Variable frequency air conditioner demand response method based on multi-agent system and power distribution network structure

Abstract

The application discloses a variable frequency air conditioner demand response method based on a multi-agent system and a power distribution network structure. The application uses a Monte Carlo method to aggregate variable frequency air conditioner loads to form a load agent to adapt to the requirements of the proposed demand response method; electrical distances are calculated according to a voltage-reactive power sensitivity matrix as a communication weight between agents to make it applicable to the power distribution network structure; and the formed agents are jointly solved for an optimal cost by using a multi-agent system and a consistency algorithm. The application realizes the aggregated utilization of the regulation capacity of the variable frequency air conditioner and evaluates the capability of the load to participate in the demand response of the power system; the application optimizes the cost of the variable frequency air conditioner cluster participating in the demand response; and the application can realize the distributed control of the system to the single variable frequency air conditioner, ensures the speed, and improves the flexibility of the load accessed from different positions of the power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power system demand response, specifically to the use of a multi-agent system suitable for distribution network structures to achieve distributed control of variable frequency air conditioners, thereby reducing the demand response cost of variable frequency air conditioners. Background Technology

[0002] In distributed control systems, the development of multi-agent systems (MAS) and related technologies meets the needs of regulating these power resources. References include [Li Xianglong, Fu Xiao, Zhu Jie, et al. Distribution network operation control strategy based on multi-agent system theory [J]. Journal of Electric Power System and Automation, 2020, 32(9):49-54.] and [Zeng Shuang, Liang Anqi, Wang Liyong, et al. Multi-agent peak shaving and valley filling strategy considering the flexibility of photovoltaic-storage-electric-thermal synergistic system [J]. China Electric Power, 2023, 56(2):133-142.]. Multi-agent systems can divide complex systems into agents that can operate autonomously and communicate with each other. Through coordinated control among agents, they can solve distributed system optimization problems that traditional centralized systems cannot effectively address, thereby improving problem processing speed and system real-time performance.

[0003] Currently, the application of multi-agent related technologies in power systems is mostly concentrated in microgrid energy management. [10-12]Spatial load forecasting, as referenced in [LI J., WANG D., ZHANG S., et al. Spatial load forecasting based on multi-agent system theory[J]. Sci. Technol. Innov. Herald 2018, 1: 15–16.], and source-load state detection, as referenced in [YANG W., ZHANG C., SHEN L., et al. Automatic load state detection technology of intelligent multi-agent client]. In the fields of server[J].Autom.Instrument.2019,4:188–191., load aggregators (LAs) based on the power market and switching costs are often considered as agents. The literature [Cheng Lin, Wan Yuxiang, Zhang Fang, et al. Operation mode and control strategy of air conditioning service based on load aggregator business[J]. Automation of Electric Power System,2018,42(18):8-14.] and the literature [Tang Zao, Liu Junyong, Liu Youbo, et al. Load control and distribution network reconfiguration under the participation of air conditioning aggregators[J]. Automation of Electric Power System,2018,42(2):42-48.] use system power balance, agent adjustable capacity upper and lower limits as constraints to optimize response speed and achieve the purpose of actively supporting the distribution network.

[0004] However, these studies generally face the problems of single load type and weak versatility. Flexible loads are more often used as an auxiliary means to increase the inertia of new energy power generation systems, especially to stabilize the strong fluctuations and anti-peak characteristics of wind power output. In addition, the various types of loads lack the ability to communicate with each other and do not fully utilize the potential of the load side to jointly solve problems such as demand response. It is difficult to handle the output level, priority or trade-off of multiple loads. Reference [XUE Song, ZHANGFan, FANMenghu, et al. Assessment Model of Demand Side Load Resources Promoting Non-hydro Renewables Integration and Consumption[C] / / 2018China International Conference on Electricity Distribution(CICED),2018,Tianjin,China:2880-2883.].

[0005] Meanwhile, in traditional multi-agent systems, the communication weights between agents are limited by the adjacency matrix in graph theory, and are usually only assigned values ​​of 0 or 1 [Wang Beibei, Hu Xiaoqing, Gu Weiyang, et al. Distributed collaborative control strategy for large-scale air conditioning loads participating in peak shaving under hierarchical control architecture [J]. Proceedings of the CSEE, 2019, 39(12):3514-3527.], which cannot reflect the electrical coupling relationship between agents and weakens the applicability of multi-agent systems to power systems. This invention uses relevant power system cluster partitioning indicators to assign values ​​to the communication weights between agents in multi-agent systems to improve the distinguishability of communication between agents. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention discloses a variable frequency air conditioning demand response method based on a multi-agent system and power distribution network structure.

