Virtual power plant multi-agent cooperative control method and system based on layered and partitioned architecture

Abstract

The invention relates to the technical field of electric power supply, in particular to a virtual power plant multi-agent cooperative control method and system based on a layered and partitioned architecture, and the method comprises the steps: collecting the historical operation data of distributed resources in each partition; calculating the output stability of each resource based on historical data, and introducing the stability as a penalty term into a scheduling optimization objective function; solving the target function by adopting an improved particle swarm algorithm which adaptively adjusts a search strategy according to the output stability to obtain an optimal output power distribution plan of the current period; and issuing and executing the plan according to partitions. According to the method, the resource stability is quantified and the optimization model is integrated, so that the scheduling scheme automatically avoids excessive dependence on unreliable resources, and the optimal balance of economy and reliability is realized under a layered and partitioned architecture in combination with a self-adaptive optimization algorithm.

Description

Technical Field

[0001] This invention relates to the field of power supply. More specifically, this invention relates to a multi-entity collaborative control method and system for virtual power plants based on a hierarchical and partitioned architecture. Background Technology

[0002] A virtual power plant is a technological form that aggregates dispersed power resources such as distributed power sources, energy storage systems, and controllable loads through advanced information and communication technologies and software systems, and participates in grid operation and power market coordination and optimization as a special power plant.

[0003] However, traditional virtual power plant scheduling optimization ignores the stability differences exhibited by each distributed resource during historical operation. This may lead to scheduling results that overly rely on resources with large output fluctuations and unreliability, thereby increasing the operational risk of the system and reducing power supply reliability. Summary of the Invention

[0004] To address the technical problems of traditional virtual power plant scheduling methods ignoring differences in resource stability and making it difficult to achieve reliable collaborative control under a hierarchical and partitioned architecture, this invention provides solutions in the following aspects.

[0005] In a first aspect, the present invention provides a multi-entity collaborative control method for a virtual power plant based on a hierarchical and partitioned architecture. The virtual power plant adopts a hierarchical architecture including a central control layer and at least one regional control layer, and its distributed resources are divided into multiple partitions according to a preset logic. The method is executed by the central control layer and includes the following steps: Historical operational data of distributed resources in each partition of the virtual power plant is collected, including the sequence of dispatch command values ​​and their corresponding actual output power sequences. The collected historical operational data is preprocessed to form a historical dataset. Based on the historical dataset, calculate the output stability of each distributed resource; establish a scheduling optimization objective function, wherein the scheduling optimization objective function includes a penalty term with the output stability as a parameter; The scheduling optimization objective function is solved using a solution algorithm, wherein the solution algorithm adjusts the solution strategy according to the output stability to obtain an output power allocation plan for distributed resources in each partition within the current scheduling period; The output power allocation plan is distributed to the corresponding regional control layer by region to achieve multi-entity collaborative control.

[0006] Preferably, the layered architecture further includes a local execution layer that is communicatively connected to the regional control layer; the regional control layer decomposes the received output allocation plan into executable instructions for specific resources in the local execution layer.

[0007] Preferably, after obtaining the historical data set, the historical data set is divided into a plurality of continuous time periods according to a scheduling period, and each time period corresponds to a scheduling instruction value and an actual output power sequence corresponding thereto.

[0008] Preferably, the output stability obtaining process is: Selecting any distributed resource as a target resource and selecting any time period as a target time period, calculating a variation coefficient of an actual output power sequence of the target resource in the target time period, and obtaining self-output stability of the target resource in the target time period by taking a negative value of the variation coefficient and performing normalization processing; Calculating a relative deviation of an actual output power mean value of the target resource in the target time period from the scheduling instruction value, and obtaining instruction tracking fidelity of the target resource in the target time period by performing exponential function operation on the relative deviation; Taking a product of the self-output stability and the instruction tracking fidelity as a contribution value of the target resource in the target time period, and obtaining output stability of the target resource by taking an arithmetic mean value of the contribution values of the target resource in all time periods.

[0009] Preferably, the process of establishing the scheduling optimization objective function is: Pre-acquiring a traditional cost objective function and an output power plan value allocated to each distributed resource in a current scheduling period; Calculating a product of a difference between 1 and the output stability of a single distributed resource and the output power plan value allocated to the single distributed resource, and summing up the products calculated by traversing all distributed resources to obtain a stability penalty term; Multiplying the traditional cost objective function and the stability penalty term to constitute the scheduling optimization objective function.

