Distributed photovoltaic autonomous control method, system and equipment for power distribution network and medium

Abstract

The invention relates to the technical field of novel power system control, and provides a power distribution network distributed photovoltaic autonomous control method, system and device and a medium, and the method comprises the steps: building a voltage control optimization model according to a linear power flow model of a target power distribution network by taking the maximization of photovoltaic consumption and the minimization of reactive power regulation cost as optimization targets; performing multi-scene time sequence simulation and node voltage out-of-limit monitoring according to the linearized power flow model, and when the node voltage is out of limit, obtaining a photovoltaic control strategy sample according to the voltage control optimization model; and training a preset neural network model to obtain a photovoltaic local control model based on a plurality of photovoltaic node control strategy training sets obtained by grouping the photovoltaic control strategy samples, and performing power output control on the photovoltaic inverter according to the photovoltaic local control model. According to the invention, only photovoltaic local measurement information is used to realize voltage efficient autonomy and photovoltaic absorption effects approaching global optimum, and the safety, stability and economical efficiency of high-proportion new energy power distribution network operation are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of novel power system control technology, and in particular to a method, system, equipment and medium for autonomous control of distributed photovoltaic power distribution networks. Background Technology

[0002] With the increasing penetration of distributed photovoltaic (PV) power in distribution networks, the randomness and volatility of PV output pose serious challenges to the safe and stable operation of these networks, particularly the frequent occurrence of voltage overruns at PV grid connection points and backflow issues. Therefore, efficient and precise control of distributed PV inverters is crucial for maintaining voltage stability in the distribution network.

[0003] Currently, distributed photovoltaic (PV) control is mainly divided into two categories: centralized control and local control. Centralized control is usually based on the Optimal Power Flow (OPF) algorithm. The distribution network center management system collects measurement data from the entire network, calculates the globally optimal command, and then sends it to each PV inverter for voltage regulation. Local control (such as traditional voltage-reactive power droop control) relies entirely on local measurement information for feedback regulation of PV inverters. However, although centralized control can achieve system-level global optimization, its nonlinear OPF calculation complexity is high and it is difficult to adapt to the millisecond-level rapid fluctuations in PV output. It also requires a high-bandwidth, low-latency communication network, which has high network construction and maintenance costs, and long-distance communication is susceptible to delays, packet loss, or interruptions, resulting in low reliability. Local control, on the other hand, has the advantages of no communication required, fast response speed, and simple deployment. However, due to the lack of global system information, it can only obtain local suboptimal solutions and cannot take into account the coordinated optimization of the entire network. This can lead to over-regulation or under-regulation of individual distributed PV units due to lack of coordination, making it difficult to maximize PV consumption while ensuring voltage safety, resulting in poor economic efficiency. Therefore, there is an urgent need to provide a distributed photovoltaic control method that can achieve global collaborative optimization and autonomous voltage management under conditions without communication. Summary of the Invention

[0004] The purpose of this invention is to provide a distributed photovoltaic (PV) autonomous control method for distribution networks. By using a centralized global collaborative control strategy obtained through offline learning driven by optimal power flow data, a sample set of local autonomous control strategies for each PV node is constructed to train the PV local control model. This enables local online autonomous control of the operating PV nodes. It can achieve near-global optimal voltage efficiency and PV absorption effect by using only distributed PV local measurement information without relying on real-time communication networks, effectively improving the safety, stability and economy of high-proportion renewable energy distribution networks.

[0005] To achieve the above objectives, it is necessary to provide a method, system, equipment, and medium for autonomous control of distributed photovoltaic power in power distribution networks.

