Power distribution network group intelligent cooperative operation terminal

Abstract

The invention relates to a power distribution network group intelligent cooperative operation terminal comprising a modeling module used for modeling a power distribution network which comprises distributed equipment and non-distributed equipment; and the decision tree module is used for constructing a decision number model for the modeled power distribution network, and performing layer-by-layer regulation and control according to the decision number model to realize cooperative operation. Compared with the prior art, the method has the advantages of realizing instant interconnection between a power grid and the outside, realizing control of equipment class groups and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of power distribution networks, and in particular to a group intelligent collaborative operation terminal for power distribution networks. Background Art

[0002] In microgrids and distributed energy systems, island mode operation is one of the key scenarios, especially in remote areas or emergency situations, when the main power grid is unavailable, it is crucial to ensure the stability and reliability of power supply. The distributed coordinated control optimization strategy based on the island scenario aims to achieve efficient energy utilization and stable operation of the system through intelligent management of distributed photovoltaic (PV), wind power, diesel generator sets (DEG) and power loads with island characteristics. However, the coordinated complementarity of multiple energy sources increases the complexity of active distribution networks. Since different energy networks have different models and characteristics, and interact and influence each other, the traditional method of analyzing each energy network separately has been difficult to adapt to the requirements of the existing distribution system.

[0003] In addition, in actual scenarios, the popularity of electric vehicles (EVs), which integrate the three functions of power supply, energy storage equipment and power load, has caused more uncertainty in the source-storage-load characteristics of the power grid. Although the modeling of electric vehicles has been relatively complete, most of the existing optimization models have failed to consider the load brought by electric vehicles. The number of electric vehicles connected to the grid has shown an increasing trend year by year, and optimizing their scheduling is also a topic that needs to be considered in actual scenarios. The document "Using Agent-Based Customer Modeling for the Evaluation of EV Charging Systems" simulates the charging behavior of electric vehicle drivers through an agent-based model (ABM) and evaluates the impact of different policy interventions on sustainable charging and renewable energy integration, but it is not coupled with the distributed energy system scenario.

[0004] Different from small-scale energy systems, safety must be strictly guaranteed for application scenarios that contain residential loads, maritime loads, and other components that are highly relevant to the national economy and people's livelihood. However, existing control measures are slightly insufficient in dealing with emergencies. The document "Research on Smooth Switching of Microgrid Operation StatesBased on Improved Droop Control" proposes a three-level control architecture containing multiple microgrid active distribution networks, and regards the distribution network energy management system (EMS) as the upper dispatch center of the microgrid group to achieve optimal operating costs. However, the required data calculation volume is large and the communication time is also long. Once the EMS platform fails, the entire system may collapse. The document "Research on Smooth Switching of Microgrid Operation StatesBased on Improved Droop Control" proposes a multi-agent hierarchical control architecture for microgrid groups. Based on the collaborative and autonomous characteristics of the agents, regional autonomy between microgrids is achieved, but the coordination optimization mechanism between microgrid groups and the communication topology design of the multi-agent system are not analyzed.

[0005] In the existing technology, the closest implementation scheme usually adopts an optimization strategy based on model predictive control (MPC) or mixed integer linear programming (MILP), combined with intelligent algorithms such as genetic algorithm (GA) or particle swarm optimization (PSO) to achieve the optimal operation of distributed energy systems. The document "Research on Design and Operation Optimization of Multi-energy Complementary Distributed Energy System in Data Center" proposes a multi-energy complementary distributed energy system integrating multiple renewable energy sources and energy storage devices as its energy supply system, establishes an optimization target system including indicators such as life cycle cost, carbon emissions, energy consumption, power grid purchase rate and heat waste rate, formulates two operation strategies considering load characteristics, and uses an improved multi-objective locust optimization algorithm to solve the system capacity configuration. In order to analyze the impact of optimization objectives and operation strategies on system optimization, a variety of optimization design schemes are formed by combining different optimization objectives and operation strategies, and the weights of evaluation indicators are determined by using the hierarchical analysis method and entropy weight method, and the evaluation and ranking are carried out using the multi-criteria compromise solution sorting method.

