CN-122026536-A - Virtual power plant scheduling control method considering segmentation capability and energy storage state optimization

CN122026536ACN 122026536 ACN122026536 ACN 122026536ACN-122026536-A

Abstract

The invention discloses a virtual power plant scheduling control method considering segmentation capability and energy storage state optimization, and relates to the technical field of virtual power plants and energy storage cooperative regulation and control. The method comprises the steps of obtaining power grid stepped regulation requirements, receiving real-time state data which is asynchronously reported by an energy storage system based on capacity stepped event driving, generating an initial scheduling strategy containing an expected action sequence according to the state data and a requirement instruction, constructing an operation safety boundary set based on an energy accumulation limit and a contract power boundary of the energy storage system, and performing out-of-limit evaluation and correction on the expected action sequence by utilizing a control barrier function to generate a target control parameter which is forcedly constrained in the boundary set. The invention is used for solving the problems of communication congestion caused by high-frequency concurrent reporting of mass distributed energy storage and safety of extremely easy breakdown of physical limits of bottom equipment and initiation of auxiliary service default in conventional AI scheduling.

Inventors

JIA WEI
WANG ZHENGFENG
WANG CHENGYU
FU PENG
CHEN HAO
ZHANG WEI
CHEN TIANYU
LIANG KUN

Assignees

国网安徽省电力有限公司合肥供电公司
国网安徽省电力有限公司宣城供电公司
国网安徽省电力有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. A virtual power plant scheduling control method taking segmentation capability and energy storage state optimization into consideration is characterized by comprising the following steps: Acquiring a step-type regulation demand instruction of a power grid, and receiving real-time allowance state data asynchronously reported by an energy storage system based on a capacity step event driving mechanism; Generating an initial scheduling strategy of the energy storage system according to the state data and the demand regulation instruction, wherein the initial scheduling strategy comprises an expected action sequence in a target period; Constructing an operation safety boundary set for restraining the energy storage system, wherein the boundary set is determined based on a contract power boundary corresponding to the regulation demand instruction and an energy storage limit threshold value of the energy storage system; And performing out-of-limit evaluation and correction on the expected action sequence based on a preset control obstacle function, and generating target control parameters constrained in the boundary set in a forced way.
2. The method of claim 1, wherein receiving real-time headroom status data asynchronously reported by the energy storage system based on a capacity step event driven mechanism comprises: analyzing the segmented declaration capacity requirement in the demand adjustment instruction to instruct the energy storage system to configure a plurality of non-equidistant discrete trigger thresholds; receiving a state pulse signal asynchronously reported by an energy storage system when the current energy accumulation depth crosses any one of the discrete trigger thresholds, wherein the pulse signal comprises a time stamp and a current available capacity parameter; the real-time headroom status data is updated based on the status pulse signal.
3. The method of claim 1, wherein generating an initial scheduling policy for an energy storage system based on the status data and the regulatory requirement instructions comprises: Extracting an adjusting direction, a minimum execution time span and a compensation excitation parameter in the adjusting demand instruction; Inputting the real-time allowance state data and the time-of-use electricity price data into a preset Markov decision process model; and solving the Markov decision process model by using the reinforcement learning algorithm with the aim of maximizing the expected benefit of the aggregate participation response, and outputting an expected action sequence containing the segmented energy throughput state and the corresponding power.
4. A method according to claim 3, wherein the adjustment direction comprises a power down scenario and a power up scenario, the expected sequence of actions satisfying mutually exclusive state constraints in different adjustment directions: In a power downlink scenario, the expected action sequence within the target period is defined as an energy feedback state or a standby state; in a power up scenario, the expected sequence of actions within the target period is defined as an energy injection state or a standby state.
5. The method of claim 1, wherein the constructing a set of operational safety boundaries that constrains the energy storage system comprises: Determining the upper limit and the lower limit of the maximum bidirectional transmission power and the energy accumulation depth of the energy storage system to form a physical constraint subset; Extracting a promised adjustment power dead zone corresponding to each response time period according to the confirmed adjustment demand instruction, and forming a contractual constraint subset; and carrying out intersection solution on the physical constraint subset and the contract constraint subset to obtain a dynamic operation safety boundary set corresponding to the current moment.
6. The method of claim 5, wherein the performing out-of-limit evaluation and correction of the expected motion sequence based on the preset control obstacle function comprises: converting the dynamic operation safety boundary set into a forward unchanged set of the control system; Constructing a continuous and differentiable boundary function for the invariant set, wherein the sign of the boundary function is used for representing whether the system state of the energy storage system is positioned in the invariant set; And judging whether the expected action sequence meets the forward evolution condition corresponding to the boundary function, and if not, correcting the expected action sequence based on quadratic programming.
7. The method of claim 6, wherein modifying the expected sequence of actions based on quadratic programming comprises: Taking the error between the minimized corrected instruction and the expected action sequence as an optimization target, and taking the evolution derivative of the boundary function as a safety constraint condition that the evolution derivative is greater than or equal to zero to construct a quadratic programming model; Carrying out online solving on the quadratic programming model to obtain a corrected deviation value; And adding the corrected deviation value to the expected action sequence, and outputting a target control parameter which enables the energy storage system to be strictly maintained in the unchanged set.
8. The method according to claim 7, characterized in that the quadratic programming model is expressed in particular as: Wherein, the For the control parameter of the object to be solved, In order to anticipate a sequence of actions, As a system state variable of the energy storage system, As a continuously differentiable boundary function, As an evolving derivative of the boundary function with respect to time, Is a strictly increasing extension class K function and meets 。
9. The method of claim 1, further comprising, after generating the target control parameter, issuing it to an end effector of the energy storage system, comprising: Packaging the target control parameters according to a preset data frame format; And the energy interactive inversion unit is transmitted to the energy storage system through the industrial communication bus, so that the energy interactive inversion unit executes transient power output according to the target control parameter and under a preset droop control coefficient.
10. The method according to any one of claims 1 to 9, further comprising: after the target control parameters are issued, the actual energy throughput flow rate and the actual energy accumulation depth of the energy storage system are obtained; An actual execution bias in a current response period is calculated based on the energy throughput flow rate, and an initial state parameter of a next control period is updated according to the actual energy accumulation depth.

