing literature has not yet considered the impact of this complex dynamic change on the accuracy of aggregated regulation, and there are few studies on dynamic clustering based on the above characteristics of multi-loads; in addition, the dynamic clustering results of multi-loads directly affect the formulation of the optimal regulation scheme, and conversely, the implementation of the optimal regulation scheme will change the adjustable capacity in each period and thus affect the clustering results. It is urgent to establish a coupling relationship between the two. Summary of the invention
[0006] The present invention provides a method and device for optimizing and controlling multi-load hierarchical and grouped dynamic clustering. The present invention considers the influence of complex dynamic changes of the adjustable capacity of multi-load on the control accuracy, carries out dynamic clustering of multi-load based on K-means, and formulates corresponding optimization control and instruction allocation schemes at the cluster and individual levels. The coupling relationship between dynamic clustering and optimization control is established through recursive iteration, which improves the control speed of massive multi-load while ensuring the control accuracy and the utilization of adjustable potential. See the following description for details:
[0007] In a first aspect, a multi-load stratification and grouping optimization control method of dynamic clustering is provided, the method comprising:
[0008] Based on the adjustable capacity model, the adjustable capacity parameters of multiple loads are obtained. According to the adjustable capacity parameters of multiple loads and combined with the geographical locations of multiple loads, a clustering feature set is constructed. Dynamic clustering based on the K-means principle is carried out on the clustering feature set to obtain clustering results, and a dynamic aggregated load aggregation model is constructed according to the dynamic clustering results.
[0009] Taking the load aggregate model as the equivalent constraint, a cluster control optimization model is established. Taking the optimal economic performance of massive multi-loads as the goal, the cluster control optimization model is solved to obtain the optimal control instructions for the aggregate.
[0010] Based on the optimal control instructions of the aggregate, the power distribution control model of multiple loads in the group is established and solved to obtain the control schemes of each multiple load in the group;
[0011] When the clustering results and the control scheme converge, the final clustering results and multi-element load control scheme are obtained.
[0012] The control method is implemented based on a multi-load hierarchical grouping control application architecture, and the multi-load hierarchical grouping control architecture is:
[0013] 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations, backup energy storage and load power consumption of multiple loads, evaluating and calculating the adjustable capacity of multiple loads, uploading the geographical locations, backup energy storage and load power consumption of multiple loads of the terminal layer to the group layer, and receiving and responding to the dynamic grouping and intra-group control instructions issued by the group layer;
[0014] 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. It receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster.
[0015] 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregate as the features change dynamically, and sends the dynamic clustering instructions to the cluster layer; receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer;
[0016] 4) The network layer includes the power grid, which is responsible for providing boundary conditions of time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
[0017] The load aggregation model of the dynamic aggregation is:
[0018] Input X(t) = {x1(t), x2(t), …, x N (t)}t∈{1,…,T} and normalize it to get X * (t), select the cluster number set K = (K1, K2, ..., K Nk ), Nk is the number of candidate K values, and the initial cluster number K is set to K n ;
[0019]
[0020] Where: The characteristic quantity of multivariate load i N is the number of multiple loads; T is the total number of dispatch periods; * (t) randomly selects K features as the initial centroid, whose set u0(t) = {u1(t),u2(t),…,u K (t)}, the center of mass of the kth load aggregate at time t is u k (t);
[0021] To minimize the multivariate load Its corresponding center of mass u k The sum of the Euclidean distances of (t) is used as the target to build an optimization model, and the objective function is:
[0022]
[0023] The optimization model needs to satisfy the following constraints:
[0024]
[0025] Calculate the number of The mean of is taken as the new centroid; for the new centroid, if u' k (t) = u k (t), k=1→K, output the multivariate load clustering result at time t;
[0026] If t = T, the clustering result C of T time periods is output;
[0027] The optimal number of clusters K is determined using the sum of squared errors and the Calinski Harabasz index. op .
[0028] Among them, the clustering regulation optimization model is:
[0029] minF=C om +C grid +C CE -C tra
[0030] The daily operating cost F of massive multi-loads includes the operation and maintenance cost C om , Power purchase cost C grid 、Carbon emission cost C CE and electricity trading income C tra ;
[0031] The power allocation control model within the group is:
[0032] According to the polymer control instructions Determine the energy storage charge and discharge state in the polymer. Then the aggregate k is a charging group like The discharge group Otherwise, it is an idle group.
[0033] P based on multivariate loads i CL (t) and its energy storage SOC i Indicator Construction Evaluation Matrix And standardize and normalize A to get
[0034] Sure and The weight vector W determines the weighted norm matrix
[0035] Sure and Positive and negative ideal solutions for the jth index of internal multivariate loads
[0036] calculate or The distance between each index of internal multivariate load i and its positive and negative ideal solutions
[0037] calculate and Topsis score β of internal multivariate load i evaluation index i , forming a charging and discharging priority queue
[0038] according to and The sequencing controls the charging and discharging of the multiple load backup energy storage in the polymer.
[0039] Among them, when the clustering results and the control plan do not converge, the adjustable capacity is updated, and the recursive iteration of dynamic aggregation-clustering control-intra-cluster allocation is continued until convergence; when the clustering results and the control plan converge, the iteration is stopped; finally, the optimal clustering results and multi-load control plan are obtained.
[0040] In a second aspect, a multi-load stratified and grouped optimization control device with dynamic clustering is provided, the device comprising:
[0041] The solution and construction module obtains the adjustable capacity parameters of multiple loads based on the adjustable capacity model, constructs a clustering feature set based on the adjustable capacity parameters of multiple loads and in combination with the geographical locations of multiple loads, performs dynamic clustering based on the K-means principle on the clustering feature set to obtain clustering results, and constructs a dynamic aggregated load aggregation model based on the dynamic clustering results;
[0042] The module for solving the optimal control instruction of the aggregate is used to establish a cluster control optimization model with the load aggregate model as the equivalent constraint, and solve the cluster control optimization model to obtain the optimal control instruction of the aggregate with the goal of optimizing the economic efficiency of massive multi-loads;
[0043] Solve the control scheme modules of various loads in the group, establish and solve the power distribution control model of various loads in the group based on the optimal control instructions of the aggregate, and obtain the control schemes of various loads in the group;
[0044] The multi-element load control module is used to obtain the final clustering result and the multi-element load control plan when the clustering result and the control plan converge.
[0045] The control device is implemented based on a multi-load hierarchical grouping control application architecture, and the multi-load hierarchical grouping control application architecture is:
[0046] 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations, backup energy storage and load power consumption of multiple loads, evaluating and calculating the adjustable capacity of multiple loads, uploading the geographical locations, backup energy storage and load power consumption of multiple loads of the terminal layer to the group layer, and receiving and responding to the dynamic grouping and intra-group control instructions issued by the group layer;
[0047] 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. It receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster.
