CN115986748A - A data-driven microgrid voltage control method - Google Patents

Abstract

The invention discloses a data-driven microgrid voltage control method, which relates to the technical field of microgrid application and comprises the following steps: dividing the microgrid into a plurality of sub-networks which are mutually connected and coupled through tide based on a distributed architecture, and constructing an agent corresponding to the internal control of the sub-networks aiming at each sub-network; constructing a combined control model with the reactive output of the photovoltaic inverter as a main part and the active output of the energy storage system as an auxiliary part, and constructing a barrel-shaped voltage barrier function model by combining a V-shaped voltage barrier function model and a U-shaped voltage barrier function model; taking voltage constraint based on the network total power loss and the voltage barrier function model as target control, balancing voltage deviation and network power loss by using a weighting sum algorithm, and finally establishing a micro-grid distributed VVC model; the micro-grid distributed VVC model is fitted into a partially observable Markov decision POMG model, and a state action equation is modified from adaptation of discrete action to adaptation of continuous action so as to adapt to the real-time control requirement.

Description

A data-driven voltage control method for microgrids

Technical Field

The present invention relates to the technical field of microgrid applications, and in particular to a data-driven microgrid voltage control method.

Background Art

Microgrid (microgrid) refers to a small power generation and distribution system composed of distributed power sources, energy storage devices, energy conversion devices, related loads and monitoring and protection devices. It is an autonomous system that can achieve self-control, protection and management. It can be connected to the external power grid or run in isolation. It is an important part of the smart grid. Microgrid has a dual role: for the power grid, as a smart load with variable size, microgrid provides dispatchable loads for the local power system, can respond within seconds to meet system needs, and provide strong support to the large power grid in a timely manner; it can maintain the system without affecting the customer's load; it can reduce (extend) the replacement of the distribution network, adopt the IEEE1547.4 standard, guide the operation of distributed power islands, and eliminate the technical obstacles caused by certain special operating requirements. For users, microgrids, as a customizable power source, can meet the diverse needs of users, such as enhancing local power supply reliability, reducing feeder losses, supporting local voltages, improving efficiency by utilizing waste heat, providing voltage sag correction, or serving as uninterruptible power services. Microgrids not only solve the problem of large-scale access to distributed power sources, give full play to the advantages of distributed power sources, but also bring many other benefits to users. Microgrids will fundamentally change the traditional way of dealing with load growth, and have great potential in reducing energy consumption, improving the reliability and flexibility of power systems, etc.

In order to solve the power management problems such as voltage regulation and reducing network power loss in microgrids, some voltage/var control (VVC) methods have been developed. Among them, inverters and energy storage devices have been widely studied and used to participate in microgrid regulation control due to their increasing availability, flexibility and fast response speed of advanced power electronics technology. There are currently three main control frameworks for microgrids: centralized control, local control and distributed control. Distributed methods have been increasingly studied because they do not require complex communication networks and can take into account existing privacy, complexity, scalability and global issues.

A common research idea for distributed control is to transform non-convex problems into convex optimization problems through reasonable approximation and simplification, and then use distributed algorithms to solve them. In order to cope with the uncertainty of distributed renewable energy and load demand, model-based distributed control methods usually need to predetermine the optimal solution. However, when new situations arise, traditional model control methods must solve optimization problems, which requires a lot of calculations. In addition, model-based control methods usually require accurate and comprehensive grid parameters, which are often difficult to achieve.

Summary of the invention

In order to solve the deficiencies mentioned in the above background technology, an object of the present invention is to provide a data-driven microgrid voltage control method.

The purpose of the present invention can be achieved by the following technical solution: a data-driven microgrid voltage control method, the method comprising the following steps:

Based on the distributed architecture, the microgrid is divided into multiple sub-networks that are interconnected and coupled through power flow, and an intelligent agent for the internal control of each sub-network is constructed;

Construct a joint control model with the reactive output of photovoltaic inverter as the main function and the active output of energy storage system as the auxiliary function. A single intelligent agent corresponding to each subnetwork controls all photovoltaic inverters and energy storage devices in the subnetwork, and effectively controls the voltage of the microgrid by controlling the reactive output of photovoltaic inverter and the active output of energy storage system.

