Distributed control method, system and equipment for transformer area temperature control load and storage medium

Abstract

The invention provides a distributed control method, system and device for a transformer area temperature control load and a storage medium, and relates to the technical field of power system operation and control. A station area temperature control load group operation scheduling framework is constructed, the indoor temperature change process of a residential user is modeled based on an ETP model, the schedulable capacity range of temperature control loads is constrained, the schedulable capacity range comprises the starting time constraint and the power boundary of the temperature control loads, and the temperature control loads comprise fixed-frequency and variable-frequency air conditioners; based on historical indoor daily temperature curve and daily load curve data of the residential users, temperature change rate, energy consumption preference and optimal temperature interval indexes are extracted, and clustering is carried out through a similarity measurement method of multi-dimensional vector feature fusion; based on the schedulable capability of each temperature control load group, a distributed real-time optimization scheduling model is constructed, a scheduling strategy is solved in combination with a multi-agent depth deterministic strategy gradient algorithm, and real-time accurate regulation and control of the transformer area temperature control load are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system operation and control, and particularly relates to a distributed control method, system and device for a temperature-controlled load in a transformer area and a storage medium. BACKGROUND

[0002] Under the background of current energy structure transformation, the construction of new power system with new energy as the main body is gradually promoted, which makes the proportion of renewable energy generation in the system continue to rise. The inherent volatility and uncertainty of renewable energy generation represented by wind and light generation bring great challenges to the stability and reliability of power grid operation. Traditional thermal power units can adjust their power generation to cope with system load fluctuations, while renewable energy generation has weak controllability, and it is difficult to achieve supply and demand balance regulation of new power system only by relying on the generation side. The existing technology has excavated more adjustable potential of flexible load from the demand side. As a typical user-side flexible load, the temperature-controlled load (TCL) has the characteristics of large scale, wide distribution and high density, and its power demand has obvious time characteristics and seasonal characteristics. In the winter and summer seasons, the use frequency and power demand of the temperature-controlled load will increase sharply, which is easy to cause the steep increase of power grid load and the superposition of power consumption peak, and causes impact on the safe operation of power grid.

[0003] At present, the scheduling method for the temperature-controlled load in the transformer area mainly adopts a centralized optimization strategy, that is, the dispatching center collects a large amount of information such as power grid operation parameters and user power consumption data in the day-ahead or real-time stage, uniformly calculates and issues control instructions, which has the following technical problems: 1. High communication cost and poor real-time performance: centralized scheduling needs to frequently transmit a large amount of data, which causes heavy load on the communication link, increases the communication cost, and affects the real-time performance of control due to information transmission delay. 2. High optimization dimension and large calculation complexity: if a distributed scheduling strategy is adopted, due to the significant difference in power consumption behavior of residential users, it is necessary to develop control strategies for individual households, which leads to exponential growth of optimization variable dimension, easily causes "dimension disaster" problem, makes the algorithm fall into local optimum, and is difficult to globally converge. 3. It is difficult to balance scheduling accuracy and efficiency: the existing method often sacrifices scheduling accuracy when reducing calculation complexity, and faces the problem of low calculation efficiency when pursuing high accuracy, which makes it difficult to realize fast and accurate load control.

[0004] In view of the above problems, existing research attempts to classify user load through clustering methods, such as group division based on power consumption habits, equipment parameters or temperature comfort requirements, so as to reduce the dimension of optimization problem. However, the existing clustering methods mostly rely on a single feature (such as load curve), and fail to fully consider the physical characteristics (such as temperature dynamic change) of temperature-controlled load and user individualized demand, which leads to poor clustering effect and affects the accuracy of subsequent scheduling strategy.

[0005] Therefore, there is an urgent need for a distribution control method for temperature-controlled loads in a transformer area that can balance communication efficiency, computational complexity and scheduling accuracy, to solve problems such as heavy communication pressure in centralized scheduling and difficulty in distributed scheduling optimization, and to realize fast and accurate regulation of temperature-controlled loads, thereby assisting the safe and stable operation of new power systems. SUMMARY

[0006] The main purpose of the present application is to provide a distributed control method, system, device and storage medium for temperature-controlled loads in a transformer area, to solve the above technical problems.

[0007] To solve the above technical problems, the technical solution adopted by the present application is: a distributed control method for temperature-controlled loads in a transformer area, comprising the following steps: S1: Constructing a transformer area temperature-controlled load group operation scheduling architecture, analyzing scheduling information, control signals and power flow on the grid side, aggregator side and user side; S2: Modeling the indoor temperature variation process of residential users based on an equivalent thermal parameter model, and constraining the schedulable ability range of temperature-controlled loads, including the opening time constraint and power boundary of temperature-controlled loads, the temperature-controlled loads including fixed-frequency and variable-frequency air conditioners; S3: Based on the historical indoor daily temperature curve and daily load curve data of residential users, extracting temperature change rate, energy preference and optimal temperature interval indicators, clustering through a similarity measurement method of multi-dimensional vector feature fusion, and conservatively processing the state boundary of temperature-controlled loads in the cluster; S4: Based on the schedulable ability of each temperature-controlled load group in S2, constructing a distributed real-time optimization scheduling model: maximizing the response to transformer area photovoltaic and scheduling center task from a global perspective, minimizing the temperature-controlled load electricity cost from an individual perspective, and embedding the optimization model into a multi-agent deep deterministic policy gradient algorithm framework to optimize and solve the scheduling strategy, realizing real-time and accurate regulation of temperature-controlled loads in the transformer area.