[0048] 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregate as the features change dynamically, and sends the dynamic clustering instructions to the cluster layer; receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer;
[0049] 4) The network layer includes the power grid, which is responsible for providing boundary conditions of time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
[0050] In the third aspect, a device for dynamically clustering multi-load stratification and grouping optimization and control comprises: a processor and a memory, wherein the memory stores program instructions, and the processor calls the program instructions stored in the memory to enable the device to execute any one of the methods described in the first aspect.
[0051] In a fourth aspect, a computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes any one of the methods described in the first aspect.
[0052] The beneficial effects of the technical solution provided by the present invention are:
[0053] 1) Considering the spatiotemporal distribution changes of the two types of multi-load clustering characteristics, namely geographical location and adjustable capacity, the present invention proposes a multi-load dynamic clustering method based on the K-means principle, and based on the optimal control instructions of the dynamic aggregate, establishes a multi-load power distribution control model within the group to obtain the control schemes of each multi-load within the group; Compared with the optimization control based on the traditional clustering algorithm, the present invention can ensure the accuracy of the utilization of adjustable potential and the formulation of the control scheme;
[0054] 2) Compared with the centralized optimization and control method of massive multi-loads, the present invention can improve the speed of solving the optimization and control problem while ensuring the accuracy of control; at the same time, the present invention can realize the accurate distribution of cluster control schemes among multi-load individuals through power distribution control of multi-loads in the aggregate;
[0055] 3) The present invention can enable massive and diverse loads to participate in the economic optimization and regulation of the power grid, thereby reducing the electricity costs of diverse loads and improving the economic benefits of power grid operators while meeting the power system's demand for flexible resources. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 It is a flowchart of the implementation process of the multi-load stratification and grouping optimization control method based on dynamic clustering;
[0057] Figure 2 It is a composition diagram of the backup energy storage capacity of multiple loads;
[0058] Figure 3 The structural diagram for the realization of dynamic clustering multi-load stratified grouping control;
[0059] Figure 4 Flowchart for dynamic clustering of multivariate loads;
[0060] Figure 5 Flowchart for solving the power distribution control model within the aggregate;
[0061] Figure 6 It is a schematic diagram of the distribution of multiple loads in the area;
[0062] Figure 7 It is the relationship diagram between the number of clusters K and the clustering effect;
[0063] Figure 8 It is the dynamic clustering result of massive multi-loads and the distribution diagram of equivalent adjustable capacity;
[0064] Fig. 9 A comparison chart of adjustable capacity and backup energy storage capacity for three scenarios;
[0065] Fig.10 Result diagram of optimization and control of power grid for massive and diverse loads;
[0066] Fig.11 It is the control command response diagram based on the power distribution control model. DETAILED DESCRIPTION
[0067] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.
[0068] Considering the reserve energy storage potential of multiple loads, carrying out economic regulation of multiple loads has become an important measure to reduce the cost of power grid operators. Due to the explosion of decision variables and slow solution speed in the centralized regulation of massive multiple loads, they can be aggregated into load aggregate regulation to reduce the scale of calculation. However, the complexity of the spatiotemporal distribution of multiple loads leads to dynamic changes in their controllable potential. Traditional clustering ignores its changes, reduces the accuracy of the regulation scheme, and limits the utilization level of the controllable potential of multiple load reserve energy storage.
[0069] To this end, the embodiment of the present invention proposes a multi-load hierarchical and grouped optimization control method with dynamic clustering. Firstly, based on the two characteristics of the multi-load energy storage controllable capacity and geographical location in the spatiotemporal dimension, a multi-load dynamic clustering method based on K-means is proposed, and the load aggregate is corrected considering the change of the controllable capacity, and a dynamic aggregation model is performed on it; then, for the obtained dynamic load aggregate, a cluster control economic optimization model and an intra-group power allocation control model based on TOPSIS (superiority and inferiority distance method) evaluation are constructed at the cluster and individual control levels, respectively, so as to solve the corresponding control scheme; considering the mutual influence of clustering and control in the proposed method, the optimal clustering and control scheme is solved by recursive iteration of dynamic aggregation-cluster control-intra-group allocation; finally, the proposed method is verified by examples to improve the control speed of massive multi-loads while ensuring the accuracy of control and the utilization of controllable potential.
[0070] Example 1
[0071] The embodiment of the present invention proposes a dynamic clustering multi-load stratification and grouping optimization control method, such as Figure 1 FIG. 1 is a flowchart of an implementation process of the method according to an embodiment of the present invention, wherein the method comprises the following steps:
[0072] 101: Considering the power consumption characteristics of multiple loads, a multi-load adjustable capacity model is constructed, and a multi-load hierarchical and clustered control architecture with dynamic clustering is established;
[0073] Among them, the above-established multi-load adjustable capacity model provides adjustable capacity parameters for subsequent clustering and regulation for the embodiment of the present invention, and the dynamic clustering multi-load hierarchical and grouped regulation architecture is the application scenario architecture provided by the embodiment of the present invention.
[0074] 102: Based on the adjustable capacity model, the adjustable capacity parameters of the multi-loads are obtained, and a clustering feature set is constructed according to the adjustable capacity parameters of the multi-loads and in combination with the geographical locations of the multi-loads. Dynamic clustering based on the K-means principle is carried out on the clustering feature set to obtain clustering results, and a dynamic aggregation load aggregate model is constructed according to the dynamic clustering results; wherein, dynamic clustering is carried out based on the dynamically changing clustering feature set, and the feature set is continuously updated during the clustering process to correct the clustering results, and a load aggregate model is constructed based on the clustering results, and the aggregate is used as an equivalent subject to participate in subsequent regulation.
[0075] 103: Taking the load aggregate model as the equivalent constraint, a cluster control optimization model is established. Taking the optimal economic performance of massive multi-loads as the goal, the cluster control optimization model is solved to obtain the optimal control instructions of the aggregate;
[0076] 104: Based on the optimal control instructions of the aggregate, establish and solve the power distribution control model of multiple loads in the group, and obtain the control scheme of each multiple load in the group;
[0077] By evaluating the charging and discharging priorities of the energy storage of multiple loads within the aggregate, the aggregate control instructions obtained by group control are responded to in order of priority, and the control scheme at the individual level of the multiple loads within the aggregate is obtained.
[0078] 105: Based on the control scheme formulated in steps 101-104, determine whether the clustering result and the control scheme converge; if converged, obtain the final clustering result and the multi-load control scheme; otherwise, return to step 101 to update the adjustable capacity, and continue to perform recursive iteration of dynamic aggregation-clustering control-intra-cluster allocation until it converges.
[0079] Among them, the obtained multi-load control scheme can improve the speed of multi-load control and reduce the electricity cost of massive multi-loads while ensuring the accuracy of control and effective utilization of controllable potential.
[0080] Example 2
[0081] 1. Basic Model and Implementation Architecture
[0082] 1. Modeling of multi-load adjustable capacity
[0083] 1) Common county multi-load power consumption model
[0084] The embodiment of the present invention takes the common electric oven, machine well and 5G base station models in the county distribution network as examples to analyze the adjustable capacity of multiple loads.