Combining V-type and U-type voltage barrier function models, a barrel-type voltage barrier function model is constructed;

The voltage constraint based on the total network power loss and voltage barrier function model is used as the target control, and the voltage deviation and network power loss are weighed using the weighted sum algorithm. A spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals is established. A network flow constraint model based on the Newton-Raphson method is established. A spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals is integrated. A joint control model with the photovoltaic inverter as the main reactive output and the energy storage system as the auxiliary active output is established. The control target after weighing the voltage deviation and network power loss and the network flow constraint model are established to establish a distributed VVC model of the microgrid.

The distributed VVC model of the microgrid is fitted into a partially observable Markov decision POMG model, and the state-action equation is improved from adapting to discrete actions to adapting to continuous actions to meet the real-time control requirements;

The constructed spatiotemporal uncertainty model based on PV and load forecast intervals is transformed into a stochastic programming model, and a network solution failure penalty is added to the return reward of each training set to improve the network solution success rate;

Using the multi-agent deep deterministic policy gradient MADDPG algorithm, the algorithm process is designed and applied to microgrids.

Preferably, the distributed architecture does not require a central coordinator to collect all the information generated in the network, and needs to consider global coordination issues; in distributed optimization, the sub-network represented by each constructed intelligent agent only exchanges limited boundary physical information and global reward returns with its adjacent sub-networks, collectively seeking a global optimal solution, and during each operation, the trained controller implements locally measured VVC within the corresponding sub-network.

Preferably, the joint control model in which the photovoltaic inverter is mainly reactive and the energy storage system is supplemented by active output is as follows:

表示光伏输出的实时有功功率，

表示光伏逆变器输出的实时无功功率，

表示光伏逆变器的复功率，

表示t时的光伏有功功率输出边界，δ光伏逆变器无功功率容量因数，

和

表示储能系统可以发出和吸收的最小和最大有功功率，

表示在t时储能的有功功率输出值，

表示t时储能的实时容量，

表示储能的最大容量。

Indicates the real-time active power output of photovoltaic power.

Indicates the real-time reactive power output by the photovoltaic inverter.

represents the complex power of the PV inverter,

represents the photovoltaic active power output boundary at time t, δ is the photovoltaic inverter reactive power capacity factor,

and

Indicates the minimum and maximum active power that the energy storage system can generate and absorb,

represents the active power output value of energy storage at time t,

represents the real-time capacity of energy storage at time t,

Indicates the maximum capacity of energy storage.

Photovoltaic inverters give priority to providing reactive power when needed. When the reactive power compensation capacity is insufficient, the energy storage system will take action. The reactive power output of each photovoltaic inverter is also limited to a preset proportion of its apparent power capacity. A positive value indicates the injection of reactive power into the grid, and a negative value indicates the absorption of grid reactive power. The setting of the energy storage system is similar to that of the inverter. A positive value indicates the injection of active power into the grid, and a negative value indicates the absorption of grid active power. The remaining power of the energy storage system is always positive.

Preferably, the barrel voltage barrier function model is as follows:

In the formula, _va is the real-time voltage of the node; v _ref is the network voltage reference value, which is 1.00 pu; l _v ( _va ) is the real-time reward of the node voltage;

The voltage barrier function model combines the advantages of the V-type and the U-type: on the one hand, it has a slow gradient within the safety range to obtain better voltage conditions; on the other hand, the larger gradient outside the safety range ensures faster strategy guidance.

Preferably, the control target after weighing the voltage deviation and network power loss is:

为网络支路集合；r_ij和x_ij分别表示i和j节点之间支路的电阻和电抗；v_i,t和v_j,t分别表示t时刻i节点和j节点的电压幅值；Where lv(vi,t) is the real-time voltage barrier function value of node i at time t; N is the number of network nodes; _Nm is the set of network nodes;

is the network branch set; r _ij and x _ij represent the resistance and reactance of the branch between nodes i and j respectively; v _i,t and v _j,t represent the voltage amplitudes of nodes i and j at time t respectively;

Then, the weighted sum algorithm is used to transform the multi-objective function into an equivalent single-objective function with weighted factors. The objectives are normalized using Utopia points and Nadir points. For any subnetwork m, the weighted sum of the normalized objectives is expressed as:

式中，

表示归一化后m网络t时刻的电压偏差，

表示归一化后m网络t时刻的网络有功损耗，α和β表示归一化的系数。

In the formula,

represents the normalized voltage deviation of the m network at time t,

represents the normalized network active loss of m network at time t, and α and β represent the normalized coefficients.