[0008] In a preferred embodiment, S1 constructing the operation scheduling architecture specifically comprises: S11: Clearly defining the power flow: the user side perceives the load state in real time through the temperature-controlled load management system, and interacts with the aggregator; S12: Determining the interaction mode of each subject scheduling information: the aggregator side receives the grid scheduling task and decomposes it to the user side, and the user side feeds back the scheduling result; S13: The user side installs photovoltaic in the transformer area, and the power supply is jointly borne by the grid side transformer and the transformer photovoltaic, forming a stable power transmission mode; S14: Based on the temperature-controlled load management system, the information of each user temperature-controlled load in the building is sensed and control commands are issued, and signal interaction with the aggregator is realized.

[0009] In a preferred embodiment, the modeling in S2 by the equivalent thermal parameter model specifically comprises: S21: analyzing the room temperature variation process by a second-order equivalent thermal parameter model, and the model expression is: ; In the formula, , , are the indoor gas temperature, solid temperature and outdoor air temperature at time t respectively; , are the equivalent heat capacities representing the heat storage performance of the indoor solid and gas; , are the equivalent thermal resistances representing the heat exchange rates between the indoor gas and the outdoor gas, and between the indoor solid and the indoor gas; is the working state of the temperature-controlled load; is the working power of the temperature-controlled load; S22: establishing the dispatchable capacity boundary of the temperature-controlled load, solving the thermodynamic equation in the dispatching period, ignoring the solid temperature change, approximating to a first-order thermodynamic equation and discretizing the equation: ; S23: the dispatchable capacity range constraints of the temperature-controlled load include: For a fixed-frequency air conditioner, the opening duration constraint expression is: ; ; ; Among them, is the minimum start-up duration, is the maximum start-up duration, and , are the upper and lower limits of the set optimal temperature interval; is the natural increase variation of the room temperature; is the natural decrease variation of the room temperature; For a variable-frequency air conditioner, the temperature-controlled load power can be regarded as continuous and adjustable, and the load power constraint is: ; Among them, is the lower limit of the dispatchable power, and the upper limit of the dispatchable power is is: .

[0010] In a preferred embodiment, S3 specifically comprises: S31: Obtain the indoor temperature time series of all residential users in the transformer area and the power time series of the temperature control load, extract the sub-feature temperature change rate curve and the daily load curve, and normalize the data of the indoor daily temperature curve, the daily load curve and the sub-feature temperature change rate curve to convert them into a multi-dimensional vector, the expression is: ; wherein is the indoor gas temperature of the temperature control load at the moment, is the indoor temperature change rate of the temperature control load at the moment, is the power of the temperature control load at the moment, , so a single trajectory composed of n vector points in the data set is expressed as , ; S32: Use the bidirectional Hausdorff distance to measure the similarity of the multi-dimensional feature total trajectory curve set of the temperature control load, the expression is: ; wherein represents the Euclidean distance between two points, and the calculation formula is: ; , then the similarity between any two trajectories is: ; S33: Use the AP clustering algorithm to aggregate the temperature control load of residential users, update the attraction degree matrix and the belonging degree matrix through iteration until convergence, and output the clustering result, the formula is: ; wherein is the degree to which the temperature control load j is suitable as the clustering center of the temperature control load i , indicating the attraction degree from the temperature control load j to the temperature control load i ; i is the degree to which the temperature control load j is suitable as the clustering center of the temperature control load i , indicating the belonging degree from j to ; is the value in the i row j column of the similarity matrix ​​​​​​​​Update the values ​​in the attraction and belonging matrices: ; Output clustering results , The cluster is recorded in the middle. k Medium temperature control load number.

[0011] In the preferred scheme, the clustering results are further processed: Based on the clustering results, all temperature-controlled loads were divided into y A cluster, for the fixed-frequency air conditioners in the cluster, the cluster k The minimum opening time is The maximum open time is For clusters k Medium-frequency inverter air conditioners, the lower limit of their adjustable power remains at [value missing]. The maximum schedulable power is ; Upper and lower limits of the optimal temperature range within the same cluster , The temperature control load corresponding to the cluster center shall be used as the standard.

[0012] In the preferred embodiment, the objective function of the distributed real-time optimization scheduling model of S4 includes: establishing a mathematical model for optimizing the scheduling of temperature-controlled loads in the distribution area. The objective function is divided into two layers: the upper layer objective is to maximize the response of photovoltaic and scheduling tasks in the distribution area, and the lower layer objective is to minimize the electricity cost of the cluster. Specifically: The upper layer coordinates the consumption of photovoltaic power in all temperature-controlled areas and responds to dispatch tasks issued by the dispatch center. The expression is: ; in The maximum number of scheduling attempts, This is the penalty coefficient for failing to meet the scheduling target. To maximize photovoltaic output at all times, The task load is distributed to the dispatch center at each time point. The duration of participation in scheduling is expressed as follows: ; The lower-level target expression is: ; in This refers to the real-time electricity price.