[0085] ①Electric baking room power consumption model
[0086] In the process of baking tobacco and tea leaves, traditional baking rooms use coal as raw material and are supplemented by electricity for air blowing. This model causes great pollution to the environment. Against this background, electric tobacco baking technology came into being. Electric tobacco baking technology is safe and reliable, environmentally friendly and pollution-free. It is economical and efficient, can effectively reduce energy costs, and is a clean alternative to traditional baking technology. The essence of an electric baking room is to reduce the moisture content in tobacco and tea leaves by baking. In this process, the power consumption of the electric baking room load can be expressed as:
[0087]
[0088] Where m is the water content in tobacco or tea leaves; W e The amount of water in tobacco or tea leaves that can be evaporated by consuming 1 kWh of electricity; S e is the calorific value of standard coal for electricity; η e is the electrical efficiency of the electric oven; C p,m is the isobaric heat capacity of water; T1 is the ambient temperature; T2 is the temperature during baking.
[0089] In summary, the theoretical total power consumption and actual power consumption of the entire process of the electric baking room are:
[0090] P EBR =P e,1 +P e,2 (4)
[0091] ②Power consumption model of pumped well
[0092] Farmers' irrigation behavior will have a great impact on the well load, especially the time when the irrigation well is opened and the duration of irrigation. When the soil moisture is below a certain threshold, farmers tend to open the well to maintain a good growth environment for crops; when the soil moisture is above a certain threshold, farmers choose to stop irrigation. The power consumption model of the well is estimated based on the actual water output and the water-to-electricity conversion coefficient, which can be expressed as:
[0093] P IP =Q / K IP (5)
[0094] In the formula, K IP is the water-to-electricity conversion coefficient of the irrigation well, and Q is the water output of the well per unit time.
[0095] The time when the pumped well starts working obeys the normal distribution, and its probability density function is:
[0096]
[0097] In the formula, μ s is the expected value of the time when the well starts working, which is 13.0; s s is the corresponding standard deviation, which is 4.9.
[0098] Well working time t d It obeys the log-normal distribution, and its probability density function is:
[0099]
[0100] In the formula, μ d The expected value of the working time of the well is 4.45; s d is the corresponding standard deviation, which is 0.42.
[0101] ③5G base station power consumption model
[0102] Power consumption of base station i at time t P i 5G (t) are as follows:
[0103]
[0104] Where: P i active , P i sleep They are the working and sleeping power consumption of the base station respectively; are the basic circuit power and maximum transmission power of base station i respectively; is the communication load of base station i at time t.
[0105] 2) Multi-load adjustable capacity model
[0106] The backup energy storage of multiple loads needs to retain some power to maintain its normal operation when the power is cut off. Therefore, the backup energy storage capacity of multiple loads can be divided into minimum backup capacity and adjustable capacity. The dynamic changes in the power consumption of multiple loads will lead to differences in the backup energy storage and power demand, such as Figure 2 As shown in (a), during high power consumption periods, the required minimum backup capacity increases and the corresponding adjustable capacity decreases; the opposite is true during low power consumption periods.
[0107] Backup energy storage capacity E of multiple load i at time t i (t) is as follows:
[0108]
[0109] Where: dispatch period △t = 1h; λ E is the energy storage self-discharge rate; η ch , η disc is the energy storage charging and discharging efficiency; P i ch (t), P i disc (t) are the charging and discharging power of the energy storage of the multi-element load i at time t.
[0110] From formula (8), we can get the value of maintaining the operation of multiple loads i when power outage occurs at time t. res Power required As shown in formula (10):
[0111]
[0112] Where: T res is the backup duration of energy storage.
[0113] Comprehensive consideration E i CL (t) and the lower limit of energy storage capacity Minimum available spare capacity Then combined with the energy storage capacity E at time t i (t) Calculation of adjustable capacity
[0114]
[0115] Based on formula (11), the maximum charge and discharge power of the backup energy storage within △t can be calculated:
[0116]
[0117] Where: P i + , P i - They are the maximum charging and discharging power limits of the backup energy storage for the multi-element load i respectively; is the rated capacity of energy storage.
[0118] The multi-load is strongly affected by seasonal fluctuations and user behavior, and its spatial distribution is relatively dispersed, which will further affect the spatiotemporal distribution of power consumption, thereby exacerbating the adjustable capacity of the multi-load in formula (11). The complexity of space-time dynamics.
[0119] 2. Implementation architecture of dynamic clustering multi-load stratification and grouping control method
[0120] The embodiment of the present invention divides the multi-load layered and grouped control into four layers from bottom to top: “end-group-cloud-network”. Figure 3 The functions of each layer are as follows:
[0121] 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations of multiple loads Backup energy storage SOC i And load power consumption P i CL (t) and evaluate the calculation of multivariate loads Upload the above parameters to the group layer, and receive dynamic grouping and intra-group control instructions issued by the response group layer;
[0122] 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. At the same time, it receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster.
[0123] 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregates as the features change dynamically, and sends dynamic clustering instructions to the cluster layer; at the same time, it receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer;
[0124] 4) The network layer is the power grid, which is responsible for providing boundary conditions such as time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
[0125] 2. Dynamically Aggregated Multi-Load Aggregate Model
[0126] 1. Multivariate load dynamic clustering method
[0127] To obtain similar The load aggregate equivalent model and constraints formed by the aggregation of multiple loads ensure To effectively utilize and accurately formulate control plans, the embodiment of the present invention proposes a dynamic clustering method based on K-means.
[0128] 1) Dynamic clustering of multivariate loads based on K-means
[0129] Dynamic clustering process Figure 4 As shown, the steps are as follows:
[0130] Step 1: Input X(t) = {x1(t), x2(t), …, x N (t)}t∈{1,…,T}, and normalize it using formula (13). Input cluster number set K=(K1,K2,…,K Nk ), where Nk is the number of candidate K values, and the initial cluster number K is set to K n (initial n = 1);
[0131]
[0132] Where: The characteristic quantity of multivariate load i N is the number of multivariate loads; T is the total number of scheduling periods, and in the embodiment of the present invention, T=24h; Normalized feature set
[0133] Step 2: In X *(t) randomly selects K features as the initial centroid, whose set u0(t) = {u1(t),u2(t),…,u K (t)}, where the center of mass of the kth load aggregate at time t is u k (t);
[0134] Step 3: Minimize the multivariate loads Its corresponding center of mass u k The sum of the Euclidean distances of (t) is used as the target to build an optimization model for clustering. The objective function is shown in formula (14):
[0135]
[0136] This formula represents Assign to the nearest u k (t) The aggregate k to which it belongs.
[0137] Subsequent group control involves the balance constraint of the backup energy storage capacity of multiple loads across time periods (9). Therefore, the dynamic load aggregate, as the control subject, must have the same number of multiple loads in each time period. Therefore, the optimization model satisfies the constraint condition (15):
[0138]
[0139] Where: n k (t) is the number of multivariate loads in load aggregate k. The constraint is implemented as follows: if the number of multivariate loads in aggregate k is less than n k (t), then the multivariate load i is assigned to the aggregate, otherwise u is removed from the centroid set. k (t) and return to equation (14) to be Assign centroids until all N multivariate loads are assigned.