Preferably, the process of establishing a spatiotemporal uncertainty model based on photovoltaic and load forecast intervals is as follows:

Before each operation period, a spatial uncertainty scenario is randomly generated within a given prediction interval through Monte Carlo sampling. Secondly, in each operation cycle, a temporal uncertainty scenario is generated through Monte Carlo sampling, and the time lag is considered through the temporal uncertainty interval. In each temporal uncertainty scenario, the voltage deviation and network loss are calculated, and the two targets are corrected by the scenario occurrence probability. The corrected normalized target is expressed as:

为随机规划中所以情形归一化值的和(m网络t时刻的平均电压偏差/网损)；

为u情形下归一化值的和；ξ_u表示u情形的发生概率。

is the sum of the normalized values of all cases in the stochastic planning (average voltage deviation/network loss at time t in network m);

is the sum of the normalized values under situation u; ξ _u represents the probability of occurrence of situation u.

Preferably, the process of improving the state-action equation from adapting to discrete actions to adapting to continuous actions is as follows:

The state-action function is as follows, which is used to indicate the current state or the comprehensive return of the state-action pair:

Where τ _i represents the history of agent i; a _{- i} = × _{j ≠ i} a _j . In order to adapt to continuous actions, the state-action function is changed to the following form:

通过蒙特卡罗抽样获得，为：Where, the Gaussian distribution based on a′ _i is represented by π _i (a′ _i |τ _i ),

Obtained through Monte Carlo sampling, it is:

Preferably, the process of adding a network solution failure penalty to improve the network solution success rate is as follows:

Add network solution failure penalty function F:

F＝-f,t _f ＜T _max

Where f is the penalty constant, which is a large positive number; _tf is the duration of successful network solution in the training set that was interrupted due to network solution failure; _Tmax is the maximum time step of each training set;

At this time, the reward for each training set that is interrupted due to network solution failure is adjusted to R _mf :

R _mf = R _m + F

Where _Rm is the normal reward value obtained before the training set is interrupted.

Preferably, a device comprises:

one or more processors;

A memory for storing one or more programs;

When one or more of the programs are executed by one or more of the processors, the one or more processors implement the data-driven microgrid voltage control method as described above.

Preferably, a storage medium comprising computer executable instructions, wherein the computer executable instructions are used to perform a data-driven microgrid voltage control method as described above when executed by a computer processor.

Beneficial effects of the present invention:

The present invention is based on a data-driven approach and a distributed architecture, and adopts a combination of centralized training and decentralized execution. An effective operation control strategy is obtained by offline training of historical data, so that real-time joint output decisions can be made online according to the real-time operating conditions of the network, thereby ensuring the reliability, economy and safety of the network during operation.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.

FIG1 is a schematic diagram of the overall network framework of the present invention;

FIG. 2 is a schematic diagram of the algorithm flow framework of the present invention.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

As shown in FIG1 , a data-driven microgrid voltage control method includes the following steps:

Step (1): Based on a distributed architecture, the microgrid is divided into multiple sub-networks that are interconnected and coupled through power flows, and an intelligent agent for internal control of each sub-network is constructed.

Considering the problems of communication and computational burden, privacy issues, etc. in the traditional centralized VVC framework, as well as the problem that the traditional local control method cannot be globally coordinated, a regional coordinated microgrid VVC framework is constructed, as shown in Figure 1. The microgrid is divided into multiple sub-networks that are interconnected and coupled through power flows based on a distributed architecture, and an intelligent agent corresponding to its internal control is constructed for each sub-network. The sub-network represented by each constructed intelligent agent only exchanges limited boundary physical information and global reward returns with its adjacent sub-networks, and collectively seeks a global optimal solution. During each operation, the trained controller implements the locally measured VVC in its corresponding sub-network.

Step (2): Construct a joint control model with the reactive output of the photovoltaic inverter as the main control and the active output of the energy storage system as the auxiliary control. The single intelligent agent corresponding to each subnetwork controls all photovoltaic inverters and energy storage devices in the subnetwork, and effectively controls the voltage of the microgrid by controlling the reactive output of the photovoltaic inverter and the active output of the energy storage system.

The inverter control model is shown in (1)-(3). The reactive power output of each inverter is also limited to a preset ratio of its apparent power capacity. A positive value indicates reactive power injection into the grid, and a negative value indicates reactive power absorption from the grid.