[0010] Preferably, after obtaining the scheduling optimization objective function, a constraint condition is set for the scheduling optimization objective function, the constraint condition includes an output balance constraint and a resource output upper and lower limit constraint; the output balance constraint is that a sum of the output power plan values allocated to all distributed resources is greater than or equal to total power required in the current scheduling period; and the resource output upper and lower limit constraint requires that the output power plan value allocated to each distributed resource is within a safe operation range allowed by physical equipment thereof.

[0011] Preferably, the solving algorithm is a particle swarm optimization algorithm; and the adjusting solving strategy according to the output stability includes: In an iteration process of the particle swarm optimization algorithm, for each particle in the population, based on a correlation between a candidate output power allocation plan represented by the particle and a sequence formed by the output stability of all distributed resources, and a ratio of a current scheduling optimization objective function value of the particle to a global optimal scheduling optimization objective function value, a regulation factor is calculated. adjust the individual learning factor and the global learning factor of the particle according to the adjustment factor.

[0012] Preferably, the iteration is terminated when a preset maximum iteration number is reached, and a candidate output power allocation plan that minimizes the scheduling optimization objective function value when the iteration is terminated is taken as the final output power allocation plan.

[0013] Preferably, actual operation data generated by each distributed resource when executing the output power allocation plan at the end of the current scheduling period is taken as historical operation data to be collected in the next scheduling period.

[0014] In a second aspect, the present application provides a virtual power plant multi-agent collaborative control system based on a hierarchical partitioning architecture, comprising: a processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement any of the virtual power plant multi-agent collaborative control methods based on a hierarchical partitioning architecture.

[0015] The present application has the following beneficial effects: 1. The present application quantitatively evaluates the historical output stability of each distributed resource and introduces it as a penalty term into the optimization objective, so that the scheduling scheme can automatically avoid excessive dependence on unstable resources while pursuing economy, thereby reducing the power deviation risk from the source of decision-making and significantly improving the overall operation reliability of the virtual power plant.

[0016] 2. The present application designs an adaptive improved particle swarm algorithm for the complexity of multi-agent coordination under a hierarchical partitioning architecture. The algorithm dynamically adjusts the search strategy using resource stability information, balances the global exploration and local development capabilities, and thus can find a high-quality scheduling scheme more quickly and accurately, ensuring the efficiency and effectiveness of collaborative control. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a method flowchart of steps S1-S4 in the virtual power plant multi-agent collaborative control method based on a hierarchical partitioning architecture according to an embodiment of the present application.

[0018] Figure 2 is a structural schematic diagram of the virtual power plant multi-agent collaborative control system based on a hierarchical partitioning architecture according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all of the embodiments.

[0020] To cope with the challenges of wide distribution, large number, and complex control, the present application adopts a "central-regional-local" three-layer control architecture and manages in combination with resource logical partitioning: Central control layer: as the decision core, deployed in the cloud platform or master station system. Responsible for global data aggregation, resource stability evaluation, cross-zone coordination optimization, and total scheduling plan generation.

[0021] Regional control layer: divided into multiple control zones (such as A zone, B zone) according to the grid structure or geographical boundaries, and each zone is provided with a regional controller. Responsible for receiving central layer instructions, aggregating and managing resources within the zone, decomposing and forwarding instructions, monitoring running state and reporting data.

[0022] Local execution layer: composed of local control units of various distributed resources (such as photovoltaic inverters, energy storage PCS, and load controllers). Responsible for executing specific power regulation instructions and collecting real-time local operation data for uploading.

[0023] Resource partitioning: based on regional division, logical grouping can be further performed according to resource types (such as photovoltaic and energy storage) or reliability levels, facilitating fine modeling and collaborative strategy formulation.

[0024] Reference Figure 1 The steps of the method of the present application will be described in detail. The method of the present application is executed in a fixed scheduling period (which can be set according to actual system requirements, for example, 15 minutes, 30 minutes or 1 hour). In each current scheduling period, the central control layer performs the following steps S1-S4 to generate an executable output power plan for the current scheduling period.