[0006] Among the existing implementation cases of similar topics, some use microgrid control based on model predictive control (MPC) to predict photovoltaic output and load demand in the next few hours. MPC can adjust the output of diesel generator sets in advance to ensure the balance of electricity supply and demand while reducing fuel consumption and emissions. Most existing studies use intelligent algorithm optimization: combined with GA or PSO, it is used to solve dynamic optimization problems of microgrids, such as quickly responding to changes in grid frequency or handling high penetration of renewable energy access.

[0007] Current digital twin models need to be highly accurate to reflect the dynamics of the real system, but overly complex models may lead to excessive consumption of computing resources and affect real-time decision-making. When considering the intermittent nature of distributed photovoltaic and wind power, accurate weather forecasts and power generation models may require a large amount of computing resources, limiting the practicality of the model in actual operating terminals. Existing microgrid models have small capacity and limited regulation capabilities. When high-penetration distributed devices are decentralized and connected to the distribution network (especially electric vehicles with strong randomness), once large-scale load mutations or line failures occur, the reliability of the microgrid will be greatly reduced. In addition, there is currently a lack of instant decision-making models that can effectively reduce both the economic cost of electricity consumption and carbon emissions. Summary of the invention

[0008] The purpose of the present invention is to provide a distribution network group intelligent collaborative operation terminal that realizes group collaborative operation.

[0009] The purpose of the present invention can be achieved by the following technical solutions:

[0010] A distribution network group intelligent collaborative operation terminal comprises the following steps:

[0011] Modeling module: used for modeling a distribution network, wherein the distribution network includes distributed devices and non-distributed devices;

[0012] Decision tree module: used to construct a decision tree model for the modeled distribution network, perform layer-by-layer regulation according to the decision tree model, obtain the output power of each device in the distribution network, and realize coordinated operation;

[0013] The decision tree model includes load base layer, intermediate decision layer and control center layer:

[0014] Control center layer: As the root node, it is used to build the overall control optimization model, solve the control power ratio of distributed equipment types and the control power ratio of non-distributed equipment types, and input them into the intermediate decision layer;

[0015] Intermediate decision layer: As an intermediate node, it is used to plan and solve decision variables according to the total control proportion of distributed equipment types and the total control proportion of non-distributed equipment types, and consider power balance, and input them into the load base layer;

[0016] Load base layer: as a leaf node, it is used to build an intelligent agent for the distributed equipment and non-distributed equipment according to the load characteristics of the distributed equipment, and the intelligent agent controls the output power of each device according to the allocated decision variables.

[0017] Furthermore, the distributed equipment includes photovoltaic equipment, wind power equipment and electric vehicles, and the non-distributed equipment includes diesel power generation equipment and island power load equipment.

[0018] Furthermore, the distributed devices are constructed as digital twins by combining a mechanism model with a data-driven model, and the non-distributed devices are modeled by a statistical law model.

[0019] Furthermore, the overall control optimization model includes an objective function and constraints. The objective function aims to minimize the operating cost. The expression of the objective function is:

[0020]

[0021] C PV (t) = W PV P PV (t)Δt

[0022] C WT (t) = W WT P WT (t)Δt

[0023] C DG (t) = W DG P DG (t)Δt+μ(t)[μ(t)-μ(t-Δt)]C on +[1-μ(t)][-μ(t)