[0007] This invention establishes a variable frequency air conditioning model based on typical power resource air conditioning, and then aggregates it to create a load-side agent that conforms to the characteristics of power resource load curves. This achieves preliminary integration of these resources and preliminary optimization of equipment capacity and load curves. Then, based on a consensus algorithm, it proposes using electrical distance from the cluster partitioning index as the communication weight between agents to establish a multi-agent system, further improving the control speed and utilization rate of the distribution network for loads. Finally, the changes and related characteristics of each agent under demand response scenarios are analyzed, verifying the effectiveness of the multi-agent system technology in demand response.

[0008] The variable frequency air conditioning demand response method based on multi-agent systems and distribution network structures includes the following steps:

[0009] Step 1: Aggregate the air conditioning load output models;

[0010] For the load output model, an improved variable frequency air conditioner load output model based on the on-off air conditioner model is adopted, and the variable frequency air conditioner load aggregated output model is obtained by Monte Carlo method.

[0011] Step 2: Treat the variable frequency air conditioning load aggregation output model as individual agents; improve the communication weights between agents and reassign them with normalized electrical distance to suit the power distribution network structure, and establish a multi-agent system;

[0012] The electrical distance normalized based on sensitivity in the cluster partitioning index is used as the inter-agent communication weight;

[0013] Step 3: By using a multi-agent system to jointly solve the optimal demand response cost among agents using communication weights, a method is proposed for a multi-agent system to participate in demand response by leveraging the distribution network structure.

[0014] Step 4: For distributed control of the agents, the optimized load reduction is allocated to each agent, and the agents control individual devices so that these loads can quickly respond to demand through the multi-agent system. Embedded devices are used to monitor the user load status and generate feedback information.

[0015] Further, the specific method of step 1 is as follows: Use the lumped parameter method model of room heat transfer to establish a single on-off air conditioner model. Starting from the on-off air conditioner model, add the indoor and outdoor temperature difference factor to optimize it into a variable frequency air conditioner load output model. Use the Monte Carlo method to make the air conditioner conform to the average distribution in the initial temperature distribution and the initial set temperature also conform to the average distribution. Expand the base number of air conditioners and aggregate to obtain the aggregated air conditioner load output model.

[0016] The lumped parameter method model for heat transfer in a room is as follows:

[0017]

[0018] θ in (0)=θ in0 ,θ wall (0)=θ wall0

[0019] In the formula: C in C wall These are the equivalent heat capacity of the room air and the equivalent heat capacity of the room walls, respectively; θ in θ wall θ out These are the indoor air temperature, the room wall temperature, and the outdoor temperature, respectively; R in R wall These are the equivalent thermal resistance of the air in the room and the equivalent thermal resistance of the walls in the room, respectively; P ac (t) represents the current operating power of the air conditioner, which has the following states.

[0020]

[0021] In the formula: P Ncool Rated cooling capacity; P Nheat S(t) represents the rated thermal power; S(t) represents the current on / off state of the air conditioner.

[0022] For on / off air conditioners, the following on / off states are available.

[0023] By controlling the switch to change according to the difference between the outdoor temperature and the set temperature, the operating mechanism of the inverter air conditioner in summer is simulated, which can adapt to the situation where the load in the agent includes both inverter air conditioners and on / off air conditioners.

[0024]

[0025] In the formula: T set The set temperature for the air conditioner; T limit This is to improve the temperature control accuracy of the air conditioner.

[0026] It should be noted that the above formula only applies to summer cooling; the operating mechanism of air conditioners differs slightly in winter. The operating mechanism of inverter air conditioners in winter is shown below.

[0027]

[0028] The Monte Carlo method is used to evenly distribute the air conditioning parameters and initial room temperature within a certain range. Combined with the outside temperature variation curve, a daily load curve for aggregated air conditioning load is obtained. When the air conditioning system performs demand response, if peak shaving is required in summer, then T... set Rise, fill the valley then T set Decline, winter T set The change is the opposite. In T set After the corresponding changes occur, the power of the aggregated air conditioning load will also change accordingly.