[0140] Step 4: Use formula (16) to calculate the The mean of , which is taken as the new centroid;
[0141]
[0142] Where: C k (t) is the multivariate load in load aggregate k Composed of a collection.
[0143] Step 5: For the new centroid, if equation (17) is satisfied:
[0144] u' k (t) = u k (t), k=1→K (17)
[0145] Then the output multivariate load clustering result at time t is C(t) = {C1(t), C2(t),…, C K (t)}, u(t)={u1(t),u2(t),…,u K (t)}, otherwise repeat the above steps 3 and 4.
[0146] Step 6: If t = T, then output the clustering result C of T time periods, as shown in formula (11). Otherwise, enter the next time period and return to step 1;
[0147] C={C1(t),C2(t),…,C K (t)} t∈{1,…,T} (18)
[0148] Step 7: For the clustering result C, use the error sum of square (SSE) and Calinski Harabasz (CH) index to determine the optimal number of clusters K op .
[0149] Among them, the optimal number of clusters K op It is within the inflection point range where the SSE curve changes significantly. At the same time, the higher the CH value, the better the clustering effect. Comprehensive judgment can be used to obtain K op , get K op Then the dynamic clustering ends, otherwise let K = K n+1 And return to step 1 to continue clustering.
[0150] 2) Optimal number of clusters selection
[0151] The calculation method of SSE is shown in formula (19):
[0152]
[0153] Among them, I SSE (t) represents the multivariate load characteristics within each load aggregate The sum of the squares of the distances to the corresponding centroids. SSE The smaller (t) is, the better the clustering effect is, that is, multiple loads with close geographical locations and similar controllable capacities are divided into the same aggregate. The calculation method of CH index is shown in formula (20):
[0154]
[0155] Among them, tr B With tr W The expression is as follows:
[0156]
[0157] Among them, u(t) is the overall centroid of the massive multivariate load feature set at time t, as shown in formula (22):
[0158]
[0159] Among them, the CH index represents the ratio of the separation degree and compactness of each load polymer, so I CH The larger (t) is, the better the clustering effect is.
[0160] From equations (19)-(21), we can see that the dynamic changes in clustering results will lead to the I SSE (t) and I CH (t) changes accordingly, so the embodiment of the present invention adopts I of T time periods SSE (t), I CH The average value of (t) is taken as the final I SSE with I CH , the expression is as follows:
[0161]
[0162] Considering the above indicators, we can get K op , which can carry out dynamic clustering with optimal effect.
[0163] 2. Multivariate load aggregate model
[0164] Based on the dynamic clustering results, the embodiment of the present invention establishes a multi-load aggregate model, which is used as an equivalent constraint condition to participate in subsequent optimization and regulation. The multi-load-energy storage monomer model is as follows:
[0165]
[0166] Where: SOC i (t) is the state of charge of the backup energy storage of the multi-element load i at time t; SOC i,min (t) and SOC i,max (t) are the lower and upper limits of the charge state of the backup energy storage of the multi-element load i at time t respectively; are binary parameters that measure the charge and discharge status of the backup energy storage of the multi-load i at time t. Then the energy storage is only charged, otherwise it is only discharged.
[0167] For simplicity, the embodiment of the present invention assumes that the λ of each multi-load backup energy storage E , η ch With η disc The same can be used as the equivalent parameters of multi-load polymers. In addition, the energy storage charge and discharge states in polymer k are consistent, and the state parameters can be used and Unified representation, that is The polymer energy storage is charged. Similarly, as shown in formula (25):
[0168]
[0169] Based on the above assumptions, all multivariate load models in the group are linearly superimposed according to formula (17), and the load aggregation model can be obtained as follows:
[0170]
[0171] Where: E kΣ (t) and are the equivalent reserve energy storage capacity, equivalent charge and discharge power and their upper limits of polymer k at time t, which are obtained by linearly superposing the reserve energy storage in the polymer; SOC k (t), SOC k,max (t) and SOC k,min (t) are the average values of the SOC of each energy storage in the polymer and its upper and lower limits, as shown in formula (27):
[0172]
[0173] 3. Dynamic Clustering Multi-Load Layered and Grouped Control Method
[0174] 1. Multi-load grouping control optimization model
[0175] Under the time-of-use electricity price mechanism of the power grid, an optimization model is constructed with the goal of minimizing the daily operating cost F of massive multi-loads, and the optimization and regulation of multi-load aggregates is realized between "clouds and clusters". Among them, F includes the operation and maintenance cost C om , Power purchase cost C grid 、Carbon emission cost C CE and electricity trading income C tra , as shown in formula (28):
[0176] minF=C om +C grid +C CE -C tra (28)
[0177] Among them, C om The calculation is as follows:
[0178]
[0179] Among them, C grid The calculation is as follows:
[0180]
[0181] Among them, CCE The calculation is as follows:
[0182]
[0183] Among them, C tra The calculation is as follows:
[0184]
[0185] In formulas (28)-(32): Unit operation and maintenance cost of energy storage for multi-load aggregates; e grid (t) are the power purchased by aggregate k from the upper grid and the grid time-of-use electricity price at time t; e CE , μ CE are carbon trading price and grid carbon emission factor respectively.
[0186] The equivalent constraints for each multi-load aggregate are as follows:
[0187] 1) Constraints on polymer energy storage:
[0188] The constraints of polymer energy storage include: capacity balance constraint, charge and discharge power constraint, SOC constraint and charge and discharge state constraint, which can be seen in formula (26);
[0189] 2) Aggregate power balance constraints:
[0190]
[0191] Where: It is the sum of the power consumption of multiple loads in aggregate k.
[0192] 3) Transmission power constraints:
[0193]
[0194] Where: and They are respectively the upper and lower limits of the transmission power allowed by the current regional power grid.
[0195] 2. Power distribution control model of multiple loads within the cluster
[0196] For multiple load power consumption P i CL(t) and the two-dimensional evaluation index composed of the SOC of the backup energy storage. The embodiment of the present invention constructs a power distribution control model at the "group-end" layer, and uses the TOPSIS method to quantify the charging and discharging priority of the backup energy storage of multiple loads in the aggregate, and controls the backup energy storage to charge and discharge at rated power according to the priority order, thereby avoiding the situation where the multiple load energy storage with high SOC is charged and the multiple load energy storage with low SOC is discharged, and realizes the reasonable distribution of the load aggregate control instructions among the multiple loads in the aggregate. The model solution process is as follows: Figure 5 As shown, the specific steps are as follows:
[0197] Step 1: According to Determine the energy storage charge and discharge state in the polymer: Then the aggregate k is a charging group like The discharge group Otherwise, it is an idle group.