[0024] +μ(t-Δt)]C off

[0025] C buy (t) = σ(t)W buy (t)P buy (t)Δt

[0026] In the formula, C sys is the total operating cost, C PV (t) is the photovoltaic operation cost at time t, C WT (t) is the wind power operation cost at time t, C DG (t) is the maintenance cost of diesel generator, C prot is the maintenance cost of distributed equipment, C buy (t) is the external electricity purchase cost, W PV is the unit power generation cost of photovoltaic power generation, W WT is the unit power generation cost of wind power, W DG is the unit power generation cost of the diesel generator, W buy (t) is the external electricity purchase price at time t, P PV (t) is the photovoltaic power at time t, P WT (t) is wind power, P DG (t) is the diesel generator power, P buy(t) is the power purchased from the outside, Δt is the decision time interval, generally 15 minutes, μ(t) is the starting state of the diesel generator at time t, C on , C off is the start-stop cost, t-Δt is the previous moment, σ(t) is a 0-1 variable, indicating whether electricity is purchased from the outside;

[0027] The constraints include:

[0028] (1) Power balance constraints

[0029]

[0030] Where P PVi ′ is the power of the i-th photovoltaic intelligent body, P WTi ′ is the power of the ith photovoltaic agent, P EVi ′ is the power of the i-th wind power agent, P DGi is the power of the i-th diesel generator, P buy The power purchased from the outside, P load is the total operating power load, k is the number of PV agents;

[0031] (2) Equipment operation constraints

[0032] min{P item}≤P item ≤max{P item}

[0033] In the formula, the subscript item represents each device, P item is the operating power of the equipment, min{P item}、max{P item} are the upper and lower limits of the equipment's operating power.

[0034] Furthermore, the intermediate decision layer includes two intermediate nodes, namely a distributed intermediate decision platform and a non-distributed intermediate decision platform.

[0035] The distributed intermediate decision platform uses a planning solver to solve the decision variables of the intelligent agent of the distributed device according to the proportion of the control power of the distributed device type and considers the power balance, and transmits the decision variables to the intelligent agent of the distributed device;

[0036] The non-distributed intermediate decision platform uses a planning solver to solve the decision variables of the intelligent body of the non-distributed device according to the control power proportion of the non-distributed device type and takes power balance into consideration, and transmits them to the intelligent body of the non-distributed device. At the same time, an LSTM network is used to predict the output power of each load unit in the non-distributed device in the future.

[0037] Furthermore, the steps of constructing the agent include:

[0038] (1) For the intelligent agents of distributed devices:

[0039] According to the load characteristics of distributed equipment, the data dimension reduction method is used to extract the main components of each distributed equipment;

[0040] Clustering the principal components using a clustering algorithm to group distributed devices with similar principal components into one cluster;

[0041] Build an agent for all distributed devices in the same cluster;

[0042] (2) For non-distributed devices:

[0043] An intelligent agent is constructed for each load unit in the non-distributed equipment.

[0044] Furthermore, the clustering operation expression of the clustering algorithm is:

[0045] f({PV i |0≤i≤n1})={PV′1,PV′2,…,PV′ k},k<<n1

[0046] f({WT i |0≤i≤n2})={WT′1,WT′2,…,WT′ j},j<<n2

[0047] f({EV i |0≤i≤n3})={EV′1,EV′2,…,EV l '},l<<n3

[0048] In the formula, f is the clustering operation, k, j, l are the number of photovoltaic, wind power and electric vehicle agents, PV i is the ith photovoltaic agent, WT i is the i-th wind power intelligent agent, EV i is the i-th electric vehicle agent, n1, n2, n3 are set constants.

[0049] Furthermore, the intelligent agents of the distributed devices include the following:

[0050] Neural network: used to predict the main components and output power of each distributed device in the future;

[0051] Communication module: used for two-way communication with each intelligent agent and information source to achieve joint control;

[0052] 0-1 decision module: used to control the start and stop of each distributed device according to the decision variables, where "0" means stop and "1" means start.

[0053] Furthermore, the agent of the non-distributed device includes:

[0054] Communication module: used for two-way communication with each intelligent agent and information source to achieve joint control;

[0055] 0-1 decision module: used to control the start and stop of each distributed device according to the decision variable, where "0" means stop and "1" means start.