[0013] In the preferred embodiment, the framework of the multi-agent deep deterministic policy gradient algorithm includes: embedding an optimization model within the multi-agent deep deterministic policy gradient algorithm framework, establishing a state space, observation space, action space, and reward function, and setting each cluster k as an agent, wherein: The state space is: ; The observation state space is: ; The motion space is: ; The reward function for a single agent is: ; in These are the weight coefficients of the two-layer objective function; Set a maximum number of training rounds, train the optimized scheduling model based on the algorithm framework until convergence, obtain the optimized scheduling strategy for the temperature-controlled load of each cluster under the transformer area, and deploy the trained scheduling strategy to each temperature-controlled load management system for real-time scheduling and control.

[0014] A distributed control system for temperature-controlled loads in a distribution area includes: The scheduling architecture module is used to construct the operation scheduling architecture of the temperature-controlled load group in the distribution area, and to analyze the scheduling information, control signals and power flow of the grid side, aggregator side and user side; The ETP model module is used to model the indoor temperature change process of residential users based on the equivalent thermal parameter model and constrain the schedulable capacity range of temperature-controlled loads, including the start-up time constraints and power boundaries of the temperature-controlled loads, which include fixed-frequency and variable-frequency air conditioners. The multidimensional vector feature module is used to extract temperature change rate, energy preference and optimal temperature range indicators based on the historical indoor daily temperature curve and daily load curve data of residential users. It performs clustering through a similarity measurement method of multidimensional vector feature fusion and conservatively processes the state boundary of temperature-controlled loads within the cluster. The scheduling module is used to construct a distributed real-time optimization scheduling model based on the schedulable capabilities of each temperature-controlled load group in the ETP model module. From an overall perspective, it maximizes the response of photovoltaic power stations and tasks issued by the dispatch center, and from an individual perspective, it minimizes the electricity cost of temperature-controlled loads. The optimization model is embedded into a multi-agent deep deterministic policy gradient algorithm framework to optimize the scheduling strategy and achieve real-time and accurate control of temperature-controlled loads in the power station area.

[0015] An electronic device, comprising a memory and a processor; The memory is used to store computer programs; The processor is used to implement the distributed control method for temperature-controlled loads in a distribution area when executing the computer program.

[0016] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the distributed control method for temperature-controlled loads in a distribution area.

[0017] This invention provides a distributed control method for temperature-controlled loads in a distribution area. It constructs an operational scheduling architecture for temperature-controlled load groups in the distribution area, analyzes scheduling information, control signals, and power flow from the grid side, aggregator side, and user side. Based on the ETP model, it models the indoor temperature change process of residential users and constrains the dispatchable capacity range of temperature-controlled loads, including start-up time constraints and power boundaries. The temperature-controlled loads include fixed-frequency and variable-frequency air conditioners. Based on historical daily indoor temperature and load curves of residential users, it extracts indicators such as temperature change rate, energy preference, and optimal temperature range, and performs clustering using a similarity measurement method based on multi-dimensional vector feature fusion. Based on the dispatchable capacity of each temperature-controlled load group, it constructs a distributed real-time optimization scheduling model, and combines a multi-agent deep deterministic policy gradient algorithm to solve the scheduling strategy, achieving real-time and precise control of temperature-controlled loads in the distribution area. This method reduces communication costs, improves scheduling real-time performance, reduces and optimizes variable dimensions, unifies the dispatchable capacity processing of temperature-controlled loads within the same cluster, simplifies the optimization model while ensuring control safety, and improves computational efficiency. It also enhances the adaptability and robustness of temperature-controlled load control. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the distributed control method of the present invention; Figure 2 This is a schematic diagram of the single-unit temperature-controlled load group scheduling operation of the present invention; Figure 3 This is a detailed overall flowchart of the fast distributed control method of the present invention; Figure 4 This is a schematic diagram of the multidimensional vector feature fusion measurement method of the present invention. Detailed Implementation

[0019] Example 1 like Figures 1-4 As shown, a distributed control method for temperature-controlled loads in a distribution area includes the following steps: S1: To clarify the transmission paths of scheduling information and control signals and the direction of power flow among the grid side, aggregator side and user side, and to ensure effective scheduling and control of temperature-controlled loads, a scheduling architecture for temperature-controlled load groups in the distribution area is constructed, and the scheduling information, control signals and direction of power flow among the grid side, aggregator side and user side are analyzed.

[0020] S2: The indoor temperature change process of residential users is modeled based on the Equivalent Thermal Parameters (ETP) model, and the schedulable capacity range of temperature-controlled loads is constrained, including the start-up time constraint and power boundary of the temperature-controlled loads, which include fixed-frequency and variable-frequency air conditioners.

[0021] S3: Based on the resident user historical indoor daily temperature curve and daily load curve data, the temperature change rate, energy preference and optimal temperature interval index are extracted, clustering is carried out through the similarity measurement method of multi-dimensional vector feature fusion, and the state boundary of the temperature control load in the cluster is conservatively processed.