[0029]

[0030] In the formula: P ac1,k (t) represents the power of the k-th air conditioning unit belonging to agent i before executing the demand response; P ac2,k (t) represents the power of the k-th air conditioning unit of agent i after demand response.

[0031] The obtained P DAC,i (t) is the variable frequency air conditioning load aggregate output model.

[0032] Furthermore, the method for calculating the inter-agent communication weight in step 2 is as follows:

[0033] The communication weights between agents in the MAS are assigned using relevant power system cluster partitioning indices, ensuring that the cluster partitioning indices are strongly correlated only with the network structure. First, the electrical distance between nodes is calculated based on the voltage-reactive power sensitivity matrix, serving as the basis for the inter-agent communication weights.

[0034] The relationships between relevant variables in a power distribution network can be represented by the following matrix:

[0035]

[0036] In the formula: H, N, M, and L are the Jacobian matrices between the corresponding variables.

[0037] Inverting the matrix yields:

[0038]

[0039] In the formula: S VQ This is the voltage-reactive power sensitivity matrix.

[0040] As can be seen from the formula, by injecting a certain amount of reactive power into a node and calculating the change in voltage amplitude at that node, the voltage-reactive power sensitivity matrix can be obtained.

[0041] The electrical distance is calculated using the S-axis of the remaining nodes. VQ The value is related to the node's own S. VQ The ratio of values.

[0042]

[0043] In the formula: d ij ΔV represents the electrical distance between the two nodes; ΔV represents the magnitude of the node voltage change.

[0044] The relationship between two nodes in a network is also affected by other nodes in the network. Considering the influence of all nodes in the network, the electrical distance between node i and node j is:

[0045]

[0046] The communication weights between agents should satisfy the condition that the smaller the electrical distance, the greater the weight. Therefore, the communication weights between agents should be normalized, as shown below.

[0047]

[0048] In the formula: l ij Weights for inter-agent communication.

[0049] Furthermore, the mechanism by which the multi-agent system in step 3 jointly solves for the optimal demand response cost among agents using communication weights is as follows:

[0050] First, the Laplace matrix is ​​calculated using communication weights.

[0051] The process of transferring first-order dynamic information and finding a consistent solution between agents in continuous time can be achieved by iterating the Laplace matrix:

[0052]

[0053]

[0054] In the formula: L ij A represents an element in the Laplace matrix; ij These are the elements in the adjacency matrix, which is the communication matrix between agents, and their weights are the inter-agent communication weights l. ij .

[0055] In practice, due to the influence of sampling and time delay, the first-order information transmission and consensus solution solving process of the system have discrete-time characteristics. Therefore, a consensus solution should be achieved through an iterative row stochastic matrix.

[0056]

[0057]

[0058] In the formula: D ij represents the elements in a row random matrix.

[0059] Furthermore, the demand response problem is solved using the following method.

[0060] The optimization objective is to minimize the total cost of demand response.

[0061]

[0062] In the formula: C represents the demand-side cost of the power company; P Di For the actual load reduction of a single user; F Di Cost per agent.

[0063] The Lagrange relaxation method is used to optimize this objective.

[0064] F D * =F D (P D1 ,P D2 ,…,P DN )-λf(P D1 ,P D2 ,…,P DN )

[0065]

[0066] Then we have:

[0067]

[0068]

[0069] The Lagrange relaxation method shows that the objective has an optimal solution when the first derivative of the cost of each agent with respect to the amount of load reduction required by that agent is consistent. That is, it is possible to use the communication weights between agents to calculate the row stochastic matrix of the multi-agent system, and then use the information of the first derivative of the load reduction with respect to the cost to iterate through the row stochastic matrix to find the consistent solution of this first-order information, that is, the consistent solution of the incremental cost.

[0070] Furthermore, step 4 implements the distributed control method for the agent as follows:

[0071] The incremental cost changes in real time, which corresponds to the load reduction amount that changes in real time. After the multi-agent system solves the demand response problem, it allocates the optimal load reduction amount to each agent and controls individual devices through the agents to generate feedback information, while waiting for new demand response targets from the system.