[0056] Furthermore, the data dimension reduction method is principal component analysis.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) The present invention constructs a decision tree model based on the distribution network. The decision tree model is divided into a control center layer, an intermediate decision layer and a load base layer, and performs layer-by-layer regulation. The control center layer serves as a root node to perform regulation planning from a global perspective. The intermediate decision layer serves as an intermediate node to perform planning and solving for distributed device types and non-distributed device types respectively, and outputs decision variables. The load base layer serves as a leaf node, receives the decision variables of the upper layer, and performs intelligent regulation on each device in the distribution network, thereby realizing collective collaborative operation.

[0059] (2) The control center layer of the present invention performs overall planning with the goal of minimizing operating costs, allocates power requirements to the control center layer, and achieves the lowest economic cost on the basis of realizing group intelligent control; the control center layer considers the need for each device to meet power balance, and uses a planning solver to solve the distributed device class and the non-distributed device class separately to obtain the decision variables of each device, so as to achieve real-time integration and real-time decision-making of information resources, and avoid excessive calculation time that reduces the risk resistance of the power grid.

[0060] (3) When constructing an intelligent agent, the load base of the present invention combines principal component analysis with a clustering algorithm to group distributed devices with similar principal components into a cluster to construct an intelligent agent, which can reduce prediction variables without sacrificing prediction accuracy. The intelligent agent can perform future predictions, communicate, and control the start and stop of equipment according to decision variables, thereby realizing intelligent regulation of each device in the base.

[0061] (4) The present invention uses a decision tree model to communicate between layers, realize the sharing of data resources and linkage decision-making, and instantly connect with the outside world, thereby improving robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1This is the distribution network group intelligent collaborative operation terminal architecture of the present invention. DETAILED DESCRIPTION

[0063] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0064] This embodiment provides a distribution network group intelligent collaborative operation terminal, such as Figure 1 As shown, the terminal includes a modeling module and a decision tree model building module.

[0065] Modeling module:

[0066] Used to digitally model the distribution network. The distribution network includes distributed equipment and non-distributed equipment. Distributed equipment includes photovoltaic equipment, wind power equipment, and electric vehicles, and non-distributed equipment includes diesel generators, island power load equipment, etc.

[0067] For distributed power generation equipment, considering that its working environment will not change much in a short period of time, it can be built into a digital twin by combining mechanism modeling and data modeling, so as to build a daily output curve for photovoltaic and wind power based on the existing data set and combined with data enhancement methods. For electric vehicles, the Monte Carlo method is introduced to simulate the behavior of electric vehicle owners to predict the number and capacity of vehicles connected to the grid. For non-distributed equipment, statistical laws are used for modeling.

[0068] Decision tree module:

[0069] This module models the digital distribution network as a decision tree model. The decision tree model has three layers, namely the load base layer, the intermediate decision layer and the control center layer.

[0070] (1) Load base

[0071] In the load base layer, each type of load is abstracted into a micro-agent. Each agent can make decisions with the upper-level nodes and can also transfer information and energy with each other.

[0072] 1) Cluster-type agents

[0073] Based on their distributed characteristics, the principal components of loads such as photovoltaics, wind power and electric vehicles can be extracted through principal component analysis. Then, photovoltaic panels / wind turbines / electric vehicles with similar principal components can be grouped into a cluster through a clustering algorithm. By adjusting the relevant parameters of the clustering, energy distribution and cost calculation can be achieved.

[0074] Through principal component analysis and clustering algorithms, distributed devices can be abstracted into fewer "representative" devices. If the function f represents the clustering operator, then:

[0075] f({PV i |0≤i≤n1})={PV′1,PV′2,…,PV′ k},k<<n1 (1)

[0076] f({WT i |0≤i≤n2})={WT′1,WT′2,…,WT′ j},j<<n2 (2)

[0077] f({EV i |0≤i≤n3})={EV′1,E′2,…,EV l ′},l<<n3 (3)

[0078] Thus, k, j, and l grassroots photovoltaic, wind power, and electric vehicle agents are obtained respectively. Each agent contains the following modules:

[0079] ① Small neural network. Based on the upper-level decision variables and the comparison between historical data and current data, it can instantly predict the value of the principal component and power output of the agent within a period of time.