[0198] Step 2: Based on the n in aggregate k at time t k (t) P of multiple loads i CL (t) and its energy storage SOC i Two types of indicators to construct an evaluation matrix where a ij is the jth index value of multivariate load i. The TOPSIS evaluation index can be divided into positive and negative types. The larger the positive index and the smaller the negative index, the higher the scheme score. From formulas (8)-(12), we can see that: P i CL The smaller (t) is, the SOC i The larger the value is, the lower the backup energy storage charging priority of the multi-element load i is and the higher the discharge priority is, that is, the correlation between the charging and discharging priority and the score is opposite. This model assumes that the discharge priority is positively correlated with the score, that is, P i CL (t) is a negative indicator and needs to be processed positively. The processing method is shown in formula (35):
[0199]
[0200] Then A is standardized and normalized to obtain
[0201] Step 3: Determine and The weight vector W, where P CL The weights of SOC are ω load ,ω SOC , and then determine the weighted normative matrix
[0202]
[0203] Step 4: Determine and Positive and negative ideal solutions for the jth index of internal multivariate loads
[0204] Step 5: Calculate separately or The distance between each index of internal multivariate load i and its positive and negative ideal solutions
[0205]
[0206] Step 6: Calculate separately and Topsis score β of two-dimensional evaluation index of internal multivariate load i i ;
[0207]
[0208] Among them, charging priority and β i Negatively correlated, so for β i Arrange in ascending order to form a charging priority queue On the contrary, for β i Arrange in descending order to form a discharge priority queue
[0209] Step 7: According to and The sequencing control of the multiple load backup energy storage charging and discharging in the polymer:
[0210] 1) Yes Middle front n c Multiple load energy storage applies charging instructions (with charging) to minimize the total charging power of the multi-load energy storage in the aggregate With polymer charging instructions The error is used as the target to build an optimization model, and the objective function is as follows:
[0211]
[0212] At the same time c It must be within the range of the number of multivariate loads of aggregate k, that is, it must satisfy constraint (41):
[0213] n c ∈(1,2,…,n k (t)) (41)
[0214] 2) Yes Middle front nd Multiple load energy storage applies discharge instructions (with Discharge), the objective function and constraint conditions can be obtained as shown in equations (42)-(43):
[0215]
[0216] n d ∈(1,2,…,n k (t)) (43)
[0217] 3. Dynamic clustering multi-load stratification and grouping control solution process
[0218] The solution process of the method proposed in the embodiment of the present invention is as follows:
[0219] Step 1: Enter the maximum number of iterations iter max , set the number of iterations iter = 1 and The initial value of the iteration As shown in formula (44):
[0220]
[0221] Where: is the initial capacity of the backup energy storage of the multi-load i.
[0222] Step 2: Use the dynamic aggregation model to solve the clustering results;
[0223] Step 3: Use the clustering control optimization model to solve the control scheme of multi-load aggregates;
[0224] Step 4: Use the intra-cluster allocation control model to allocate aggregate control instructions and solve the multi-load control scheme within the aggregate;
[0225] Step 5: Perform iteration termination judgment, if iter = iter max Or the daily operating cost F satisfies the convergence criterion (45), the iteration is stopped:
[0226] |F(iter)-F(iter-1)|≤e (45)
[0227] Where: ε is the convergence accuracy. The clustering results and control schemes affected by the load no longer fluctuate greatly, and the optimal clustering results and control schemes are output; otherwise iter = iter + 1, based on the multi-load control scheme, the multi-load control scheme is combined with equations (9) and (11) to update each multi-load And return to step 2.
[0228] Example 3
[0229] The embodiment of the present invention takes the multi-load in the 2.65km*2.65km area as the research object. The area contains 2916 large loads and small loads in three different functional areas (including 36 large loads and 2880 small loads). Figure 6 The energy storage related parameters and time-of-use electricity prices are shown in Table 1-2. The power consumption difference between large loads and small loads is large, so the large loads in each region participate in hierarchical clustering optimization and regulation as independent clusters. The embodiment of the present invention only carries out dynamic clustering for massive county-level small loads.
[0230] Table 1 Parameters of energy storage
[0231]
[0232]
[0233] Table 2 Standard time of use electricity price
[0234]
[0235] 1. Optimal number of clusters K op Select Discussion
[0236] Depend on Figure 7 It can be seen that as the K value gradually increases, I SSE It gradually becomes smaller and tends to be stable. Figure 7 When K=3 to K=9, I SSE The descent speed slows down significantly, so there is an elbow inflection point; then combined with I CH The K value is screened twice: From formula (20), we can see that when K = 1, I CH The value does not exist, so it is not discussed. When K=6, the CH index value is 1837.4, which is much higher than I when K takes other values. CH value, so the embodiment of the present invention selects K op =6 as the optimal number of clusters, and K op =6 Expand analysis.
[0237] 2. Dynamic clustering results of massive multi-loads
[0238] Depend on Figure 8 It can be seen that the clustering results at each moment are different:
[0239] 1) t = 6h: The power consumption of multiple loads in each functional area is relatively low, so Figure 8In (a), the controllable capacity of each multi-load and the controllable capacity of the load aggregate are both high, and the geographical location plays a leading role in the clustering, that is, each aggregate is evenly divided according to the geographical location;
[0240] 2) t = 14h: As time goes by, Figure 8 The functional characteristics of each region in (b) determine that its power consumption is relatively high, and the adjustable capacity of each aggregate is relatively low. The adjustable capacity of aggregates 2 and 5 is relatively small, and they are still evenly divided by geographical location. However, the adjustable capacity of aggregates 1, 3, and 6 is significantly different, so their clustering is more affected by the adjustable capacity distribution, and the distribution of aggregates is no longer uniform.
[0241] 3) t = 22h: The power consumption in commercial and office areas is relatively low, and the power load is mainly concentrated in residential areas. Figure 8 In (c), the adjustable capacity of backup energy storage of aggregates 3, 4, and 5 is much higher than that of aggregates 1, 2, and 6. In addition, the power consumption of each multi-load is not only related to the geographical distribution. Under the dynamic clustering method, the cross-regional aggregation of aggregates 1, 3, and 6 is dominated by the adjustable capacity characteristics of energy storage, that is, the influence of adjustable capacity distribution on the aggregation of multi-loads is more significant.
[0242] In summary, the dynamic clustering method proposed in the embodiment of the present invention aggregates multiple loads with similar geographical locations and similar adjustable capacities at each moment, and dynamically corrects the load aggregate composition according to the change of the energy storage adjustable capacity, thereby obtaining K multiple load aggregates with different aggregate adjustable capacities; subsequently, the charging and discharging of the aggregate can be reasonably regulated based on the difference in the adjustable capacities of each aggregate, and at the same time, since the adjustable capacities of each multiple load in the aggregate are close, the energy storage charging of the multiple loads in the low-capacity aggregate and the energy storage discharging of the multiple loads in the high-capacity aggregate can be promoted, thereby ensuring the accuracy of the regulation scheme.