表示光伏输出的实时有功功率，

表示光伏逆变器输出的实时无功功率，

表示光伏逆变器的复功率，

表示t时的光伏有功功率输出边界，δ光伏逆变器无功功率容量因数。in,

Indicates the real-time active power output of photovoltaic power.

Indicates the real-time reactive power output by the photovoltaic inverter.

represents the complex power of the PV inverter,

represents the PV active power output boundary at time t, and δ represents the PV inverter reactive power capacity factor.

The PV inverter droop control scheduling model is shown in (4) and (5). The setting of the energy storage system is similar to that of the inverter. A positive value indicates that active power is injected into the grid, and a negative value indicates that active power is absorbed from the grid. The remaining power of the energy storage system is always positive.

和

表示储能系统可以发出和吸收的最小和最大有功功率，

表示在t时储能的有功功率输出值，

表示t时储能的实时容量，

表示储能的最大容量。in,

and

Indicates the minimum and maximum active power that the energy storage system can generate and absorb,

represents the active power output value of energy storage at time t,

represents the real-time capacity of energy storage at time t,

Indicates the maximum capacity of energy storage.

Photovoltaic inverters give priority to providing reactive power when needed. When the reactive power compensation capacity is insufficient, the energy storage system will take action. The reactive power output of each photovoltaic inverter is also limited to a preset proportion of its apparent power capacity. A positive value indicates the injection of reactive power into the grid, and a negative value indicates the absorption of grid reactive power. The setting of the energy storage system is similar to that of the inverter. A positive value indicates the injection of active power into the grid, and a negative value indicates the absorption of grid active power. The remaining power of the energy storage system is always positive.

Step (3): combining the V-type and U-type voltage barrier function models to construct a barrel-type voltage barrier function model;

In formula (6), the V-shaped voltage barrier function has a large gradient within the range of 0.95p.u.-1.05p.u., which can achieve better voltage conditions, but it is impossible to achieve fine adjustment within the range. In formula (7), the U-shaped voltage barrier function has a small gradient outside the range of 0.95p.u.-1.05p.u., which cannot achieve rapid entry into the safe range.

Combining the advantages of both, a barrel voltage barrier function model is proposed:

In formula (8), _va is the real-time voltage of the node; v _ref is the network voltage reference value, which is generally 1.00 pu; l _v ( _va ) is the real-time reward of the node voltage.

This model combines the advantages of V-type and U-type: on the one hand, it has a slow gradient within the safety range to obtain better voltage conditions; on the other hand, the larger gradient outside the safety range can ensure faster strategy guidance.

Step (4): Take the voltage constraint based on the total network power loss and the voltage barrier function model as the target control, use the weighted sum algorithm to weigh the voltage deviation and network power loss, establish a spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals, establish a network power flow constraint model based on the Newton-Raphson method, integrate the spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals, a joint control model with the photovoltaic inverter as the main reactive output and the energy storage system as the auxiliary active output, the control target after weighing the voltage deviation and network power loss and the network power flow constraint model, and establish a distributed VVC model of the microgrid;

Two VVC objectives are set for each subnetwork, namely voltage constraint and network active power minimization, as shown in (9). Where (10) is the voltage deviation, and (11) is the total power loss of the network at time t.

为网络支路集合，r_ij和x_ij分别表示i和j节点之间支路的电阻和电抗。v_i,t和v_j,t分别表示t时刻i节点和j节点的电压幅值。In formula (10), l _v (vi _,t ) is the real-time voltage barrier function value of node i at time t; N is the number of network nodes; _Nm is the set of network nodes; in formula (11)

is the network branch set, r _ij and x _ij represent the resistance and reactance of the branch between nodes i and j respectively. _{v i,t} and v _j,t represent the voltage amplitude of nodes i and j at time t respectively.

The classic weighted sum algorithm is used to transform the multi-objective function into an equivalent single objective function with weighted factors. Here we use Utopia points and Nadir points to normalize these objectives. For subnetwork m, the weighted sum of the normalized objectives is expressed as:

表示归一化后m网络t时刻的电压偏差，

表示归一化后m网络t时刻的网络有功损耗，α和β表示归一化的系数。In the formula,

represents the normalized voltage deviation of the m network at time t,

represents the normalized network active loss of m network at time t, and α and β represent the normalized coefficients.

Establish a spatiotemporal uncertainty model based on PV and load forecast intervals.

First, a spatial uncertainty model is established based on the prediction intervals of PV generation and load, which is then optimized at each decision time step. The time intervals in the optimized model vary with time and are related to the real-time measurements at time t.