[0072] The beneficial effects of this invention are as follows:

[0073] This invention proposes to use the normalized electrical distance as the inter-agent communication matrix. Compared with the existing method that only uses the adjacency matrix as the inter-agent communication matrix, this method reflects the electrical coupling relationship between agents, making it applicable to power distribution network structures.

[0074] The demand response method proposed in this invention, compared with the existing method of allocating the load reduction of each agent according to the maximum demand response capacity, effectively reduces the demand response cost while ensuring the solution speed, and achieves the optimal allocation of agent load reduction. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the demand response consistency solution based on a multi-agent system and distribution network structure in this invention (short dashed lines represent the first-order information changes of agent 1, solid lines represent the first-order information changes of agent 2, long dashed lines represent the first-order information changes of agent 3, and dotted lines represent the first-order information changes of agent 4).

[0076] Figure 2 The present invention is based on the load reduction tasks assigned to each agent in a multi-agent system (short dashed line represents the load reduction amount of agent 1, solid line represents the load reduction amount of agent 2, long dashed line represents the load reduction amount of agent 3, and dotted line represents the load reduction amount of agent 4).

[0077] Figure 3 This is a schematic diagram illustrating the change in total load reduction during the demand response process of this invention.

[0078] Figure 4a , 4b Figures 4a, 4b, and 4c show the effect comparison between the proposed demand response method and the traditional mode for each agent of this invention (4a, 4b, 4c, and 4d are the load curves of Agent 1, Agent 2, Agent 3, and Agent 4, respectively, and point 20 in the figure is the start time of demand response). (The solid line is the load curve of the agent when no demand response is performed, the short dashed line is the load change of the agent under the traditional mode, and the long dashed line is the load change of the agent after optimization under the method of this invention). Detailed Implementation

[0079] The present invention will be further described and illustrated below with reference to the accompanying drawings and specific embodiments. In addition to meeting the maximum regulation capacity limit of the agent, the regulation capacity of each agent in the present invention must also meet the constraints of the distribution network lines. Furthermore, the network topology can be combined in various ways.

[0080] This invention is a variable frequency air conditioner demand response method based on a multi-agent system and distribution network structure, which simulates the output of variable frequency air conditioners and realizes load aggregation;

[0081] The method for aggregating variable frequency air conditioning loads includes the following steps:

[0082] Step (1) Based on the Monte Carlo method, set the initial conditions.

[0083] Initial conditions include: ensuring that the room area distribution conforms to the corresponding probability density, that the real-time changes in the outside temperature also conform to the corresponding probability distribution, that the initial temperature is set to be evenly distributed within a certain range, and that the temperature control accuracy is 1 degree Celsius (which can be changed according to actual conditions).

[0084] Step (2) Based on the lumped parameter model of room heat transfer, simulate the output of the variable frequency air conditioning load group, and sum the power of all individual devices under the agent to calculate the total output curve of the aggregated air conditioning load. The lumped parameter model of room heat transfer is as follows:

[0085]

[0086] θ in (0)=θ in0 ,θ wall (0)=θ wall0

[0087] In the formula: C in C wall θ represents the equivalent heat capacity of the air in the room and the equivalent heat capacity of the room walls; in θ wall θ out The indoor air temperature, room wall temperature, and outdoor temperature; R in R wall P represents the equivalent thermal resistance of the air in the room and the equivalent thermal resistance of the room walls. ac (t) represents the current operating power of the air conditioner, which has the following states.

[0088]

[0089] In the formula: T set The set temperature for the air conditioner; T limit This is to improve the temperature control accuracy of the air conditioner.

[0090] The variable frequency air conditioning load aggregation output model is used as a series of agents, and the communication weights between agents are improved, including the following steps:

[0091] Step (1) Calculate the electrical distance between nodes based on the voltage-reactive power sensitivity matrix.

[0092] Normalized inter-agent communication weights include:

[0093] Inject a certain amount of reactive power into a node of the distribution network, calculate the change in voltage amplitude at that node, and calculate the voltage-reactive power sensitivity matrix.

[0094] The electrical distance is calculated using the S-axis of the remaining nodes. VQ The value is related to the node's own S. VQ The ratio of values.

[0095]

[0096] In the formula: d ij ΔV represents the electrical distance between the two nodes; ΔV represents the magnitude of the node voltage change.