[0025] S1: Collect historical operation data of distributed resources in each zone of the virtual power plant, including scheduling instruction value sequence and its corresponding actual output power sequence, and pre-process the collected historical operation data to form a historical data set.

[0026] The central control layer collects historical operation data of all distributed resources in each zone over a period of time through a communication network. The historical operation data mainly includes two types: Scheduling instruction value: records the output power of each distributed resource historically issued.

[0027] Actual output power: the output power actually measured by each distributed resource corresponding to the above scheduling instruction value.

[0028] Further, the scheduling instruction value sequence and its corresponding actual output power sequence within the collection time are obtained. The collected historical operation data is further pre-processed to obtain a historical data set.

[0029] The pre-processing of the historical operation data includes filling in missing data caused by communication interruption or the like by using linear interpolation or the like to ensure data continuity.

[0030] Further, the historical data set is divided into several regular periods in units of a response period of each dispatching instruction, each period corresponds to a dispatching instruction value and an actual output power sequence corresponding thereto.

[0031] S2: Based on the historical data set, the output stability of each distributed resource is calculated, and a scheduling optimization objective function is established, which includes a penalty term with the calculated output stability as a parameter.

[0032] Traditional scheduling optimization is only based on economic indicators (such as cost, price) for optimization, and fluctuating resources such as photovoltaic and wind power are treated equally with stable resources such as energy storage and gas turbines, without considering the essential difference in their historical operation reliability. Some resources can stabilize output and accurately follow instructions, while others fluctuate sharply and often deviate from the target.

[0033] In order to solve the above problems, the reliability of the resource is first quantified, that is, the output stability of each distributed resource is calculated based on the historical data set, which comprehensively evaluates the stability of the distributed resource in each period and its following accuracy of the dispatching instruction. It is a value between 0 and 1, and the larger the value, the more stable and reliable the corresponding distributed resource is in historical operation.

[0034] After obtaining the quantified output stability, a new scheduling optimization objective function is established, which introduces a penalty term with the output stability as a key parameter on the basis of traditional scheduling optimization (i.e. traditional economic cost term). The design of the penalty term makes it possible to generate higher risk cost when allocating power to resources with poor stability, and lower risk cost when allocating power to resources with high stability.

[0035] For example, select any distributed resource as a target resource and select any period as a target period, calculate the coefficient of variation of the actual output power sequence of the target resource in the target period, take the negative value of the coefficient of variation and normalize it to obtain the output stability of the target resource in the target period. Calculate the relative deviation of the actual output power mean of the target resource in the target period and the dispatching instruction value, and perform exponential function operation on the relative deviation to obtain the instruction tracking fidelity of the target resource in the target period.

[0036] The product of the above output stability and the above instruction tracking fidelity is taken as the contribution value of the target resource in the target period.

[0037] wherein the above self-output stability is expressed by a relationship as follows: wherein, is the self-output stability of the target resource in the target period, is the coefficient of variation of the actual output power sequence of the target resource in the target period, denotes a normalization process. Through the normalization process, the dimensional influence of different capacity resources is eliminated, fair comparison is achieved, and the case that the absolute fluctuation value of a large capacity resource may cover up the instability of a small capacity resource is avoided.

[0038] wherein the above instruction tracking fidelity is expressed by a relationship as follows: wherein, is the instruction tracking fidelity of the target resource in the target period, is the average value of the actual output power of the target resource in the target period, is the scheduling instruction value corresponding to the target period, is an exponential function with a natural constant e as the base. By using the exponential function, a nonlinear severe punishment is implemented on the scheduling deviation. The power system requires highly reliable instruction following, and the exponential function can achieve "light punishment for small deviation and heavy punishment for large deviation", which is more in line with the engineering safety requirements than linear punishment.

[0039] Finally, the contribution values of the target resource in all periods are taken as an arithmetic mean value to obtain the output stability of the target resource.

[0040] Further, the output stability of all distributed resources is obtained according to the calculation process of the above output stability of the target resource.

[0041] Exemplarily, the process of establishing the scheduling optimization objective function is as follows: Firstly, a traditional cost objective function is obtained, i.e.: wherein, is an economic cost item, such as electricity purchase cost, operation and maintenance cost, etc., is a minimum value. In addition, the output power plan value allocated to each distributed resource in the current scheduling period is also obtained.