[0006] In a first aspect, embodiments of the present invention provide a method for autonomous control of distributed photovoltaic power in a distribution network, the method comprising: A linearized power flow model of the target distribution network is constructed, and a voltage control optimization model is constructed based on the linearized power flow model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs. The target distribution network is simulated in multiple scenarios based on the linear power flow model. When a node voltage exceeds the limit each time it is detected, global control optimization is performed based on the voltage control optimization model, and corresponding photovoltaic control strategy samples are collected. All the photovoltaic control strategy samples are grouped according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets; Based on the training set of each photovoltaic node control strategy, the preset neural network model is trained to obtain the corresponding photovoltaic local control model; The corresponding power control commands are obtained in real time according to the photovoltaic local control model of each photovoltaic node, and the power output of the corresponding photovoltaic inverter is controlled according to the power control commands.

[0007] Furthermore, the step of constructing the linearized power flow model of the target distribution network includes: Based on the topology of the target distribution network, a corresponding distribution network power flow model is constructed based on node power balance analysis; the distribution network power flow model includes photovoltaic node power balance equations, load node power balance equations, and node voltage amplitude constraints; Based on a preset distribution network benchmark operating point, the sensitivity coefficient of the voltage amplitude of each node in the target distribution network relative to the injected power of each photovoltaic node is obtained. Based on each sensitivity coefficient, the power flow model of the distribution network is linearized according to Taylor series expansion to obtain the linearized power flow model. The linearized power flow model is a linear model that describes the relationship between the node voltage change and the photovoltaic power adjustment.

[0008] Furthermore, the steps for constructing a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs include: Based on the absolute values ​​of active and reactive power of each photovoltaic node, an optimization objective function is constructed with the goal of maximizing photovoltaic absorption and minimizing reactive power regulation costs. The voltage control optimization model is obtained based on the optimization objective function and the preset optimization constraints; the preset optimization constraints include photovoltaic inverter capacity limit constraints, voltage safety constraints, and photovoltaic power adjustable range constraints.

[0009] Furthermore, the step of performing multi-scenario time-series simulation of the target distribution network based on the linearized power flow model includes: The historical runtime sequence data of the target distribution network is obtained, and a time-series simulation environment is constructed using the historical runtime sequence data as time-varying boundary conditions. The historical runtime sequence data includes the net active power injection and net reactive power injection of all nodes at each discrete time point. Based on the timing simulation environment and the linearized power flow model, power flow recursion calculation is performed based on a preset time step, and it is determined whether the voltage amplitude of each node at each time step exceeds the preset voltage amplitude range.

[0010] Furthermore, the photovoltaic control strategy sample includes control strategies for each photovoltaic node; the control strategy includes electrical characteristic data and corresponding optimal power setpoints; The step of performing global control optimization based on the voltage control optimization model and collecting corresponding photovoltaic control strategy samples each time a node voltage over-limit is detected includes: The node load data, photovoltaic foundation output data, and photovoltaic node electrical characteristic data at the moment when the voltage exceeds the limit are obtained respectively; the photovoltaic node electrical characteristic data includes the actual voltage amplitude, actual injected active power, and actual injected reactive power of each photovoltaic node; Based on the node load data and the photovoltaic foundation output data, the voltage control optimization model is solved to obtain the optimal power setting value for each photovoltaic node at the moment when the corresponding node voltage exceeds the limit; the optimal power setting value includes the optimal active power setting value and the optimal reactive power setting value. The optimal power setting value of each photovoltaic node and the electrical characteristic data of the photovoltaic node at the time when the voltage of each node exceeds the limit are summarized to obtain the photovoltaic control strategy sample.

[0011] Furthermore, the step of grouping all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets includes: Based on the principle of grouping control strategies for the same photovoltaic node into a group, all photovoltaic control strategy samples are classified to obtain multiple photovoltaic node control strategy sample sets. The electrical feature data of each control strategy sample in the control strategy sample set of each photovoltaic node are spliced ​​together to obtain the corresponding electrical feature vector sample, and the corresponding optimal power set value is used to generate the label vector of the electrical feature vector sample. The corresponding photovoltaic node control strategy training set is generated by summarizing the electrical feature vector samples and label vectors corresponding to each photovoltaic node control strategy sample set.