Considering the locational variability, short-term intermittency and fluctuation of renewable energy generation and load, spatial and temporal uncertainties need to be addressed to ensure operational constraints.

Firstly, a spatial uncertainty model is established based on the predicted photovoltaic power generation time interval and load, and the possible uncertainty at t0 is limited to the prediction interval:

表示空间不确定性逆变器最大功率点/母线i有功、无功功率需求的上下限。In the formula,

It represents the upper and lower limits of the active and reactive power requirements of the spatially uncertain inverter maximum power point/bus i.

In the above spatial uncertainty model, communication delay will cause response delay of the inverter. In order to eliminate the influence of these delays, a temporal uncertainty model is established and each decision time step is optimized, which can be expressed as:

示逆变器最大功率点/母线有功功率和无功功率需求的时间不确定性上下限。这个时间不确定性模型的间隔随时间而变化，并且与t时的实时测量有关。In the formula,

The upper and lower limits of the time uncertainty of the inverter maximum power point/bus active power and reactive power demand are shown. The interval of this time uncertainty model varies with time and is related to the real-time measurement at time t.

The interval of the temporal uncertainty model varies with time and is related to the real-time measurement at time t.

The network power flow constraint model based on the Newton-Raphson method is established as shown in formula (18).

where p _i,t and q _i,t represent the active power and reactive power injected by node i at time t, respectively; vi _,t represents the voltage amplitude of node i at time t; g _ij and b _ij represent the conductance and susceptance of the bus between node i and node j; θ _ij,t represents the voltage phase angle difference between node i and node j at time t; and N is the index of all nodes in the microgrid.

The distributed VVC model of the microgrid is established by integrating the spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals, the joint control model with the photovoltaic inverter as the main reactive output and the energy storage system as the auxiliary active output, the balanced voltage deviation and network power loss control objectives, and the network flow constraint model, as shown in Equation (19).

In formula (19), M is the network partition index; T is the total duration; and _vi,t is the real-time voltage.

Step (5): Fit the distributed VVC model of the microgrid to a partially observable Markov decision POMG model by improving the state-action equation from adapting to discrete actions to adapting to continuous actions to adapt to real-time control requirements;

和

分别表示节点实时电压、节点连接负载的有功消耗、节点连接负载的无功消耗、实时光伏注入电网的有功功率和实时储能剩余能量；连续的动作空间包含

和

分别表示逆变器无功输出功率和储能有功输出功率。First, set the observation space to contain v _i,t ,

and

They represent the real-time node voltage, the active power consumption of the node connected load, the reactive power consumption of the node connected load, the real-time active power injected into the grid by photovoltaic power generation, and the real-time residual energy of energy storage. The continuous action space contains

and

They represent the inverter reactive output power and energy storage active output power respectively.

Determine that the goal of each actor is to maximize its expected benefit within the time frame T, that is,

The state action equation is improved from adapting to discrete actions to adapt to continuous actions to meet real-time control requirements.

The state-action function is used to indicate the current state or the comprehensive return of the state-action pair:

Where τ _i represents the history of agent i; a _-i = × _{j ≠ i} a _j . To accommodate continuous actions, we change it to the following form:

通过蒙特卡罗抽样近似获得，可以改写为：Where, the Gaussian distribution based on a′ _i is represented by π _i (a′ _i |τ _i ). In practical applications,

It is obtained by Monte Carlo sampling approximation and can be rewritten as:

Step (6): Convert the constructed spatiotemporal uncertainty model based on PV and load forecast intervals into a stochastic programming model, and add a network solution failure penalty to the return reward of each training set to improve the network solution success rate;

The spatial uncertainty scenario is generated by Monte Carlo sampling according to the prediction interval of equations (13) and (14). Here, the time lag is considered and the temporal uncertainty scenario is generated by Monte Carlo sampling according to equations (15) and (16). The voltage deviation and network loss are corrected by the scenario occurrence probability. _ξu is the occurrence probability of scenario u∈U, which is normalized:

为随机规划中所以情形归一化值的和(m网络t时刻的平均电压偏差/网损)。

为u情形下归一化值的和。In formula (24), U considers U time uncertainty scenarios at each decision time step, and _ξu is the occurrence probability of scenario u∈U.

It is the sum of the normalized values of all situations in the stochastic planning (average voltage deviation/network loss of m network at time t).

is the sum of the normalized values in case u.