[0042] Then, on the basis of the above traditional cost objective function, the product of the difference between the output stability of the single distributed resource and the output stability of the 1 and the output power plan value allocated to the single distributed resource is calculated, and the sum of the products calculated by traversing all distributed resources is obtained to obtain a stability penalty term. The stability penalty term is multiplied by the above traditional cost objective function to form the above scheduling optimization objective function.

[0043] The scheduling optimization objective function Expressed by a relational expression as: In the formula, is the output stability of the distributed resource , is the output power planning value of the distributed resource , is the total number of distributed resources.

[0044] Wherein, As a stability penalty term, its working principle is as follows: is the resource instability coefficient, which is directly derived from the output stability calculated in the previous step. The larger the resource instability coefficient is, the lower the reliability of the distributed resource is; The stability penalty term for a single distributed resource is composed of the resource instability coefficient, and its physical meaning is that allocating a large output power ( ) to an unreliable distributed resource ( ) will result in a higher risk cost.

[0045] Through the scheduling optimization objective function composed of the above, when the optimization algorithm is looking for a solution that minimizes , if it tries to reduce by greatly increasing the output power planning value of the unstable resource, it will result in a sharp rise in the above stability penalty term, which may instead worsen the scheduling optimization objective function . Therefore, the algorithm is forced to make a mathematical optimal trade-off between “economic benefits” and “stability risk costs”.

[0046] In addition, in order to ensure the realizability of the optimization result and the safety of the system, a constraint condition is added to the above scheduling optimization objective function, including an output balance constraint and a resource output upper and lower limit constraint. The output balance constraint is that the sum of the output power planning values allocated to all distributed resources is greater than or equal to the total power required in the current scheduling period. The resource output upper and lower limit constraint requires that the output power planning value allocated to each distributed resource is within the safe operating range allowed by its physical device, preventing device overload or operation in unstable conditions, and ensuring device safety and the executability of the scheduling instruction.

[0047] S3: Use a solving algorithm to solve the scheduling optimization objective function, wherein the solving algorithm adjusts the solving strategy according to the output stability to obtain an output power distribution plan for the distributed resources in each partition within the current scheduling period.

[0048] The above S2 has already constructed a scheduling optimization objective function, and then the scheduling optimization objective function is solved. Commonly used solution algorithms include the standard particle swarm optimization algorithm. However, its learning factor and other parameters are usually fixed values. In the application scenario of this invention, it is impossible to dynamically adjust according to the real-time status (such as particle mass and convergence status) during the solution process, which leads to slow convergence speed or easy getting trapped in local optima.

[0049] To address the aforementioned issues, this invention dynamically transforms the calculated output stability information from static evaluation parameters at the model level into an intelligent strategy guiding the algorithm's search process. Specifically, based on output stability, the algorithm's internal search mechanisms, such as learning factors, are adjusted in real-time and adaptively, enabling the solution itself to possess risk perception and stability-oriented capabilities, thereby achieving a closed loop from defining standards to efficiently achieving those standards.

[0050] During the iterative process of the particle swarm optimization algorithm, for each particle in the swarm, the Pearson correlation coefficient between the candidate output power allocation plan represented by the particle and the sequence formed by the output stability of all distributed resources is calculated. In addition, the optimal scheduling optimization objective function value that can be achieved by all particles in the entire particle swarm and the scheduling optimization objective function value of the candidate output power allocation plan corresponding to a single particle are obtained after the iteration of the particle.

[0051] Furthermore, based on the elements obtained above, a regulation factor is calculated, which can be expressed as a relational formula: In the formula, For the first The second iteration The adjustment factor for each particle. For the first The Pearson correlation coefficient between the candidate output power allocation scheme represented by each particle and the sequence consisting of the output stability of all distributed resources. In the first The optimal scheduling objective function value that can be achieved by all particles in the entire particle swarm after the next iteration. In the first In the nth iteration The objective function value for scheduling optimization of candidate output power allocation plans for each particle.