[0022] S4: Based on the schedulable ability of each temperature control load group in S2, a distributed real-time optimization scheduling model is constructed: from the overall point of view, the maximum response to the photovoltaic of the substation and the task issued by the scheduling center, from the individual point of view, the minimum temperature control load power cost, and the optimization model is embedded into the multi-agent deep deterministic policy gradient algorithm framework to optimize and solve the scheduling strategy, and the real-time accurate regulation and control of the temperature control load in the substation is realized.

[0023] The embodiment establishes a real-time distributed scheduling architecture of the temperature control load group in the substation considering the terminal resident user temperature control load, temperature control load aggregator and other subjects, analyzes the similarity of the indoor temperature change rate and other indicators based on the historical power consumption data of the resident user, converts to multi-dimensional vector space clustering, and conservatively processes the state boundary of the temperature control load, thereby reducing the variable dimension of the scheduling problem, reducing the excessive dependence on information in the traditional centralized control process, and realizing the real-time accurate and rapid control of the temperature control load in the substation.

[0024] The application realizes high-precision, low-delay and low-cost optimization regulation and control of the temperature control load through the innovative architecture of "clustering dimension reduction + distributed optimization + reinforcement learning".

[0025] The embodiment takes the minimization of system operation cost as the goal, constructs a distributed real-time optimization scheduling model according to the schedulable ability of each temperature control load group, considers the two goals of maximizing the response to the photovoltaic of the substation and the task issued by the scheduling center, and minimizing the power cost of the temperature control load, embeds the optimization model into the multi-agent deep deterministic policy gradient algorithm framework to solve the scheduling strategy, so as to realize the real-time and accuracy of the generation of the temperature control load scheduling strategy in the substation through the centralized training and distributed execution of the scheduling strategy.

[0026] In the preferred scheme, step S1 of constructing the operation scheduling architecture specifically includes: S11: Clearly analyze the power flow: install photovoltaic in the substation on the user side, and through the temperature control load management system, perform state sensing and control command issuing such as load start-stop and power adjustment on the real-time switch state, running power, indoor temperature and other information of each user temperature control load in the building, and transmit signals to the temperature control load aggregator. The power supply of each user load in the building is jointly borne by the transformer on the grid side and the photovoltaic in the substation, forming a stable power supply mode to ensure stable transmission of the power required for the operation of the temperature control load.

[0027] S12: Determine the information exchange mode of each entity's dispatch: Analyze the information exchange mode of each entity's dispatch. On the aggregator side, the temperature-controlled load aggregator receives the dispatch tasks issued by the grid-side dispatch center, decomposes the dispatch tasks accordingly, formulates dispatch strategies for the temperature-controlled loads in the area, and sends the information to the user side; the user side then feeds back the status information and dispatch results of the temperature-controlled loads to the aggregator, realizing two-way information flow.

[0028] S13: On the user side, photovoltaic systems are installed on a per-transformer-area basis. The power supply is shared by the grid-side transformer and the photovoltaic system in the transformer area, forming a stable power transmission mode.

[0029] S14: Relying on the localized deployment of the temperature control load management system and reinforcement learning algorithm, the system can sense the real-time on / off status, operating power, indoor temperature and other information of the temperature control load of each user in the building, and issue control commands such as load start / stop and power adjustment. At the same time, it can achieve signal interaction with the temperature control load aggregator. Each user installs a smart meter to ensure real-time information collection and automatic control.

[0030] Figure 1 shows the scheduling and operation architecture of a temperature-controlled load group within a single unit area.

[0031] This paper employs an equivalent thermal parameter (ETP) model to analyze and model the temperature change process in a room, thus providing a physical basis for analyzing the schedulable capability of the temperature control load model. This is due to the continuously increasing demands on the accuracy and frequency of temperature control load regulation.

[0032] In the preferred embodiment, step S2, modeling using an equivalent thermal parameter model, specifically includes: S21: The second-order ETP model taking into account changes in indoor temperature and solid temperature is as follows: ; In the formula , , The indoor gas temperature, solid temperature, and outside air temperature are respectively represented by time t. , The equivalent heat capacity is used to characterize the thermal storage performance of solid and gaseous materials in indoor environments. , The equivalent thermal resistance is used to characterize the heat transfer rate between indoor and outdoor gases, and between indoor solids and indoor gases. This refers to the operating status of the temperature-controlled load; This refers to the operating power of the temperature-controlled load.

[0033] For inverter air conditioners, It is always 1, only when Adjustments are made to the above; for fixed-frequency air conditioners, their... For constant, only to adjust, the air conditioner is 0 when it is off and 1 when it is on.

[0034] This embodiment establishes the boundary of schedulable capacity for a single refrigeration load in a temperature-controlled load (for example, an air conditioner). In a scheduling period, the maximum upward adjustment power of the temperature-controlled load is determined by the minimum power value required to reduce the current ambient temperature to the set lower threshold, and the maximum downward adjustment power is determined by the maximum power value required to raise the current temperature to the set upper threshold. Therefore, the theoretical lower bound and the theoretical upper bound of the real-time power of the operation of the temperature-controlled load of each resident user are related to the power of the load and the upper and lower thresholds of the temperature, so the schedulable capacity in this period is obtained by solving the thermodynamic equation, and the scheduling period is set to 5-10 minutes.