[0012] Furthermore, the step of obtaining the corresponding power control command in real time based on the photovoltaic local control model of each photovoltaic node includes: Local measured electrical characteristic data of each photovoltaic node are acquired, and the local measured electrical characteristic data are spliced ​​to obtain the corresponding real-time electrical characteristic vector; the local measured electrical characteristic data includes voltage amplitude measurement value, actual photovoltaic active power output value and actual photovoltaic reactive power output value; The real-time electrical feature vectors of each photovoltaic node are input into the corresponding photovoltaic local control model to predict the output power and obtain the power control command; the power control command includes the target active power setpoint and the target reactive power setpoint.

[0013] Secondly, embodiments of the present invention provide a distributed photovoltaic autonomous control system for a power distribution network, the system comprising: The model building module is used to build a linearized power flow model of the target distribution network, and based on the linearized power flow model, to build a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs. The sample acquisition module is used to perform multi-scenario time-series simulation of the target distribution network according to the linear power flow model, and to perform global control optimization according to the voltage control optimization model each time a node voltage over-limit is detected, thereby acquiring the corresponding photovoltaic control strategy sample. The training set construction module is used to group all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets; The model training module is used to train the preset neural network model according to the training set of the control strategy of each photovoltaic node, so as to obtain the corresponding photovoltaic local control model. The photovoltaic control module is used to obtain the corresponding power control commands in real time according to the photovoltaic local control model of each photovoltaic node, and to control the power output of the corresponding photovoltaic inverter according to the power control commands.

[0014] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0015] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0016] This invention provides a method, system, computer equipment, and storage medium for autonomous control of distributed photovoltaic (PV) power in a distribution network. The method constructs a linearized power flow model of the target distribution network. Based on this model, a voltage control optimization model is built with the optimization objectives of maximizing PV absorption and minimizing reactive power regulation costs. Multi-scenario time-series simulations are performed on the target distribution network using the linearized power flow model. Each time a node voltage exceeds its limit, global control optimization is performed based on the voltage control optimization model to collect corresponding PV control strategy samples. All PV control strategy samples are grouped according to PV nodes to obtain multiple PV node control strategy training sets. Based on each PV node control strategy sample set, a preset neural network model is trained to obtain a corresponding PV local control model. Power control commands are obtained in real time based on the PV local control models of each PV node, and power output control is performed on the corresponding PV inverters according to these commands. Compared with existing technologies, this distributed photovoltaic autonomous control method for distribution networks constructs a sample set of local autonomous control strategies for each photovoltaic node by acquiring a centralized global collaborative control strategy based on offline learning driven by optimal power flow data. This sample set is used to train the photovoltaic local control model and achieve local online autonomous control of the operating photovoltaic nodes. It can achieve near-global optimal voltage efficiency and photovoltaic absorption effect by using only distributed photovoltaic local measurement information without relying on real-time communication networks. This effectively improves the safety, stability and economy of high-proportion renewable energy distribution networks while reducing the construction and maintenance costs of communication networks. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the autonomous control method for distributed photovoltaic power grids in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the deep residual network in an embodiment of the present invention; Figure 3 This is a schematic diagram of the training process of the photovoltaic local control model in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of the distributed photovoltaic autonomous control system for the power distribution network in an embodiment of the present invention; Figure 5 This is an internal structural diagram of the computer device in an embodiment of the present invention; The attached figures are labeled as follows: 1. Model building module; 2. Sample acquisition module; 3. Training set building module; 4. Model training module; 5. Photovoltaic control module. Detailed Implementation

[0018] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. Obviously, the embodiments described below are only part of the embodiments of this invention and are used to illustrate the invention, but are not intended to limit the scope of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] The distributed photovoltaic (PV) autonomous control method for distribution networks provided by this invention can be understood as a method for efficient autonomous control of distributed PV node voltages that approaches the global optimum, based on the current application status of centralized control which relies heavily on real-time communication networks, has high construction and maintenance costs, and large computational delays, as well as the lack of global system information and difficulty in achieving network-wide collaborative optimization in existing local control systems. The following embodiments will provide a detailed description of the distributed PV autonomous control method for distribution networks of this invention.