[0243] 3. Comparison between dynamic clustering and traditional clustering
[0244] Centralized optimization control has the characteristics of slow calculation speed and high accuracy. In order to compare the impact of dynamic clustering and traditional clustering on control accuracy, the embodiment of the present invention takes the centralized control results as a benchmark. Therefore, the embodiment of the present invention constructs the following three scenarios for comparison:
[0245] Scenario I: centralized control of multiple loads (equivalent to the number of clusters K = 2880);
[0246] Scenario II: Traditional clustering multi-load stratification and grouping control (K=6);
[0247] Scenario III: Dynamic clustering of multi-load stratification and grouping control (K=6).
[0248] Among them, traditional clustering considers the overall trend of the adjustable capacity curve of multiple loads, and clusters them based on the geographical location and the relative deviation of the adjustable capacity curve. The resulting aggregate is fixed and will not be corrected with the change of adjustable capacity. Fig. 9 shown.
[0249] Depend on Fig. 9 It can be seen that: in the period of 1:00-10:00, the adjustable capacity and backup energy storage capacity of scenarios II and III are consistent with those of scenario I. This is because the backup energy storage capacity of the multiple loads is relatively high at the initial moment and is in a charging state. The energy storage charging power of scenarios II and III is subject to the same upper limit of energy storage capacity, and it is difficult to produce deviations; however, after 11:00, the energy storage discharge is affected by the equivalent adjustable capacity of the load aggregate. Combined with formula (12), it can be seen that the upper bound of the discharge constraint of the multiple load aggregate in formula (26) is different. At this time, the multiple loads in the aggregate of scenario II are more likely to have insufficient adjustable potential and discharge exceeding the adjustable capacity limit. This situation is particularly significant after 17:00, which increases the error between the adjustable capacity and the backup energy storage capacity.
[0250] However, compared with scenario II, scenario III uses a dynamic clustering method to obtain consistent adjustable capacity within the aggregate, but different equivalent adjustable capacities between aggregates. A reasonable control scheme can be formulated based on the differences in equivalent adjustable potential of each aggregate. Therefore, at most times in scenario III, the errors between adjustable capacity and backup energy storage capacity are small, and the overall trend of the two is closer to scenario I. Therefore, the embodiments of the present invention can effectively ensure the full utilization of adjustable potential and the accuracy of the control scheme.
[0251] In addition, in order to verify the effectiveness of this strategy for economic operation of multiple loads, the following is introduced:
[0252] Scenario IV: Mains power supply ignoring the potential of backup energy storage.
[0253] The comparison of daily operating costs of massive multi-loads in four scenarios is shown in Table 3.
[0254] Table 3 Comparison of daily operating costs of massive multi-loads under different scenarios
[0255] Tab.3Comparison of daily operating costs of massive multiple loads in different scenarios
[0256]
[0257] 1) The operating costs of multiple loads under scenarios II and III are 311.18 yuan and 312.83 yuan respectively, which are lower than those of scenario I. This is because scenario I strictly considers the operating constraints of each multiple load, while scenarios II and III carry out optimization and regulation based on the equivalent constraints of their respective load aggregates. Their feasible domains deviate from scenario I, that is, in actual regulation, it may not be possible to consider all the individual operating constraints of multiple loads, so the optimal solution obtained within the corresponding feasible domain also deviates accordingly.
[0258] 2) The deviations between scenarios II and III and scenario I are 0.80% and 0.27% respectively. This is because the dynamic clustering method can be used to classify the Changes dynamically modify the cluster composition and update the aggregate And equivalent constraints (21), such as Figure 8 Therefore, compared with traditional clustering, dynamic clustering can provide more accurate equivalent constraints for optimization and control, and improve the accuracy of clustering and control;
[0259] 3) Compared with scenario IV, the daily operating cost of scenario III is reduced by 10.13%. Therefore, the strategy in this paper utilizes idle energy storage of multiple loads to participate in grid regulation, which can reduce the electricity cost of multiple loads and improve the economic benefits of grid operators.
[0260] 4. Massive and diverse loads participate in the discussion of power grid regulation results
[0261] Depend on Fig.10 It can be seen that: during the electricity price valley period of 01:00-06:00, the backup energy storage of the multiple loads is charged to reserve the electricity required for subsequent discharge, and electricity is purchased from the upper-level power grid to maintain the supply and demand balance of the multiple loads. At this time, the SOC of the energy storage of each multiple load aggregate gradually increases, and charging stops after reaching the upper limit; during the electricity price peak periods of 11:00-12:00, 16:00 and 18:00-19:00, the backup energy storage of the multiple loads is discharged, thereby reducing the amount of purchased electricity to achieve arbitrage, and the SOC of each aggregate energy storage decreases accordingly; during the electricity price normal periods of 15:00, 17:00 and 21:00-24:00, the energy storage charging is used to supplement the power consumption of the discharge, so as to ensure that at the end of the regulation, the SOC of each aggregate remains at the initial level, which is convenient for the regulation of the next day.
[0262] To sum up: massive and diverse loads use their own controllable energy storage to carry out economic regulation with the power grid, and arbitrage in the process of low storage and high release, which can reduce the operating costs of power grid operators.
[0263] Table 4 Calculation time for each number of polymers K
[0264] Table 4Calculation time for the number of clusters K
[0265]
[0266]
[0267] From Table 4, we can see that:
[0268] 1) As the K value increases, the corresponding calculation time T total Gradually increases; in the process of K increasing from 3 to 6, due to the small number of polymers, T total The rise is small;
[0269] 2) As the K value increases, T total The increase is significant, for example, T corresponding to K = 9 total Compared with the optimal clustering number (K op =6) increased by 417.26%, and the calculation speed decreased significantly;
[0270] 3) Compared with K=2880 (centralized optimization control), K op =6 multi-load stratification and grouping control time is reduced to 2.34% of the former, while Fig. 9 As shown, the control accuracy is still high. Therefore, the embodiment of the present invention can significantly improve the calculation speed of massive multi-load control problems while ensuring the control accuracy.
[0271] 5. Analysis of response results of power distribution within the group
[0272] Based on the power distribution control model, the response of each multi-load in the group to the load aggregate control command can be realized. The response results are as follows: Fig.11 As shown in the figure, it can be seen that the multi-load in the group can track the aggregate control instruction more accurately, and the relative error is the largest at 16:00, which is only 6.06%. Therefore, the power allocation control model at the group level proposed in the embodiment of the present invention has high accuracy.
[0273] Example 4
[0274] A multi-load stratified grouping optimization control device with dynamic clustering, the device comprising:
[0275] The solution and construction module obtains the adjustable capacity parameters of multiple loads based on the adjustable capacity model, constructs a clustering feature set based on the adjustable capacity parameters of multiple loads and in combination with the geographical locations of multiple loads, performs dynamic clustering based on the K-means principle on the clustering feature set to obtain clustering results, and constructs a dynamic aggregated load aggregation model based on the dynamic clustering results;
[0276] The module for solving the optimal control instruction of the aggregate is used to establish a cluster control optimization model with the load aggregate model as the equivalent constraint, and solve the cluster control optimization model to obtain the optimal control instruction of the aggregate with the goal of optimizing the economic efficiency of massive multi-loads;
[0277] Solve the control scheme modules of various loads in the group, establish and solve the power distribution control model of various loads in the group based on the optimal control instructions of the aggregate, and obtain the control schemes of various loads in the group;
[0278] The multi-element load control module is used to obtain the final clustering result and the multi-element load control plan when the clustering result and the control plan converge.