[0035] S22: Establish the schedulable capacity boundary of the temperature-controlled load, solve the thermodynamic equation in the scheduling period, ignore the temperature change of the solid, approximate it as a first-order thermodynamic equation, and discretize it: .

[0036] S23: The schedulable capacity range constraints of the temperature-controlled load include: For a fixed-frequency air conditioner, the expression of its on duration constraint is: ; ; ; wherein, is the minimum start-up duration, is the maximum start-up duration, and , is the set optimal temperature interval upper and lower limits; is the natural increase of room temperature; is the natural decrease of room temperature.

[0037] For a variable-frequency air conditioner, the temperature-controlled load power can be considered as continuous, and the load power constraint is: .

[0038] wherein, the schedulable power lower limit is , the schedulable power upper limit is : .

[0039] This embodiment accurately calculates the schedulable boundary of the temperature-controlled load through the equivalent thermal parameter (ETP) model, ensures that the indoor temperature is always in the comfort interval during the regulation process, avoids the risk of over-regulation or under-regulation, and improves the safety.

[0040] In actual operation scenarios, the above theoretical boundary is taken as a constraint condition of the aggregation optimization problem, and the indoor temperature is reasonably regulated and controlled in the load distribution process to maintain it within the safe operation interval, so as to effectively avoid the risk of exceeding the limit caused by improper power regulation.

[0041] In the preferred scheme, step S3 specifically comprises: S31: Obtain the indoor temperature time sequence of all resident users in the transformer area and the power time sequence of the temperature control load, extract the sub-feature temperature change rate curve and the daily load curve, and normalize the data of the indoor daily temperature curve, the daily load curve and the sub-feature temperature change rate curve to convert them into a vector in a multi-dimensional space, and the expression is: ; wherein is the indoor gas temperature of the temperature control load at the moment, is the indoor temperature change rate of the temperature control load at the moment, is the power of the temperature control load at the moment, and thus a single trajectory composed of vector points in the data set is expressed as , and the total trajectory curve set composed of the resident temperature control loads is . S32: Similarity of the multi-dimensional feature total trajectory curve set of the temperature control load n is measured by using the bidirectional Hausdorff distance, and the Hausdorff distance is to find the maximum value of each vector point in to each vector point in , and the expression is: ;

[0042] wherein represents the Euclidean distance between two points, and the calculation formula is: ; and the similarity between any two trajectories is: . . .

[0043] ​​​​​The embodiment adopts an AP (Affinity Propagation) clustering algorithm to aggregate the temperature control load of the resident user. The attraction matrix R in the AP algorithm represents the tendency of the temperature control load to select other temperature control loads as clustering centers, and the belonging matrix A represents the tendency of being selected as a clustering center by other temperature control loads. The clustering method is not sensitive to the initial point and does not need to set the number of clustering clusters, and is updated by iteration until convergence. First, the similarity matrix S is calculated as input The damping factor is set The maximum number of iterations is set to control the convergence speed and stability of the control algorithm, the attraction matrix R and the belonging matrix A are initialized as zero matrices, and then the attraction matrix R and the belonging matrix A are updated.

[0044] S33: The AP clustering algorithm is adopted to aggregate the temperature control load of the resident user. The attraction matrix and the belonging matrix are updated by iteration until convergence, and the clustering result is output. The formula is: ; Wherein is the temperature control load j is the temperature control load i The appropriate degree of the clustering center, indicating the attraction degree from the temperature control load j to the temperature control load i ; is the temperature control load i selects the temperature control load j as the clustering center, indicating the belonging degree from i to j ; is the value of the similarity matrix in the row i column. j

[0045] The values in the attraction matrix and the belonging matrix are updated: ; The clustering result is output, in which the temperature control load number in the cluster is recorded. k

[0046] S34: Obtain the clustering result for further processing: Based on the clustering result, all temperature control loads are divided into y clusters. For the fixed-frequency air conditioner in the cluster, the minimum opening duration in the cluster k is , and the maximum opening duration is ; for the variable-frequency air conditioner in the cluster k , the lower limit of the schedulable power is still ​​The upper limit of the schedulable power is ; upper and lower limits of the optimal temperature interval in the same cluster , The clustering center corresponds to the temperature control load as the standard, and the uniformity is maintained.

[0047] As shown in Figure 4 , it is a specific method of the multi-dimensional vector feature fusion measurement method in step S3 of the embodiment. Taking the daily load curve and indoor temperature curve of 5 household users as an example, the indoor gas temperature and indoor temperature change rate at each time can be extracted in the indoor temperature curve, and the power of the temperature control load at the same time can be extracted in the daily load curve. The vector composed of the three index normalized values is mapped into a three-dimensional space, that is, the vector of user 4 at each time can be connected into a curve in the three-dimensional space, and thus the trajectory curve of each user in the three-dimensional space can be obtained as shown on the right of the figure. Based on this, the similarity between users can be obtained by calculating the Hausdorff distance between the curves in S32, and further clustering can be performed.