[0097] The relationship between two nodes in a network is also affected by other nodes in the network. Considering the influence of all nodes in the network, the electrical distance between node i and node j is:

[0098]

[0099] Step (2) normalizes the electrical distance as the inter-agent communication weight.

[0100] The communication weights between agents should satisfy the condition that the smaller the electrical distance, the greater the weight. Therefore, the communication weights between agents should be normalized, as shown below.

[0101]

[0102] In the formula: l ij This represents the weight for communication between nodes.

[0103] Solving for the optimal demand response cost among agents using a multi-agent system includes the following steps:

[0104] Step (1) Iterate through the information of each agent using the consensus algorithm.

[0105] Obtain the incremental cost information of the agent and calculate the Laplace matrix using the communication weight.

[0106]

[0107]

[0108] In the formula: L ijA represents an element in the Laplace matrix; ij It is an element in the adjacency matrix and also represents the communication weight between agents.

[0109] Calculate the row random matrix based on the Laplace matrix.

[0110]

[0111]

[0112] In the formula: D ij represents the elements in a row random matrix.

[0113] Step (2) Under the constraints of a fixed total number of demand response tasks and the demand response capabilities of the agents, the optimal solution for each agent is obtained.

[0114] The optimization objective is to minimize the total cost of demand response.

[0115]

[0116]

[0117] In the formula: C represents the demand-side cost of the power company; P Di For the actual load reduction of a single user; F Di For the cost of a single agent, P sum P represents the total number of demand response tasks. Dmini and P Dmaxi These correspond to the minimum and maximum demand response capabilities for a single user, respectively.

[0118] The Lagrange relaxation method is used to optimize this objective.

[0119] F D * =F D (P D1 ,P D2 ,…,P DN )-λf(P D1 ,P D2 ,…,P DN )

[0120]

[0121] Then we have:

[0122]

[0123]

[0124] In the formula: λ represents the final consistency information for each agent, i.e., x i, where C is the incremental cost; C is the demand-side cost of the power company; P Di For the actual load reduction of a single user; F Di Cost per agent.

[0125] The Lagrange relaxation method shows that the objective has an optimal solution when the first derivative of the cost of each agent with respect to the amount of load reduction required by that agent is consistent. That is, the row random matrix of the multi-agent system can be calculated using the communication weights between agents, and then the information of the first derivative of the load reduction with respect to the cost can be iteratively applied through the row random matrix.

[0126]

[0127] In the formula, ε is the step size, and ΔP i [k] represents the load variation of a single agent.

[0128] The consistent solution for this first-order information, i.e. the consistent solution for incremental cost, can be obtained.

[0129] Distributed control via an agent includes the following steps:

[0130] Step (1) The system iterates continuously and obtains the final solution. Then, according to the Lagrange method, the change in the obtained consistency information is used as the change in the first derivative information of the agent cost with respect to the power, and the corresponding agent load reduction amount can be derived.

[0131]

[0132]

[0133] In the formula, ΔP isum T represents the load reduction amount allocated to the agent after the iteration, and T is the iteration time. Subsequently, the system can assign tasks with corresponding demand response amounts to each agent, and the agents can control individual devices to complete the execution process.

[0134] Step (2) The agent uses embedded devices to monitor the user's load status, re-collects information about individual devices after the demand response, and waits for the next demand response.

[0135] Example

[0136] The method described in this invention has the following applications in responding to the demand of variable frequency air conditioners.

[0137] A simulation analysis was conducted using an IEEE 14-node power distribution system as an example. Reactive power was injected into each node, and optimal power flow calculations were applied to obtain the electrical distance, which was then normalized as the communication weight between agents. There are four agents: Agent 1 aggregates 1000 variable frequency air conditioners, Agent 2 aggregates 500 variable frequency air conditioners, Agent 3 aggregates 600 variable frequency air conditioners, and Agent 4 aggregates 270 variable frequency air conditioners.

[0138] The air conditioner's rated power is set to 0.1 times the room area. The user's room area conforms to a normal distribution with a mean of 80 and a variance of 10, ranging from 40 to 120 square meters. The outside temperature is simulated using a Beta distribution based on light intensity delayed by 2 hours. The initial temperature is set to 26-27 degrees Celsius, with a temperature control accuracy of 1 degree Celsius. The demand response will begin at 8 PM.