Finally, the policy gradient function is given:

In order to reduce the problem of network solution failure during training due to the problem of power network carrying capacity, a network solution failure penalty function F is added:

F＝-f,t _f ＜T _max (25)

Where f is the penalty constant, which is a large positive number; _tf is the duration of successful network solution in the training set that was interrupted due to network solution failure; _Tmax is the maximum time step of each training set.

At this time, the reward for each training set that is interrupted due to network solution failure is adjusted to:

R _mf = R _m + F (26)

In formula (26), _Rm is the normal reward value obtained before the training set is interrupted.

Step (7): Use the multi-agent deep deterministic policy gradient MADDPG algorithm to design a reasonable algorithm process and apply it to microgrids.

The specific algorithm flow is shown in Figure 2. During initialization, the capacity of the energy storage system is set to 50% of the total capacity. If the energy storage action is used, the same action value as the photovoltaic is adopted. For each training set, the buffer stores PV and load data for 480 time steps (i.e., 1 day). In addition, the reward is obtained based on the calculation results of the PandaPower software package based on the buffer data. Before providing feedback to the agent, the received state will be assigned to the observation value according to the area of each agent. Each agent only receives local observations and global rewards before making the next decision. The above process will be repeated until the end of the single training set.

After each set of actions is completed, a reward will be obtained. A training contains multiple training sets, and the number of sets required for training is manually set as needed. Behavior changes and policy model preservation occur every 40 training sets, and the target is updated every 120 sets. Before each policy model update, 10 sets of tests will be performed. The test data is based on the total data sample, and the average value of the test will be used to evaluate the effectiveness of the policy.

Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by the processor when it is run. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims (10)

1. A data-driven microgrid voltage control method, characterized in that the method comprises the following steps: Based on the distributed architecture, the microgrid is divided into multiple sub-networks that are interconnected and coupled through power flow, and an intelligent agent for the internal control of each sub-network is constructed; Construct a joint control model with the reactive output of photovoltaic inverter as the main function and the active output of energy storage system as the auxiliary function. A single intelligent agent corresponding to each subnetwork controls all photovoltaic inverters and energy storage devices in the subnetwork, and effectively controls the voltage of the microgrid by controlling the reactive output of photovoltaic inverter and the active output of energy storage system. Combining V-type and U-type voltage barrier function models, a barrel-type voltage barrier function model is constructed; The voltage constraint based on the total network power loss and voltage barrier function model is used as the target control, and the voltage deviation and network power loss are weighed using the weighted sum algorithm. A spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals is established. A network flow constraint model based on the Newton-Raphson method is established. A spatiotemporal uncertainty model based on the photovoltaic and load prediction intervals is integrated. A joint control model with the photovoltaic inverter as the main reactive output and the energy storage system as the auxiliary active output is established. The control target after weighing the voltage deviation and network power loss and the network flow constraint model are established to establish a distributed VVC model of the microgrid. The distributed VVC model of the microgrid is fitted into a partially observable Markov decision POMG model, and the state-action equation is improved from adapting to discrete actions to adapting to continuous actions to meet the real-time control requirements; The constructed spatiotemporal uncertainty model based on PV and load forecast intervals is transformed into a stochastic programming model, and a network solution failure penalty is added to the return reward of each training set to improve the network solution success rate; Using the multi-agent deep deterministic policy gradient MADDPG algorithm, the algorithm process is designed and applied to microgrids. 2. A data-driven microgrid voltage control method according to claim 1, characterized in that the distributed architecture does not require a central coordinator to collect all the information generated in the network, and needs to consider global coordination issues; in distributed optimization, the subnetwork represented by each constructed intelligent agent only exchanges limited boundary physical information and global reward returns with its adjacent subnetworks, collectively seeking a global optimal solution, and during each operation, the trained controller implements locally measured VVC in the corresponding subnetwork. 3. A data-driven microgrid voltage control method according to claim 1, characterized in that the joint control model with the photovoltaic inverter as the main reactive output and the energy storage system as the auxiliary active output is as follows:

Indicates the real-time active power output of photovoltaic power.