[0048] In a preferred scheme, the objective function of the distributed real-time optimization scheduling model in step S4 includes: establishing a mathematical model for optimizing the scheduling of the temperature control load in the transformer area, and the objective function is divided into two layers, the upper layer aims to maximize the response to the transformer area photovoltaic and scheduling task, and the lower layer aims to minimize the cluster electricity cost, specifically: The upper layer is to coordinate the consumption of the transformer area photovoltaic by all transformer area temperature control loads, and to respond to the scheduling task issued by the scheduling center, and the expression is: ; Among them is the maximum scheduling times, is the penalty coefficient of the unscheduled scheduling target, is the maximum photovoltaic output at each time, is the task amount issued by the scheduling center at each time, is the participation scheduling duration, and the expression is: ; The lower layer target expression is: ; Among them is the real-time electricity price.

[0049] In a preferred scheme, the framework of the multi-agent deep deterministic policy gradient algorithm in step S4 includes: embedding the optimization model in the multi-agent deep deterministic policy gradient algorithm framework, establishing a state space, an observation space, an action space, and a reward function, and setting each cluster k as an agent, wherein: The state space is: .

[0050] The observation state space is: .

[0051] The action space is: .

[0052] The reward function of a single agent is: .

[0053] wherein is the weight coefficient of the double-layer objective function.

[0054] The maximum training round is set, and the optimization scheduling model is trained based on the algorithm framework until convergence, to obtain the optimization scheduling strategy of the temperature control load of each cluster under the transformer area. The trained scheduling strategy is deployed to each temperature control load management system for real-time scheduling control.

[0055] The step S4 of the embodiment maximizes the upper-layer target in response to the transformer area photovoltaic output and the power grid scheduling task, promotes renewable energy consumption; the lower-layer target realizes the minimization of user electricity cost, and improves the economy; the multi-agent reinforcement learning is used for optimization, that is, the deep deterministic policy gradient (MADDPG) algorithm is used, the distributed agents are trained in cooperation, the global optimization is ensured, the rapid strategy solving is realized, the problem that the traditional optimization algorithm is easy to fall into local optimization is avoided, and therefore the balance between the scheduling precision and the calculation efficiency is improved.

[0056] The specific process of the method is shown in the following specific examples: Figure 3 . Step 1: The running parameters, related equipment parameters, rated power, and historical indoor temperature and load data of each resident user temperature control load in the transformer area are read in and converted into original sample sets for the algorithm to read.

[0057] Step 2: The scheduling period and scheduling step are set.

[0058] Step 3: The schedulable ability range of each temperature control load is calculated and .

[0059] Step 4: The indoor temperature change rate is extracted from the historical data of the resident user temperature control load, and all data of the indoor daily temperature curve, daily load curve, and sub-feature temperature change rate curve are normalized.

[0060] Step 5: The normalized data set is cleaned and arranged, and each data point is combined into a vector form, thereby obtaining a total curve set in a multi-dimensional space, and each curve corresponds to a resident user .

[0061] Step 6 Calculate the Hausdorff distance between each curve in the total curve set , and form a similarity matrix .

[0062] Step 7 Use the AP clustering method to aggregate all resident temperature control loads based on the similarity matrix, and obtain the clustering result of each cluster containing resident user temperature control load number .

[0063] Step 8 Based on the clustering result, the schedulable capacity of each cluster temperature control load is conservatively unified.

[0064] Step 9 Initialize the reinforcement learning framework, specify the maximum number of steps, the maximum number of training rounds, and define the environment, including y an agent, photovoltaic power generation curve, dispatch center task data, etc.

[0065] Step 10 For each agent , randomly initialize the Actor network parameters , randomly initialize the Critic network parameters , copy the target network , , create a shared experience replay pool ReplayBuffer.

[0066] Step 11 Start training: initialize the environment, get the initial state space , enter the time step loop ( t =1… T ).

[0067] Step 12 For each agent i , according to its own observation space: .

[0068] Call the Actor network to generate actions: .

[0069] Step 13 Pass the actions of each agent into the environment, calculate the reward , and store a single training data into the experience replay pool, and update the state and observation space.

[0070] Step 14 If the step number meets the update condition and the replay pool sample is sufficient, perform the following operations: randomly sample a batch from the experience replay pool; for each agent , use the target network to calculate the next action to update the Critic network, maximize the Critic's Q value for its own action, and update update the Actor network; finally, perform soft update of the target network: .

[0071] Step 15 If the maximum number of steps is reached, end the current training round, start the next training round until the maximum number of training rounds is reached, and output the scheduling strategy and reinforcement learning network that maximize the reward; otherwise, go back to step 12 to continue.

[0072] Embodiment 2 Further illustrated in combination with Embodiment 1, a distributed control system for a transformer area temperature control load is provided, comprising: A scheduling architecture module is configured to construct a transformer area temperature control load group operation scheduling architecture, and analyze scheduling information, control signals and power flow directions on the power grid side, the aggregator side and the user side.

[0073] An ETP model module is configured to model the indoor temperature variation process of a residential user based on an ETP model, and constrain the schedulable ability range of the temperature control load, including the start time constraint and power boundary of the temperature control load, wherein the temperature control load includes fixed-frequency and variable-frequency air conditioners.