[0139] The initial parameters for the agent are as follows:

[0140] Table 1 Initial parameters of the multi-agent system layer

[0141]

[0142] As can be seen from the table, the maximum total power reduction of the electricity resources aggregated by the various load agents is 2.85MW. Based on the demand response strategy and parameter settings in this paper, assuming the total load reduction is set at 2MW, the system solution process can be obtained as follows: Figure 1 and Figure 2 As shown.

[0143] As can be seen, due to the limitations of each agent's response capability, agents 2 and 4 reached their maximum load reduction capacity first. With continuous iterations in the solution process, agent 3 gradually took on more load reduction, while the load reduction of agents 1 and 2 decreased in the later stages. This trend is largely similar to that of incremental costs, verifying the positive correlation between incremental costs and load reduction. The total load reduction stabilized after 30 iterations, while each agent only reached a stable state after 40 iterations. This suggests that in practical applications of MAS, the system may achieve the load reduction target ahead of schedule when assigning tasks to each agent.

[0144] Figure 3 This is a schematic diagram illustrating the change in total load reduction during the demand response process of this invention.

[0145] After implementing demand response, this application example is compared with the traditional model that uses the maximum demand response volume as a percentage to allocate load reduction to each agent. The results are as follows:

[0146] Table 2 Comparison of Demand Response Effects

[0147] Agent Number Agent1 Agent2 Agent3 Agent4 Traditional demand response (kW) 840 420 500 220 Traditional demand response ratio 0.42 0.21 0.25 0.11 Optimize demand response (kW) 575.6 501.4 577.8 324 Optimize demand response ratio 0.29 0.25 0.29 0.16

[0148] As can be seen, the demand response volume and demand response ratio of each load agent differ between the two modes. The task execution process of the agent layer, i.e., the change of the load curve of the aggregation model, is as follows: Figure 4a , 4b As shown in 4c and 4d.

[0149] This invention evaluates the demand response capability of variable frequency air conditioning load aggregation under the constraints of load aggregation and the maximum demand response of distribution network nodes, obtains the parameters of the multi-agent system with the minimum demand response cost, and provides a feasible solution for the collaborative participation of agents in the system demand response.

[0150] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention. Therefore, all technical solutions obtained through equivalent substitution or transformation fall within the protection scope of the present invention.

Claims

1. A variable frequency air conditioning demand response method based on a multi-agent system and distribution network structure, characterized in that, The steps include the following: Step 1: Aggregate the air conditioning load output models; For the load output model, an improved variable frequency air conditioner load output model based on the on-off air conditioner model is adopted, and the variable frequency air conditioner load aggregated output model is obtained by Monte Carlo method. Step 2: Treat the variable frequency air conditioner load aggregation output model as individual agents; Improve the communication weights between agents and reassign them with normalized electrical distances to suit the power distribution network structure to establish a multi-agent system; The electrical distance normalized based on sensitivity in the cluster partitioning index is used as the inter-agent communication weight; The method for calculating the weight of inter-agent communication is as follows: First, the electrical distance between nodes is calculated based on the voltage-reactive power sensitivity matrix, which serves as the basis for the inter-agent communication weight. The electrical distance is calculated using the distances of the remaining nodes. The value and the node's own The ratio of values; In the formula: The electrical distance between the two nodes; The magnitude of the node voltage change; The relationship between two nodes in a network is also affected by other nodes in the network; considering the influence of all nodes in the network, the electrical distance between node i and node j is: The communication weights between agents should satisfy the condition that the smaller the electrical distance, the greater the weight. Therefore, the communication weights between agents should be normalized, as shown below. In the formula: Weighting for inter-agent communication; Step 3: By using a multi-agent system to jointly solve the optimal demand response cost among agents using communication weights, a method is proposed for a multi-agent system to participate in demand response by leveraging the distribution network structure. Step 4: For distributed control of the agents, the optimized load reduction is allocated to each agent, and the agents control individual devices so that these loads can quickly respond to demand through the multi-agent system. Embedded devices are used to monitor the user load status and generate feedback information.