[0279] Among them, the control device is implemented based on the multi-load layered and grouped control application architecture, and the multi-load layered and grouped control application architecture is:
[0280] 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations, backup energy storage and load power consumption of multiple loads, evaluating and calculating the adjustable capacity of multiple loads, uploading the geographical locations, backup energy storage and load power consumption of multiple loads of the terminal layer to the group layer, and receiving and responding to the dynamic grouping and intra-group control instructions issued by the group layer;
[0281] 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. It receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster.
[0282] 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregate as the features change dynamically, and sends the dynamic clustering instructions to the cluster layer; receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer;
[0283] 4) The network layer includes the power grid, which is responsible for providing boundary conditions of time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
[0284] Example 5
[0285] A multi-load stratified grouping optimization control device with dynamic clustering, the device comprises: a processor and a memory, the memory stores program instructions, the processor calls the program instructions stored in the memory to make the device execute the following method steps in embodiment 1:
[0286] Based on the adjustable capacity model, the adjustable capacity parameters of multiple loads are obtained. According to the adjustable capacity parameters of multiple loads and combined with the geographical locations of multiple loads, a clustering feature set is constructed. Dynamic clustering based on the K-means principle is carried out on the clustering feature set to obtain clustering results, and a dynamic aggregated load aggregation model is constructed according to the dynamic clustering results.
[0287] Taking the load aggregate model as the equivalent constraint, a cluster control optimization model is established. Taking the optimal economic performance of massive multi-loads as the goal, the cluster control optimization model is solved to obtain the optimal control instructions for the aggregate.
[0288] Based on the optimal control instructions of the aggregate, the power distribution control model of multiple loads in the group is established and solved to obtain the control schemes of each multiple load in the group;
[0289] When the clustering results and the control scheme converge, the final clustering results and multi-element load control scheme are obtained.
[0290] The control is implemented based on a multi-load hierarchical grouping control application architecture, and the multi-load hierarchical grouping control architecture is:
[0291] 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations, backup energy storage and load power consumption of multiple loads, evaluating and calculating the adjustable capacity of multiple loads, uploading the geographical locations, backup energy storage and load power consumption of multiple loads of the terminal layer to the group layer, and receiving and responding to the dynamic grouping and intra-group control instructions issued by the group layer;
[0292] 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. It receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster.
[0293] 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregate as the features change dynamically, and sends the dynamic clustering instructions to the cluster layer; receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer;
[0294] 4) The network layer includes the power grid, which is responsible for providing boundary conditions of time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
[0295] Among them, the load aggregation model of dynamic aggregation is:
[0296] Input X(t) = {x1(t), x2(t), …, x N(t)}t∈{1,…,T} and normalize it to get X * (t), select the cluster number set K = (K1, K2, ..., K Nk ), Nk is the number of candidate K values, and the initial cluster number K is set to K n ;
[0297]
[0298] Where: The characteristic quantity of multivariate load i N is the number of multiple loads; T is the total number of dispatch periods;
[0299] In X * (t) randomly selects K features as the initial centroid, whose set u0(t) = {u1(t),u2(t),…,u K (t)}, the center of mass of the kth load aggregate at time t is u k (t);
[0300] To minimize the multivariate load Its corresponding center of mass u k The sum of the Euclidean distances of (t) is used as the target to build an optimization model, and the objective function is:
[0301]
[0302] The optimization model needs to satisfy the following constraints:
[0303]
[0304] Calculate the number of The mean of is taken as the new centroid; for the new centroid, if u' k (t) = u k (t), k=1→K, output the multivariate load clustering result at time t;
[0305] If t = T, the clustering result C of T time periods is output;
[0306] The optimal number of clusters K is determined using the sum of squared errors and the Calinski Harabasz index. op .
[0307] Among them, the clustering control optimization model is:
[0308] minF=C om +C grid +C CE -C tra
[0309] The daily operating cost F of massive multi-loads includes the operation and maintenance cost C om, Power purchase cost C grid 、Carbon emission cost C CE and electricity trading income C tra ;
[0310] Among them, the power allocation control model within the group is:
[0311] According to the polymer control instructions Determine the energy storage charge and discharge state in the polymer. Then the aggregate k is a charging group like The discharge group Otherwise, it is an idle group.
[0312] P based on multivariate loads i CL (t) and its energy storage SOC i Indicator Construction Evaluation Matrix And standardize and normalize A to get
[0313] Sure and The weight vector W determines the weighted norm matrix
[0314] Sure and Positive and negative ideal solutions for the jth index of internal multivariate loads
[0315] calculate or The distance between each index of internal multivariate load i and its positive and negative ideal solutions
[0316] calculate and Topsis score β of internal multivariate load i evaluation index i , forming a charging and discharging priority queue
[0317] according to and The sequencing controls the charging and discharging of the multiple load backup energy storage in the polymer.
[0318] Among them, when the clustering results and the control plan do not converge, the adjustable capacity is updated, and the recursive iteration of dynamic aggregation-clustering control-intra-cluster allocation is continued until convergence; when the clustering results and the control plan converge, the iteration is stopped; finally, the optimal clustering results and multi-load control plan are obtained.
[0319] The execution subjects of the above-mentioned processor and memory can be devices with computing functions such as computers, single-chip microcomputers, and microcontrollers. In specific implementation, the embodiments of the present invention do not limit the execution subjects and are selected according to the needs of actual applications.
[0320] The data signal is transmitted between the memory and the processor via a bus, which is not described in detail in the embodiment of the present invention.
[0321] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, the storage medium includes a stored program, and when the program is running, the device where the storage medium is located is controlled to execute the method steps in the above embodiment. The computer-readable storage medium includes but is not limited to a flash memory, a hard disk, a solid-state hard disk, etc.
[0322] It should be noted here that the description of the readable storage medium in the above embodiment corresponds to the description of the method in the embodiment, and the embodiments of the present invention are not repeated here. In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present invention is generated in whole or in part.
[0323] The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions may be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer-readable storage medium may be any available medium that can be accessed by the computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium or a semiconductor medium, etc.
Claims
1. A dynamic clustering multi-load stratification and grouping optimization control method, characterized in that: The method comprises: Based on the adjustable capacity model, the adjustable capacity parameters of multiple loads are obtained. According to the adjustable capacity parameters of multiple loads and combined with the geographical locations of multiple loads, a clustering feature set is constructed. Dynamic clustering based on the K-means principle is carried out on the clustering feature set to obtain clustering results, and a dynamic aggregated load aggregation model is constructed according to the dynamic clustering results. Taking the load aggregate model as the equivalent constraint, a cluster control optimization model is established. Taking the optimal economic performance of massive multi-loads as the goal, the cluster control optimization model is solved to obtain the optimal control instructions for the aggregate. Based on the optimal control instructions of the aggregate, the power distribution control model of multiple loads in the group is established and solved to obtain the control schemes of each multiple load in the group; When the clustering results and the control scheme converge, the final clustering results and multi-element load control scheme are obtained.