[0074] A multi-dimensional vector feature module is configured to extract temperature change rate, energy use preference and optimal temperature interval indicators based on historical indoor daily temperature curve and daily load curve data of a residential user, perform clustering through a similarity measurement method of multi-dimensional vector feature fusion, and perform conservative processing on the state boundary of the temperature control load in the cluster. A solution scheduling module is configured to construct a distributed real-time optimization scheduling model based on the schedulable ability of each temperature control load group in the ETP model module: maximize the response to transformer area photovoltaic and scheduling center issued tasks from a global perspective, minimize the temperature control load electricity cost from an individual perspective, and embed the optimization model into a multi-agent deep deterministic policy gradient algorithm framework to optimize and solve the scheduling strategy, thereby realizing real-time and accurate regulation and control of the transformer area temperature control load.

[0075] An electronic device comprising a memory and a processor; The memory is configured to store a computer program. The processor is configured to implement the distributed control method for a transformer area temperature control load in Embodiment 1 when executing the computer program.

[0076] A computer readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, a distributed control method for a transformer area temperature control load in Embodiment 1 is implemented.

[0077] The above embodiments are only preferred technical solutions of the present application, and should not be regarded as a limitation of the present application. The protection scope of the present application should be the technical solutions recited in the claims, including equivalent replacement solutions of the technical features recited in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present application.

Claims

1. A distributed control method for a transformer area temperature-controlled load, characterized by, The method comprises the following steps: S1: constructing a temperature control load group operation scheduling architecture, and analyzing scheduling information, control signals and power flow directions of the power grid side, the aggregator side and the user side; S2: modeling an indoor temperature variation process of a residential user based on an equivalent thermal parameter model, and restricting a dispatchable capacity range of the temperature control load, including an opening time constraint and a power boundary of the temperature control load, the temperature control load including a fixed-frequency air conditioner and a variable-frequency air conditioner; S3: extracting a temperature change rate, an energy use preference and an optimal temperature interval index based on historical indoor daily temperature curves and daily load curves of the residential user, clustering through a similarity measurement method of multi-dimensional vector feature fusion, and conservatively processing a state boundary of the temperature control load in the cluster; S4: constructing a distributed real-time optimization scheduling model based on the dispatchable capacity of each temperature control load group in S2: maximizing a response to a photovoltaic in a transformer area and a task issued by a scheduling center from a whole perspective, minimizing a temperature control load power consumption from an individual perspective, embedding the optimization model into a multi-agent deep deterministic policy gradient algorithm framework to optimize and solve a scheduling strategy, and realizing real-time and accurate regulation and control of the temperature control load in the transformer area.

2. The distributed control method of the temperature control load in the transformer area according to claim 1, wherein the S1 of constructing the operation scheduling architecture specifically comprises: S11: clearly defining a power flow direction: the user side senses a load state in real time through a temperature control load management system, and interacts with the aggregator; S12: determining an interaction mode of scheduling information of each subject: the aggregator side receives a power grid scheduling task and decomposes it to the user side, and the user side feeds back a scheduling result; S13: the user side installs a photovoltaic in the transformer area, and power supply is jointly borne by a transformer in the power grid side and the photovoltaic in the transformer area, forming a stable power transmission mode; S14: based on the temperature control load management system, the information of the temperature control load of each user in a building is sensed and a control command is issued, and signal interaction with the aggregator is realized.

3. The distributed control method of the temperature control load in the transformer area according to claim 1, wherein the S2 of modeling through the equivalent thermal parameter model specifically comprises: S21: analyzing a room temperature variation process by using a second-order equivalent thermal parameter model, and a model expression is: S22: establishing a dispatchable capacity boundary of the temperature control load, solving a thermodynamic equation in a scheduling period, and ignoring a solid temperature change in the equation, approximating to a first-order thermodynamic equation and discretizing the first-order thermodynamic equation: S23: the dispatchable capacity range constraint of the temperature control load includes: ； wherein, , , are the indoor air temperature, the indoor solid temperature and the outdoor air temperature at time t, respectively; , are the equivalent heat capacities representing the heat storage performance of the indoor solid and the indoor air; , are the equivalent thermal resistances representing the heat exchange rates between the indoor air and the outdoor air, and between the indoor solid and the indoor air; is the working state of the temperature-controlled load; is the working power of the temperature-controlled load; for the fixed-frequency air conditioner, an opening time constraint expression is: ； 4. The distributed control method of the temperature control load in the transformer area according to claim 1, wherein the S3 specifically comprises: S31: obtaining indoor temperature time series of all residential users in the transformer area and power time series of the temperature control load, extracting a sub-feature temperature change rate curve and a daily load curve, and normalizing data of the indoor daily temperature curve, the daily load curve and the sub-feature temperature change rate curve to convert the data into a multi-dimensional vector, and an expression is: ； ； ； wherein, is the minimum boot-up time, is the maximum boot-up time, , is the set optimal temperature range upper and lower limits; is the room temperature natural increase change amount; is the room temperature natural decrease change amount; For variable frequency air conditioners, the temperature control load power can be considered as continuously adjustable, and the load power constraint is: ; wherein is the lower limit of the schedulable power, and is: 。 ​ ​ ​ ； wherein is the indoor air temperature of the room, is the rate of change of the indoor temperature of the room, is the power of the room, n a single trajectory consisting of vector points within the dataset is represented as , ;​​​ S32: Similarity measurement of the multi-dimensional feature total trajectory curve set of the temperature-controlled load using the bidirectional Hausdorff distance, the expression being: ； wherein represents Euclidean distance between two points, calculated as: ; The similarity between any two trajectories is then: ; S33: The AP clustering algorithm is used to aggregate the temperature-controlled load of the residents, the clustering algorithm iteratively updates the attraction degree matrix and the belonging degree matrix until convergence, and the clustering result is output, the formula being: ； in For temperature control load j Temperature control load i The suitability of cluster centers indicates the degree of temperature control load. j To temperature control load i The attractiveness; For temperature control load i Select temperature-controlled load j The suitability for cluster centers is expressed as i arrive j degree of belonging; Similarity matrix middle i OK j Column values; Updating the values in the attraction degree matrix and the belonging degree matrix: ； Output clustering results , Clusters are recorded in k Temperature control load numbers.