[0020] In one embodiment, such as Figure 1 As shown, a method for autonomous control of distributed photovoltaic power in a distribution network is provided, including the following steps: S11. Construct a linearized power flow model of the target distribution network, and based on the linearized power flow model, construct a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs; wherein, the target distribution network can be understood as a new type of power system distribution network that includes multiple distributed photovoltaic nodes in practical applications and requires voltage governance optimization, without specific limitations here.

[0021] The linearized power flow model in this embodiment can be understood as an AC power flow model based on the physical characteristics of the AC power system, the distribution network topology and parameters, and Kirchhoff's laws and power balance equations. It is easy to solve, meets real-time control requirements, and is suitable for time-series power flow analysis and operation optimization. Specifically, the steps for constructing the linearized power flow model of the target distribution network include: Based on the topology of the target distribution network, a corresponding distribution network power flow model is constructed based on node power balance analysis. This model can be understood as a power flow data model describing the coupling relationship between node voltage and photovoltaic (PV) injected power, including PV node power balance equations, load node power balance equations, and node voltage magnitude constraints. In practical applications, the construction process of the distribution network power flow model is as follows: Assuming the target distribution network is a system containing N+1 nodes, the node set is defined as follows: The set of distributed photovoltaic access nodes (referred to as photovoltaic nodes) is defined as follows: ,and Node 0 represents the substation (balance node), and the remaining nodes are distribution nodes.

[0022] For any node in the target distribution network Based on the physical characteristics of AC power systems, complex power balance equations can be constructed; among them, for any photovoltaic node (i.e. For the node-injected apparent power, the following equation is satisfied: In the formula, t Represents a discrete time step; j Represents the imaginary unit; and Representing time respectively t Photovoltaic nodes n Complex voltage phasors and complex current phasors; express The conjugate of complex numbers; and Representing time respectively t Photovoltaic nodes n The active and reactive power generated by the photovoltaic inverter; and Representing time respectively t Photovoltaic nodes n The original active and reactive loads are located.

[0023] For any ordinary load node that is not connected to photovoltaics ( and The node-injected apparent power satisfies the following equation: In the formula, and Representing time respectively t ordinary load nodes m Complex voltage phasors and complex current phasors; express The conjugate of complex numbers; and Representing time respectively t ordinary load nodes m Active and reactive loads; To ensure the safe and stable operation of the target distribution network, any node The voltage amplitude must meet the following safety range constraints: In the formula, Indicates time t node i Complex voltage phasors; Indicates time t node i The voltage amplitude; and These represent the lower limit of the minimum voltage amplitude and the upper limit of the maximum voltage amplitude that the target distribution network is allowed to operate, respectively.

[0024] Considering that the power flow model of the distribution network constructed by the above method is a nonlinear model, the solution complexity is high in practical applications and it is difficult to meet the real-time control requirements. Therefore, the following linearization method based on voltage sensitivity is adopted to linearize and reconstruct the power flow model of the distribution network.

[0025] Based on a preset distribution network reference operating point, the sensitivity coefficients of the voltage amplitude of each node in the target distribution network relative to the injected power of each photovoltaic node are obtained. Then, based on each sensitivity coefficient, the power flow model of the distribution network is linearized using Taylor series expansion to obtain the linearized power flow model. The linearized power flow model is a linear model describing the relationship between node voltage changes and photovoltaic power adjustments. The preset distribution network reference operating point can be understood as the operating status information of the target distribution network during stable operation at a specific time or under specific operating conditions (including the voltage amplitude and phase angle of each node, active power, reactive power, and current flowing through each branch, as well as the tap positions of each transformer and the switching status of each capacitor bank, etc.). Specific methods for obtaining this information can be found in relevant technologies and will not be detailed here.