2. The method for optimizing and controlling multi-load stratification and grouping based on dynamic clustering according to claim 1, characterized in that: The control method is implemented based on a multi-load stratified grouping control application architecture, and the multi-load stratified grouping control architecture is: 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations, backup energy storage and load power consumption of multiple loads, evaluating and calculating the adjustable capacity of multiple loads, uploading the geographical locations, backup energy storage and load power consumption of multiple loads of the terminal layer to the group layer, and receiving and responding to the dynamic grouping and intra-group control instructions issued by the group layer; 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. It receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster. 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregate as the features change dynamically, and sends the dynamic clustering instructions to the cluster layer; receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer; 4) The network layer includes the power grid, which is responsible for providing boundary conditions of time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
3. The method for optimizing and controlling multi-load stratification and grouping based on dynamic clustering according to claim 1, characterized in that: The load aggregation model of the dynamic aggregation is: Input X(t) = {x1(t), x2(t), …, x N (t)}t∈{1,…,T} and normalize it to get X * (t), select the cluster number set K = (K1, K2, ..., K Nk ), Nk is the number of candidate K values, and the initial cluster number K is set to K n ; Where: The characteristic quantity of multivariate load i N is the number of multiple loads; T is the total number of dispatch periods; In X * (t) randomly selects K features as the initial centroid, whose set u0(t) = {u1(t),u2(t),…,u K (t)}, the center of mass of the kth load aggregate at time t is u k (t); To minimize the multivariate load Its corresponding center of mass u k The sum of the Euclidean distances of (t) is used as the target to build an optimization model, and the objective function is: The optimization model needs to satisfy the following constraints: Calculate the number of The mean of is taken as the new centroid; for the new centroid, if u' k (t) = u k (t), k=1→K, output the multivariate load clustering result at time t; If t = T, the clustering result C of T time periods is output; The optimal number of clusters K is determined using the sum of squared errors and the Calinski Harabasz index. op .
4. The method for optimizing and controlling multi-load stratification and grouping based on dynamic clustering according to claim 1, characterized in that: The clustering control optimization model is: minF=C om +C grid +C CE -C tra The daily operating cost F of massive multi-loads includes the operation and maintenance cost C om , Power purchase cost C grid 、Carbon emission cost C CE and electricity trading income C tra ; The power allocation control model within the group is: According to the polymer control instructions Determine the energy storage charge and discharge state in the polymer. Then the aggregate k is a charging group like The discharge group Otherwise, it is an idle group. P based on multivariate loads i CL (t) and its energy storage SOC i Indicator Construction Evaluation Matrix And standardize and normalize A to get Sure and The weight vector W determines the weighted norm matrix Sure and Positive and negative ideal solutions for the jth index of internal multivariate loads calculate or The distance between each index of internal multivariate load i and its positive and negative ideal solutions calculate and Topsis score β of internal multivariate load i evaluation index i , forming a charging and discharging priority queue according to and The sequencing controls the charging and discharging of the multiple load backup energy storage in the polymer.
5. The method for optimizing and controlling multi-load stratification and grouping based on dynamic clustering according to claim 1, characterized in that: When the clustering results and the control scheme do not converge, the adjustable capacity is updated, and the recursive iteration of dynamic aggregation-clustering control-intra-cluster allocation is continued until convergence; when the clustering results and the control scheme converge, the iteration is stopped; finally, the optimal clustering results and multi-load control scheme are obtained.
6. A dynamic clustering multi-load stratification and grouping optimization control device, characterized in that: The device comprises: The solving and building module obtains the adjustable capacity parameters of multiple loads based on the adjustable capacity model, builds a clustering feature set based on the adjustable capacity parameters of multiple loads and in combination with the geographical locations of multiple loads, conducts dynamic clustering based on the K-means principle for the clustering feature set to obtain clustering results, and builds a dynamic aggregate load aggregate model based on the dynamic clustering results; the module for solving the optimal control instruction of the aggregate is used to establish a clustering control optimization model with the load aggregate model as an equivalent constraint, and solves the clustering control optimization model to obtain the optimal control instruction of the aggregate with the goal of optimizing the economic efficiency of massive multiple loads; Solve the control scheme modules of various loads in the group, establish and solve the power distribution control model of various loads in the group based on the optimal control instructions of the aggregate, and obtain the control schemes of various loads in the group; The multi-element load control module is used to obtain the final clustering result and the multi-element load control plan when the clustering result and the control plan converge.
7. The device for optimizing and controlling multi-load stratification and grouping based on dynamic clustering according to claim 6 is characterized in that: The control device is implemented based on a multi-load layered and grouped control application architecture, and the multi-load layered and grouped control application architecture is: 1) The terminal layer is located on the load side and is responsible for collecting the geographical locations, backup energy storage and load power consumption of multiple loads, evaluating and calculating the adjustable capacity of multiple loads, uploading the geographical locations, backup energy storage and load power consumption of multiple loads of the terminal layer to the group layer, and receiving and responding to the dynamic grouping and intra-group control instructions issued by the group layer; 2) The cluster layer is composed of multiple load cluster controllers. It builds a multiple load aggregate model based on the clustering instructions and uses it as a control equivalent constraint, and reports it to the cloud layer together with the clustering feature set. It receives and responds to the aggregate control instructions issued by the cloud layer, and decomposes them into multiple loads within the cluster through the power allocation control model within the cluster. 3) The cloud layer performs dynamic clustering based on the cluster feature set, modifies the aggregate as the features change dynamically, and sends the dynamic clustering instructions to the cluster layer; receives the time-of-use electricity price of the power grid and the equivalent constraints of the multi-load aggregates of the cluster layer, formulates the control schemes for each multi-load aggregate through the cluster control optimization model, and uploads them to the network layer; 4) The network layer includes the power grid, which is responsible for providing boundary conditions of time-of-use electricity prices and transmission power constraints for cloud layer decisions, and interacting with load aggregates according to the control plan.
8. A dynamic clustering multi-load stratification and grouping optimization control device, characterized in that: The device comprises: a processor and a memory, wherein program instructions are stored in the memory, and the processor calls the program instructions stored in the memory to enable the device to execute any one of the methods of claims 1-5.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 5.
Citation Information
Cited By
Intelligent power distribution and control method and device based on load group identification
CN119726750A
Rural distributed resource-oriented hierarchical reinforcement learning regulation and control method and system
CN121481055A
A hierarchical reinforcement learning regulation method and system for rural distributed resources
CN121481055B
Farm power distribution network layered optimization regulation and control method and system considering irrigation load
CN121939406A