2. The variable frequency air conditioning demand response method based on a multi-agent system and distribution network structure according to claim 1, characterized in that, The specific method for step 1 is as follows: A single on-off air conditioner model is established using the lumped parameter method model of room heat transfer. Starting from the on-off air conditioner model, the indoor and outdoor temperature difference factor is added to optimize it into a variable frequency air conditioner load output model. The Monte Carlo method is used to make the air conditioner conform to the average distribution in the initial temperature distribution and the initial set temperature also conform to the average distribution. The base number of air conditioners is expanded, and the aggregated air conditioner load output model is obtained. The lumped parameter method model for heat transfer in a room is as follows: In the formula: , These are the equivalent heat capacity of the air in the room and the equivalent heat capacity of the room walls, respectively. , , These are the indoor air temperature, the room wall temperature, and the outdoor temperature, respectively. , These are the equivalent thermal resistance of the air in the room and the equivalent thermal resistance of the walls in the room, respectively. The air conditioner is currently operating at the following power and has the following status: In the formula: Rated cooling capacity; Determine the rated thermal power; This indicates the current on / off status of the air conditioner. For switch-type air conditioners, the following on / off states are available; By controlling the switch to change according to the difference between the outdoor temperature and the set temperature, the operating mechanism of the inverter air conditioner in summer is simulated, which can adapt to the situation where the load in the agent includes both inverter air conditioners and on / off air conditioners. In the formula: Set the temperature for the air conditioner; To improve the temperature control accuracy of the air conditioner; It should be noted that the above formula only applies to summer cooling. The operating mechanism of air conditioners in winter is slightly different from that in summer. The operating mechanism of inverter air conditioners in winter is shown below. The Monte Carlo method is used to evenly distribute the air conditioning parameters and initial room temperature within a certain range. Combined with the outside temperature variation curve, a daily load curve for aggregated air conditioning load is obtained. When the air conditioning system performs demand response, peak shaving is required in summer. Rise, fill the valley Decline, winter The changes are the opposite; in After the corresponding changes occur, the power of the aggregated air conditioning load will also change accordingly; In the formula: The power of the kth air conditioning unit belonging to agent i before performing demand response; The power output after demand response for the kth air conditioning unit of agent i; The results This is the variable frequency air conditioning load aggregation output model.

3. The variable frequency air conditioning demand response method based on a multi-agent system and distribution network structure according to claim 1 or 2, characterized in that, Step 3: The mechanism by which the multi-agent system uses communication weights to jointly solve for the optimal demand response cost among agents is as follows: First, the Laplacian matrix is ​​calculated using the communication weights; The process of transferring first-order dynamic information and finding a consistent solution between agents in continuous time can be achieved by iterating the Laplace matrix: In the formula: These are the elements in the Laplace matrix; These are the elements in the adjacency matrix, which is the communication matrix between agents, and their weights are the communication weights between agents. ; In practice, due to the influence of sampling and time delay, the first-order information transmission and consensus solution solving process of the system have discrete-time characteristics. Therefore, a consensus solution should be achieved through iterative row random matrices. In the formula: represents the elements in a row random matrix.

4. The variable frequency air conditioning demand response method based on a multi-agent system and distribution network structure according to claim 3, characterized in that, The demand response problem can be solved using the following methods; The optimization objective is to minimize the total cost of demand response. In the formula: For power companies' demand-side costs; This represents the actual load reduction for a single user. For the cost of a single agent, The total number of demand response tasks, and These correspond to the minimum and maximum demand response capabilities for a single user, respectively. The Lagrange relaxation method is used to optimize this objective. Then we have: The Lagrange relaxation method shows that the objective has an optimal solution when the first derivative of the cost of each agent with respect to the amount of load that the agent needs to reduce is consistent. That is, it is possible to use the communication weights between agents to calculate the row stochastic matrix of the multi-agent system, and then use the information of the first derivative of the load reduction with respect to the cost to continuously iterate through the row stochastic matrix to find the consistent solution of the first-order information, that is, the consistent solution of the incremental cost.

5. The variable frequency air conditioning demand response method based on a multi-agent system and distribution network structure according to claim 1 or 4, characterized in that, Step 4 implements the distributed control method of the agent as follows: The incremental cost that changes in real time corresponds to the load reduction that changes in real time. After the multi-agent system solves the demand response problem, it allocates the optimal load reduction to each agent and controls individual devices through the agents to generate feedback information, while waiting for the system to have new demand response targets.

Images