[0052] In the optimal scheduling scheme, distributed resources with high output stability are expected to be allocated more output power, while distributed resources with low output stability are allocated less output power. Therefore, the closer the Pearson correlation coefficient is to 1, the more the candidate output power allocation plan of the particles conforms to the stability-first scheduling principle, and the higher the inherent rationality of the candidate output power allocation plan. Through calculation... The Pearson correlation coefficient was changed from Linear mapping to Interval.

[0053] calculate The closer the ratio is to 1, the more significant the change in the first digit. The closer the candidate output power allocation plan of the last particle is to the global optimum, the better its search performance; the smaller the ratio is than 1, the better the search performance of the last particle. The greater the difference between the previous candidate output power allocation plan for a particle and the global optimum, the worse the performance.

[0054] In each iteration, each particle calculates an independent set of learning factors for each distributed resource.

[0055] After obtaining the adjustment factors, the individual learning factors and global learning factors of each particle are further adjusted for each distributed resource. The adjusted individual learning factors and global learning factors are expressed by the following relationships: In the formula, For the first The particle in the first Distributed resources during the next iteration Individual learning factors For the first The second iteration The adjustment factor for each particle. For the first Individual particles in distributed resources Distribution characteristics along this dimension For the first The particle in the first Distributed resources during the next iteration The global learning factor It is an exponential function with the natural constant e as its base.

[0056] Among them, the calculation in the first... After the [number] iterations, the [number]th iteration... In the candidate output power allocation plan represented by each particle, the plan is to allocate power to distributed resources. The planned output power value is used as the first parameter; the calculation is performed at the... After the last iteration, the entire particle swarm is working on distributed resources. The arithmetic mean of all proposed candidate output power plans is used as the second parameter; the entire particle swarm is calculated for distributed resources. The standard deviation of all candidate planned output power values ​​proposed for planned output power allocation is used as the third parameter.

[0057] The above is obtained by dividing the difference between the first and second parameters by the third parameter (or by the sum of the third parameter and a very small positive number). .

[0058] Through the above adjustments, for high-quality particles ( Larger particles, by significantly increasing their individual learning factors, are encouraged to strengthen their self-confidence and delve deeper into their already discovered strengths; for inferior particles ( By significantly increasing its global learning factor, the small (small) solution is prompted to abandon its undesirable self-direction and quickly move towards the globally optimal solution recognized by the population.

[0059] Then, the adjusted individual learning factor and global learning factor are substituted into the particle swarm optimization algorithm, and the scheduling optimization objective function in S2 above is solved. When the maximum number of iterations is reached, such as 200, the iteration is terminated, and the output power allocation plan corresponding to the optimal particle is output.

[0060] S4: Distribute the output power allocation plan to the corresponding regional control layer by zone to achieve multi-entity collaborative control.

[0061] The S3 above only yields a mathematically optimal solution, which still needs to be converted into actual control instructions and executed in the physical system.

[0062] The central control layer encapsulates the optimized output power allocation plan into scheduling instructions according to the corresponding partitions and distributes them to the regional control layers through the communication network. After receiving the instructions, the regional control layer combines the real-time status of each distributed resource within its partition with local constraints, decomposes the output power allocation plan, and converts it into executable control commands for specific resources (such as photovoltaic inverters and energy storage converters), and distributes them to the local execution layer for execution.

[0063] The local execution layer drives the equipment to complete power regulation and feeds back the actual output power sequence, equipment status, and other operational data generated during the execution process to the central control layer in real time through the regional control layer. The central control layer performs global monitoring and coordination of the execution of plans in each zone.

[0064] Furthermore, after the current scheduling cycle is completed, the system will structure the issued output power plan value and the actual operating data (especially the actual output power sequence) generated by each distributed resource in response to the plan, and archive them to the historical database. This newly archived data, together with previous historical data, constitutes an updated dataset that more closely reflects the current system state. At the start of the next scheduling cycle, this data will be used as the "historical operating data to be collected" in step S1 to calculate the updated output stability and drive a new round of optimized scheduling.

[0065] This invention also provides a multi-agent collaborative control system for a virtual power plant based on a hierarchical and partitioned architecture. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, they implement the multi-agent collaborative control method for a virtual power plant based on a hierarchical partitioning architecture according to the first aspect of the present invention.