[0080] ② Communication module: Realize two-way communication and joint control with other intelligent agents and information sources (such as the weather station on the island).

[0081] ③0-1 decision module. Control the start and stop of each distributed device, where "0" means stop and "1" means start. Given that the number of grassroots devices is still large, only retaining 0-1 decisions for a grassroots intelligent agent can minimize the computing load and improve response sensitivity.

[0082] 2) Other Agents

[0083] Diesel generators and other controllable loads on the island that are not easy to obtain uniform characteristics do not need to be clustered when there are not many of them. Each load unit directly participates in the optimization as an intelligent agent. Its components include communication modules and 0-1 decision modules.

[0084] (2) Intermediate decision-making layer

[0085] The intermediate decision-making layer plays the role of autonomous coordination and control, especially when responding to power grid emergencies, it can make decisions immediately while providing feedback to the control center. It is divided into two types: distributed equipment and other equipment. Among them, the intermediate decision-making platform of distributed equipment is responsible for the relevant decisions of photovoltaic, wind power and electric vehicles. It is equipped with a small planning solver and communicates directly with each intelligent agent. The number of decision variables assigned to each intelligent agent does not exceed 3; the intermediate decision-making platform of other equipment is responsible for processing information such as residential loads and diesel generators, and can effectively respond to the actual situation of the power grid by using a small LSTM.

[0086] It can receive dispatching instructions issued by active distribution network-level intelligent agents and determine whether the operation plan will cause the power of lower-level intelligent agents to exceed the limit. If the distributed equipment in a certain energy network exceeds the limit, the operation plan needs to be readjusted (the highest priority optimization goal is changed to avoid exceeding the limit, and then the values ​​of various variables are calculated and updated), and the adjusted results are returned to the control center; the start and stop of loads such as diesel generators can be directly controlled based on the information from the control center at the previous moment and the information from the grassroots load at this moment (sending instructions to the most grassroots equipment to realize electric control), which brings convenience to responding to emergencies and solving linear programming problems in the control center.

[0087] (3) Control center layer

[0088] The control center layer is the root node of the entire decision tree and establishes two-way communication with the two intermediate decision nodes. Due to the characteristics of the intermediate decision layer directly manipulating 0-1 variables, the control center layer needs to consider the cost function. That is, the optimization goal of the control center is cost optimization and power balance. The overall control optimization model is:

[0089] 1) Cost Function

[0090] Only the control center needs to consider the cost function. That is, the optimization goal of the control center is cost optimization and power balance, while the lower-level nodes do not need to consider the cost.

[0091] The multi-energy optimization allocation plan takes the total system operating cost as the optimization target. The total system operating cost includes the unit power generation cost of photovoltaic power, the unit power generation cost of wind power, the maintenance cost of various distributed equipment, and the price of electricity purchased from the external power grid. The specific forms are as follows:

[0092] C sus =min {∑ t (C PV (t)+C WT (t)+C DG (t)+C prot +C buy (t)} (4)

[0093] The specific calculation form of each cost is as follows. For (5) to (7), W represents the unit power generation cost; μ and σ are 0-1 variables, where μ(t) represents the start-up state of the diesel generator at time t, which is used to calculate the start-stop cost; σ represents whether electricity is purchased from the outside; for (8), W is the external grid electricity price at time t.

[0094] C PV (t) = W PV P PV (t)Δt (5)

[0095] C WT (t) = W WT P WT (t)Δt (6)

[0096] C DG (t) = W DG P DG (t)Δt+μ(t)[μ(t)-μ(t-Δt)]C on +[1-μ(t)][-μ(t)+μ(t-Δt)]C off (7)

[0097] C buy (t) = σ(t)W buy (t)P buy (t)Δt (8)

[0098] 2) Constraints

[0099] When the power grid is operating, it is necessary to comprehensively consider constraints such as the power balance of various energy sources and the characteristics of the energy supply units. These constraints can be classified into system operation constraints and operation constraints of each energy supply unit in the system.