Indicates the real-time reactive power output by the photovoltaic inverter.

represents the complex power of the PV inverter,

represents the photovoltaic active power output boundary at time t, δ is the photovoltaic inverter reactive power capacity factor,

and

Indicates the minimum and maximum active power that the energy storage system can generate and absorb,

represents the active power output value of energy storage at time t,

represents the real-time capacity of energy storage at time t,

Indicates the maximum capacity of energy storage; Photovoltaic inverters give priority to providing reactive power when needed. When the reactive power compensation capacity is insufficient, the energy storage system will take action. The reactive power output of each photovoltaic inverter is also limited to a preset proportion of its apparent power capacity. A positive value indicates the injection of reactive power into the grid, and a negative value indicates the absorption of grid reactive power. The setting of the energy storage system is similar to that of the inverter. A positive value indicates the injection of active power into the grid, and a negative value indicates the absorption of grid active power. The remaining power of the energy storage system is always positive. 4. A data-driven microgrid voltage control method according to claim 1, characterized in that the bucket-type voltage barrier function model is as follows:

In the formula, _va is the real-time voltage of the node; v _ref is the network voltage reference value, which is 1.00 pu; l _v ( _va ) is the real-time reward of the node voltage; The voltage barrier function model combines the advantages of the V-type and the U-type: on the one hand, it has a slow gradient within the safety range to obtain better voltage conditions; on the other hand, the larger gradient outside the safety range ensures faster strategy guidance. 5. A data-driven microgrid voltage control method according to claim 1, characterized in that the control target after weighing the voltage deviation and network power loss is:

Where l _v (v _i,t ) is the real-time voltage barrier function value of node i at time t; N is the number of network nodes; N _m is the set of network nodes;

is the network branch set; r _ij and x _ij represent the resistance and reactance of the branch between nodes i and j respectively; v _i,t and v _j,t represent the voltage amplitudes of nodes i and j at time t respectively; Then, the weighted sum algorithm is used to transform the multi-objective function into an equivalent single-objective function with weighted factors. The objectives are normalized using Utopia points and Nadir points. For any subnetwork m, the weighted sum of the normalized objectives is expressed as:

In the formula,

represents the normalized voltage deviation of the m network at time t,

represents the normalized network active loss of m network at time t, and α and β represent the normalized coefficients. 6. A data-driven microgrid voltage control method according to claim 1, characterized in that the process of establishing a spatiotemporal uncertainty model based on photovoltaic and load prediction intervals is as follows: Before each operation period, a spatial uncertainty scenario is randomly generated within a given prediction interval through Monte Carlo sampling. Secondly, in each operation cycle, a temporal uncertainty scenario is generated through Monte Carlo sampling, and the time lag is considered through the temporal uncertainty interval. Under each temporal uncertainty scenario, the voltage deviation and network loss are calculated, and the two targets are corrected by the scenario occurrence probability. The corrected normalized target is expressed as:

represents the normalized average voltage deviation/network power loss under scenario u;

represents the normalized average voltage deviation/network power loss of network m at time t, that is, the sum of the normalized values in all cases in the stochastic planning; _ξu represents the probability of occurrence of scenario u. 7. A data-driven microgrid voltage control method according to claim 1, characterized in that the process of improving the state action equation from adapting to discrete actions to adapting to continuous actions is as follows: The state-action function is as follows, which is used to indicate the current state or the comprehensive return of the state-action pair:

Where τ _i represents the history of agent i; a _{- i} = × _{j ≠ i} a _j . In order to adapt to continuous actions, the state-action function is changed to the following form:

Where, the Gaussian distribution based on a′ _i is represented by π _i (a′ _i |τ _i ),

Obtained through Monte Carlo sampling, it is:

8. A data-driven microgrid voltage control method according to claim 1, characterized in that the process of adding a network solution failure penalty to improve the network solution success rate is as follows: Add network solution failure penalty function F: F＝-f,t _f ＜T _max Where f is the penalty constant, which is a large positive number; _tf is the duration of successful network solution in the training set that was interrupted due to network solution failure; _Tmax is the maximum time step of each training set; At this time, the reward for each training set that is interrupted due to network solution failure is adjusted to R _mf : R _mf = R _m + F Where _Rm is the normal reward value obtained before the training set is interrupted. 9. A device, comprising: one or more processors; A memory for storing one or more programs; When one or more of the programs are executed by one or more of the processors, one or more of the processors implement a data-driven microgrid voltage control method as described in any one of claims 1-8. 10. A storage medium comprising computer executable instructions, wherein the computer executable instructions are used to execute a data-driven microgrid voltage control method as described in any one of claims 1 to 8 when executed by a computer processor.

Images