[0066] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0067] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for multi-agent collaborative control of a virtual power plant based on a hierarchical partitioning architecture, characterized in that, The virtual power plant adopts a hierarchical architecture including a central control layer and at least one regional control layer, and distributed resources thereof are divided into multiple partitions according to a preset logic; the method is executed by the central control layer, and includes: collecting historical operation data of the distributed resources in each partition in the virtual power plant, including a sequence of scheduling instruction values and a corresponding sequence of actual output powers, and pre-processing the collected historical operation data to form a historical data set; based on the historical data set, calculating output stability of each distributed resource; and establishing a scheduling optimization objective function, wherein the scheduling optimization objective function includes a penalty term with the output stability as a parameter; solving the scheduling optimization objective function by using a solving algorithm, wherein the solving algorithm adjusts a solving strategy according to the output stability to obtain an output power distribution plan for the distributed resources in each partition in a current scheduling period; downloading the output power distribution plan to the corresponding regional control layer according to the partition to realize multi-agent collaborative control.

2. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 1, characterized in that, The hierarchical architecture further includes a local execution layer in communication connection with the regional control layer; and the regional control layer decomposes the received output distribution plan into executable instructions for specific resources in the local execution layer.

3. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 1, characterized in that, After obtaining the historical data set, the historical data set is divided into multiple continuous time periods according to a scheduling period, and each time period corresponds to a sequence of actual output powers and a corresponding sequence of scheduling instruction values.

4. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 3, characterized in that, The output stability is obtained by: selecting any distributed resource as a target resource and selecting any time period as a target time period, calculating a variation coefficient of the sequence of actual output powers of the target resource in the target time period, taking a negative value of the variation coefficient and performing normalization processing to obtain self-output stability of the target resource in the target time period; calculating a relative deviation of the average of the actual output power of the target resource in the target time period from the scheduling instruction value, and performing exponential function operation on the relative deviation to obtain instruction tracking fidelity of the target resource in the target time period; multiplying the self-output stability and the instruction tracking fidelity to obtain a contribution value of the target resource in the target time period, and taking an arithmetic average of the contribution values of the target resource in all time periods to obtain the output stability of the target resource.

5. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 1, characterized in that, The process of establishing the scheduling optimization objective function is: pre-obtaining a traditional cost objective function and an output power plan value allocated to each distributed resource in a current scheduling period; calculating a product of a difference between 1 and the output stability of a single distributed resource and the output power plan value allocated to the single distributed resource, and summing the products calculated by traversing all distributed resources to obtain a stability penalty term; multiplying the traditional cost objective function and the stability penalty term to constitute the scheduling optimization objective function.

6. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 5, characterized in that, After the scheduling optimization objective function is acquired, a constraint condition is set for the scheduling optimization objective function, the constraint condition including an output balance constraint and a resource output upper and lower limit constraint; the output balance constraint being that a sum of output power plan values allocated to all distributed resources is greater than or equal to total power required in a current scheduling period; and the resource output upper and lower limit constraint requiring that the output power plan value allocated to each distributed resource is within a safe operation range allowed by physical equipment thereof.

7. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 5, characterized in that, The solving algorithm is a particle swarm optimization algorithm; and the solving strategy is adjusted according to the output stability, and the adjusting the solving strategy according to the output stability includes: In an iteration process of the particle swarm optimization algorithm, for each particle in a population, a regulation factor is calculated based on a correlation between a candidate output power distribution plan represented by the particle and a sequence of output stabilities of all distributed resources, and a ratio of a current scheduling optimization objective function value of the particle to a global optimal scheduling optimization objective function value; According to the regulation factor, an individual learning factor and a global learning factor of the particle are adjusted.

8. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 7, characterized in that, Iteration is terminated when a preset maximum iteration number is reached, and a candidate output power distribution plan that minimizes the scheduling optimization objective function value obtained when the iteration is terminated is taken as a final output power distribution plan.

9. The virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to claim 1, characterized in that, After the output power distribution plan is executed in the current scheduling period, actual operation data generated by each distributed resource when the output power distribution plan is executed is taken as historical operation data to be collected in a next scheduling period.

10. A virtual power plant multi-agent collaborative control system based on a hierarchical partitioning architecture, characterized in that, The method includes: a processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the virtual power plant multi-agent collaborative control method based on the hierarchical partitioning architecture according to any one of claims 1-9.

Images