[0100] Power balance constraints:

[0101]

[0102] Formula (9) represents the power constraint of the active distribution network during operation. The sum of the power of various intelligent entities and the power purchased from the external power grid should meet the sum of the operating requirements of various loads on the island.

[0103] Equipment operation constraints:

[0104] min{P item}≤P item ≤max{P item} (10)

[0105] Item is a specific device or intelligent entity, which requires its electrical power to be within the rated range.

[0106] When the electricity price of the external power grid is a piecewise linear function, the total cost function can be reduced to a piecewise linear programming problem, which greatly improves the calculation rate and efficiency. At the same time, the control center layer is also allowed to carry software extensions to visualize various information, which is beneficial for users to grasp the overall operation status of the power grid. In addition, the control center is also responsible for recording the operation logs sent by the intermediate decision-making layer. After analyzing and processing the data, it is fed back to the intermediate decision-making layer during the period of low information flow in the decision tree at night, so as to further enhance the correctness of the decision-making of the intermediate decision-making layer.

[0107] In summary, this embodiment constructs a decision tree model, that is, each clustered device constitutes a decision-making base layer - an intermediate decision layer composed of intermediate nodes that control distributed devices and other types of devices respectively - and a top-level control center that regulates these two intermediate nodes. The parameters of each distributed device inside the node are abstracted into the parameters of the cluster center point, which are directly regulated by the upper-level decision nodes; a small neural network or a small planning solver is introduced at each node to achieve instant integration and instant decision-making of information resources, thereby avoiding excessive calculation time that reduces the risk resistance of the power grid.

[0108] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0109] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A distribution network group intelligent collaborative operation terminal, characterized in that: The following steps are involved: Modeling module: used for modeling a distribution network, wherein the distribution network includes distributed devices and non-distributed devices; Decision tree module: used to construct a decision tree model for the modeled distribution network, and perform layer-by-layer regulation according to the decision tree model to achieve coordinated operation; The decision tree model includes load base layer, intermediate decision layer and control center layer: Control center layer: As the root node, it is used to build the overall control optimization model, solve the control power ratio of distributed equipment types and the control power ratio of non-distributed equipment types, and input them into the intermediate decision layer; Intermediate decision layer: As an intermediate node, it is used to plan and solve decision variables according to the control power proportion of distributed equipment types and the control power proportion of non-distributed equipment types, and input them into the load base layer; Load base layer: as a leaf node, it is used to build an intelligent agent for the distributed equipment and non-distributed equipment according to the load characteristics of the distributed equipment, and the intelligent agent controls the output power of each device according to the allocated decision variables.

2. A distribution network group intelligent collaborative operation terminal according to claim 1, characterized in that: The distributed equipment includes photovoltaic equipment, wind power equipment and electric vehicles, and the non-distributed equipment includes diesel power generation equipment and island power load equipment.

3. A distribution network group intelligent collaborative operation terminal according to claim 1, characterized in that: The distributed devices are constructed as digital twins by combining a mechanism model with a data-driven model, and the non-distributed devices are modeled by a statistical law model.