5. The distributed control method of the transformer area temperature-controlled load according to claim 4, characterized by, Obtaining the clustering result for further processing: Based on the clustering results, all temperature-controlled loads were divided into y A cluster, for the fixed-frequency air conditioners in the cluster, the cluster k The minimum opening time is The maximum open time is For clusters k Medium-frequency inverter air conditioners, the lower limit of their adjustable power remains at [value missing]. The maximum schedulable power is ; Upper and lower limits of the optimal temperature range within the same cluster , The temperature control load corresponding to the cluster center shall be used as the standard.

6. The distributed control method for the temperature-controlled load of a transformer area according to claim 1, characterized in that, The objective function of the distributed real-time optimization scheduling model of S4 includes: establishing a mathematical model for the optimization scheduling of the temperature-controlled load of the transformer area, the objective function is divided into two layers, the upper layer aims to maximize the response to the photovoltaic of the transformer area and the scheduling task, and the lower layer aims to minimize the power consumption cost of the cluster, specifically: The upper layer is to cooperatively consume the photovoltaic of the transformer area by all the temperature-controlled loads of the transformer area, and to respond to the scheduling task issued by the scheduling center, the expression being: ； wherein is the maximum scheduling number, is the penalty coefficient of unfinished scheduling target, is the maximum photovoltaic output at each time, is the task amount issued by the scheduling center at each time, is the participation scheduling duration, the expression is: ; The expression of the lower layer objective is: ； wherein is the real-time electricity price.

7. The distributed control method for the temperature-controlled load of a transformer area according to claim 6, characterized in that, The framework of the multi-agent deep deterministic policy gradient algorithm includes: embedding the optimization model in the framework of the multi-agent deep deterministic policy gradient algorithm, establishing the state space, the observation space, the action space, and the reward function, and setting each cluster k as an agent, wherein: The state space is: ; The observation state space is: ; Action space is: ; The reward function of a single agent is: ； wherein is a weight coefficient of the double-layer objective function; The maximum training round is set, the optimization scheduling model is trained based on the algorithm framework until convergence, and the optimization scheduling strategy of the temperature-controlled load of each cluster under the transformer area is obtained, and the trained scheduling strategy is deployed to each temperature-controlled load management system for real-time scheduling control.

8. A distributed control system for the temperature-controlled load of a transformer area, characterized by comprising: A scheduling architecture module for constructing the operation scheduling architecture of the temperature-controlled load cluster of the transformer area, and analyzing the scheduling information, control signals and power flow direction of the power grid side, the aggregator side and the user side; An ETP model module for modeling the indoor temperature variation process of the residents based on the equivalent thermal parameter model, and constraining the schedulable ability range of the temperature-controlled load, including the start-up time constraint and the power boundary of the temperature-controlled load, wherein the temperature-controlled load includes the fixed-frequency and variable-frequency air conditioners; A multi-dimensional vector feature module for extracting the temperature change rate, energy preference and optimal temperature interval indicators based on the historical indoor daily temperature curve and daily load curve data of the residents, clustering through the similarity measurement method of multi-dimensional vector feature fusion, and conservatively processing the state boundary of the temperature-controlled load in the cluster; A solving scheduling module for constructing a distributed real-time optimization scheduling model based on the schedulable ability of each temperature-controlled load cluster in the ETP model module: maximizing the response to the photovoltaic of the transformer area and the task issued by the scheduling center from the overall perspective, and minimizing the power consumption cost of the temperature-controlled load from the individual perspective, and embedding the optimization model into the multi-agent deep deterministic policy gradient algorithm framework to optimize and solve the scheduling strategy, and realizing the real-time and accurate regulation and control of the temperature-controlled load of the transformer area.

9. An electronic device, comprising: comprising a memory and a processor; the memory is used to store a computer program; The processor is configured to implement the distributed control method for temperature-controlled load in a transformer area according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The storage medium has the computer program stored thereon, and the computer program, when executed by the processor, implements the distributed control method for temperature-controlled load in a transformer area according to any one of claims 1 to 7.

Images