[0026] In practical applications, each sensitivity coefficient is obtained by inverting the Jacobian matrix in the Newton-Raphson power flow calculation. Specifically, at the preset distribution network reference operating point, the Jacobian matrix J of the power flow equation describes the power change. ( and The linear relationship between the changes in active power and reactive power (represented by the changes in active and reactive power) and the changes in voltage amplitude is obtained by inverting the Jacobian matrix and extracting the submatrix elements corresponding to the changes in voltage amplitude and the changes in active and reactive power injection. In the formula, For any node in the target distribution network i The voltage amplitude; and Photovoltaic nodes n The injected active power and injected reactive power; and They are nodes i Voltage amplitude relative to photovoltaic node n The sensitivity coefficients of injected active power and injected reactive power; This is the symbol for partial differentials.

[0027] Based on the sensitivity coefficients calculated above, the power flow model of the distribution network can be transformed into a description of node voltage variations by using Taylor series expansion and neglecting higher-order terms. The linear equation relating the photovoltaic power adjustment amount is specifically expressed as follows: in, It is a collection of distributed photovoltaic nodes; and Photovoltaic nodes n The active power adjustment and reactive power adjustment of the photovoltaic inverter.

[0028] In this embodiment, a linearized power flow model is constructed based on the sensitivity coefficient, which can effectively improve the efficiency and reliability of generating photovoltaic control strategy samples based on multi-scenario time-series simulation. It should be noted that the above-mentioned methods for calculating the sensitivity coefficient and linearizing through Taylor series expansion can be implemented with reference to existing technologies. In addition, in practical applications, the method for linearizing the power flow model of the distribution network can also adopt existing linear approximation methods such as LinDistFlow to linearize the nonlinear power flow equations to form linear equality constraints, which will not be described in detail here.

[0029] The voltage control optimization model in this embodiment can be understood as a solution model for a global collaborative voltage governance optimization strategy driven by optimal power flow data. To improve photovoltaic absorption capacity while achieving efficient collaborative voltage autonomy, this embodiment preferably constructs a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs. Specifically, the steps for constructing the voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs include: Based on the absolute values ​​of active and reactive power of each photovoltaic node, an optimization objective function is constructed with the goal of maximizing photovoltaic power absorption and minimizing reactive power regulation costs; the optimization objective function can be expressed as: in, For photovoltaic nodes n The active power, the negative sign in front indicates that the optimization direction is to maximize active power generation in order to reduce curtailment of solar power; For processing the absolute value of reactive power The non-negative auxiliary variable introduced by this non-linear term satisfies and The aim is to minimize the losses caused by reactive power support. and These are the weighted cost coefficients for active power and reactive power, which can be determined according to actual application requirements.

[0030] Based on the optimization objective function and preset optimization constraints, the voltage control optimization model is obtained; wherein, the preset optimization constraints can be understood as constraints set in the actual voltage control optimization process to ensure the feasibility of optimization and the safety of the system. Preferably, the preset optimization constraints include the following photovoltaic inverter capacity limit constraints, voltage safety constraints, and photovoltaic power adjustable range constraints: 1) Photovoltaic inverter capacity constraint: Considering that the apparent power of the photovoltaic inverter is limited by the rated capacity (i.e., the sum of the squares of active power and reactive power must not exceed the square of the rated capacity), and that this nonlinear constraint is geometrically represented as a circular domain, this embodiment uses a polygon approximation linearization method to process it. On the active-reactive plane, a regular polygon inscribed or circumscribed to the capacity circular domain is constructed, transforming the circular domain constraint into a set of linear inequality constraints. That is, in practical applications, the linear combination value of the active power and reactive power of the photovoltaic node must not exceed the boundary threshold of the polygon in any direction, thereby ensuring that the inverter is not overloaded.