4. A distribution network group intelligent collaborative operation terminal according to claim 1, characterized in that: The overall control optimization model includes an objective function and constraints. The objective function aims to minimize the operating cost. The expression of the objective function is: C PV (t)=W PV P PV (t)Δt C WT (t)=W WT P WT (t)Δt C DG (t)=W DG P DG (t)Δt+μ(t)[μ(t)-μ(t-Δt)]C on +[1-μ(t)][-μ(t) +μ(t-Δt)]C off C buy (t)=σ(t)W buy (t)P buy (t)Δt In the formula, C sys is the total operating cost, C PV (t) is the photovoltaic operation cost at time t, C WT (t) is the wind power operation cost at time t, C DG (t) is the maintenance cost of diesel generator, C prot is the maintenance cost of distributed equipment, C buy (t) is the external electricity purchase cost, W PV is the unit power generation cost of photovoltaic power generation, W WT is the unit power generation cost of wind power, W DG is the unit power generation cost of the diesel generator, W buy (t) is the external electricity purchase price at time t, P PV (t) is the photovoltaic power at time t, P WT (t) is wind power, P DG (t) is the diesel generator power, P buy (t) is the power purchased from the outside, Δt is the decision time interval, μ(t) is the starting state of the diesel generator at time t, C on , C off is the start-stop cost, t-Δt is the previous moment, σ(t) is a 0-1 variable, indicating whether electricity is purchased from the outside; The constraints include: (1) Power balance constraints In the formula, is the power of the ith photovoltaic agent, is the power of the ith photovoltaic agent, is the power of the i-th wind power agent, is the power of the i-th diesel generator, P buy The power purchased from the outside, P load is the total operating power load, k is the number of PV agents; (2) Equipment operation constraints my{P item }≤P item ≤max{P item } In the formula, the subscript item represents each device, P item is the operating power of the equipment, min{P item }、max{P item } are the upper and lower limits of the equipment's operating power.

5. A distribution network group intelligent collaborative operation terminal according to claim 1, characterized in that: The intermediate decision layer includes two intermediate nodes, namely a distributed intermediate decision platform and a non-distributed intermediate decision platform. The distributed intermediate decision platform uses a planning solver to solve the decision variables of the intelligent agent of the distributed device according to the proportion of the control power of the distributed device type and considers the power balance, and transmits the decision variables to the intelligent agent of the distributed device; The non-distributed intermediate decision platform uses a planning solver to solve the decision variables of the intelligent body of the non-distributed device according to the control power proportion of the non-distributed device type and takes power balance into consideration, and transmits them to the intelligent body of the non-distributed device. At the same time, an LSTM network is used to predict the output power of each load unit in the non-distributed device in the future.

6. A distribution network group intelligent collaborative operation terminal according to claim 1, characterized in that: The steps of constructing the agent include: (1) For agents in distributed devices: According to the load characteristics of distributed equipment, the data dimension reduction method is used to extract the main components of each distributed equipment; Clustering the principal components using a clustering algorithm to group distributed devices with similar principal components into one cluster; Build an agent for all distributed devices in the same cluster; (2) For non-distributed devices: An intelligent agent is constructed for each load unit in the non-distributed equipment.

7. A distribution network group intelligent collaborative operation terminal according to claim 6, characterized in that: The clustering operation expression of the clustering algorithm is: f({PV i |0≤i≤n1})={PV′1,PV′2,…,PV′ k },k<<n1 f({WT i |0≤i≤n2})={WT′1,WT′2,…,WT′ j },j<<n2 f({HOME i |0≤i≤n3})={EV1′,EV2′,…,EV l ′},l<<n3 In the formula, f is the clustering operation, k, j, l are the number of photovoltaic, wind power and electric vehicle agents, PV i is the ith photovoltaic agent, WT i is the i-th wind power intelligent agent, EV i is the i-th electric vehicle agent, n1, n2, n3 are set constants.

8. A distribution network group intelligent collaborative operation terminal according to claim 6, characterized in that: The intelligent agents of the distributed devices include the following: Neural network: predict the main components and output power of each distributed device in the future; Communication module: used for two-way communication with each intelligent agent and information source to achieve joint control; 0-1 decision module: used to control the start and stop of each distributed device according to the decision variable, where "0" means stop and "1" means start.

9. A distribution network group intelligent collaborative operation terminal according to claim 6, characterized in that: The intelligent agent of the non-distributed device includes: Communication module: used for two-way communication with each intelligent agent and information source to achieve joint control; 0-1 decision module: used to control the start and stop of each distributed device according to the decision variable, where "0" means stop and "1" means start.

10. A distribution network group intelligent collaborative operation terminal according to claim 6, characterized in that: The data dimension reduction method is principal component analysis.

Images