Description

Virtual power plant scheduling control method considering segmentation capability and energy storage state optimization Technical Field The invention relates to the technical field of virtual power plants and energy storage cooperative regulation and control, in particular to a virtual power plant scheduling control method considering segmentation capability and energy storage state optimization. Background The virtual power plant provides auxiliary services such as peak clipping, valley filling, frequency supporting and the like for the power grid by aggregating massive distributed energy storage resources. When processing the step-type regulation demand instruction issued by the power grid, the system needs to monitor and overall the energy accumulation state of each node energy storage unit in real time so as to match the power response contracts of different time scales. Dynamic adjustment of large-scale heterogeneous resources puts extremely high demands on communication connectivity and physical security of control instructions. The existing virtual power plant scheduling generally adopts a global polling mechanism with a fixed period to collect the real-time state of charge of the bottom equipment, and the real-time state of charge is used as a data base for centralized regulation and control. In the decision generation link, the main flow scheme outputs an action sequence by relying on a conventional deep reinforcement learning algorithm, and attempts to guide the black box model to consider economic benefit and grid-connected response safety by applying a soft constraint mode of penalty items in an objective function. However, the scheme has obvious technical limitations that on one hand, high-frequency fixed polling can cause serious channel congestion and time delay coupling when facing massive concurrent nodes, and can not capture hidden out-of-limit risks when the stored energy approaches a limit boundary sharply, and on the other hand, soft constraint reinforcement learning based on probability distribution is extremely easy to fail under extreme power grid fluctuation, so that random strategy actions break down hardware physical limits or deviate from contract logic. In the prior art, a deterministic topological manifold correction link is difficult to construct in a bottom execution link, so that irreversible damage of a battery cell or high-volume market default are extremely easy to be caused, and stable regulation and control which takes both discrete event perception and absolute boundary defense into consideration are difficult to realize in a complex power grid scene of a system. Disclosure of Invention In order to overcome the defects of the prior art, the embodiment of the invention provides a virtual power plant scheduling control method taking the segmentation capability and the energy storage state optimization into consideration, wherein a capacity step event driving mechanism is adopted to report an asynchronous state, and an initial scheduling strategy generated by combining a power grid step instruction is constrained in a safety boundary set formed by equipment physical limit and market contract red line by utilizing a control barrier function, so that the problems that communication congestion is caused by high-frequency polling of mass distributed energy storage nodes, and irreversible damage and auxiliary service default of bottom equipment are extremely easy to be caused by conventional black box AI scheduling are solved. In order to achieve the above purpose, the present invention provides the following technical solutions: A virtual power plant scheduling control method considering segmentation capability and energy storage state optimization comprises the following steps of obtaining a step-type regulation demand instruction of a power grid, receiving real-time allowance state data asynchronously reported by an energy storage system based on a capacity step event driving mechanism, generating an initial scheduling strategy of the energy storage system according to the state data and the regulation demand instruction, wherein the initial scheduling strategy comprises an expected action sequence in a target period, constructing an operation safety boundary set for restraining the energy storage system, determining a contract power boundary corresponding to the regulation demand instruction based on an energy accumulation limit threshold of the energy storage system by the boundary set, and performing out-of-limit evaluation and correction on the expected action sequence based on a preset control barrier function to generate a target control parameter forcedly restrained in the boundary set. In a preferred embodiment, the receiving of the real-time margin state data asynchronously reported by the energy storage system based on the capacity step event driving mechanism comprises analyzing the segment reporting capacity requirement in the regulation demand instruction to in