[0031] 2) Voltage safety constraints are linear inequality constraints on voltage safety constructed using the voltage sensitivity coefficients calculated above. The specific construction logic is as follows: for each node in the target distribution network... i The predicted voltage amplitude is defined as the sum of the measured voltage amplitude at the current moment and the predicted voltage change caused by the power adjustment of all photovoltaic nodes. To ensure safety, this predicted voltage amplitude must be constrained to the minimum voltage amplitude allowed by the target distribution network. and maximum voltage amplitude In practical applications, this constraint forces the optimizer to ensure that the adjusted target grid voltage does not exceed the limit when calculating the photovoltaic power adjustment. If the current measured voltage has exceeded the limit, a sufficient reverse voltage change must be generated to pull it back to the safe range.

[0032] 3) The adjustable range constraint of photovoltaic power can be understood as the physical boundary of photovoltaic output. For the output active power, its value must be between zero and the current maximum available photovoltaic power (MPPT power), and it must not be fed back and must not exceed the power generation capacity under the current irradiance conditions. For the output reactive power, its value must be between the minimum reactive power value and the maximum reactive power value allowed by the inverter.

[0033] The voltage control optimization model constructed in this embodiment can quickly smooth out voltage fluctuations and prevent voltage over-limits while achieving global collaborative voltage governance optimization based on optimal power flow data, and can also improve the photovoltaic absorption capacity of the distribution network.

[0034] S12. Perform multi-scenario time-series simulation on the target distribution network according to the linearized power flow model, and when a node voltage over-limit is detected each time, perform global control optimization according to the voltage control optimization model to collect the corresponding photovoltaic control strategy sample; wherein, multi-scenario time-series simulation can be understood as simulating the long-term dynamic operation characteristics of the target distribution network under different seasons, different weather types and load fluctuations based on the real historical operation data of the target distribution network, and performing a distribution network operation simulation process of continuous power flow scanning and voltage over-limit monitoring in the simulation environment; specifically, the step of performing multi-scenario time-series simulation on the target distribution network according to the linearized power flow model includes: Historical operating sequence data of the target distribution network is acquired, and a time-series simulation environment is constructed using this historical operating sequence data as time-varying boundary conditions. The historical operating sequence data can be understood as the actual operating data of the target distribution network used to construct a long-term continuous simulation environment. The data duration can be determined based on actual application requirements, and operating data from the target distribution network over a continuous period of three years or longer can be selected as the basis for analysis. Specifically, the historical operating sequence data includes the net active power injection (i.e., the difference between the actual photovoltaic power generation and the active load at that node) and the net reactive power injection (i.e., the difference between the reactive power output and the reactive load at that node) of all nodes at each discrete time point: For a "photovoltaic node," the net injection (net active power injection / net reactive power injection) can be expressed as "actual photovoltaic power - load power"; for a "normal load node," its actual photovoltaic power is considered to be 0, and the corresponding net injection can be expressed as "0 - load power." The time-series simulation environment constructed in this embodiment can be understood as inputting historical runtime time-series data as time-varying boundary conditions into the distribution network simulation model to construct a continuous time-series simulation environment with a specific time step (e.g., 15 minutes), which can completely simulate the long-term dynamic operating characteristics of the target distribution network under different state conditions. It should be noted that the distribution network simulation model is constructed based on existing distribution network simulation technology, which will not be described in detail here.

[0035] Based on the time-series simulation environment and the linearized power flow model, power flow recursion calculation is performed based on a preset time step, and it is determined whether the voltage amplitude of each node exceeds the preset voltage amplitude range at each time step. The power flow recursion calculation based on the preset time step can be understood as performing power flow propagation calculations sequentially based on the linearized power flow model using the Newton-Raphson method at each time step. Specific power flow propagation calculations can be directly implemented using existing technologies and will not be detailed here. The preset voltage amplitude range can be determined based on the allowable safe voltage range of the target distribution network. After completing the power flow propagation calculation at each time step, the voltage amplitude status of each node can be checked for over-limit detection. If the voltage amplitude of a node is outside the preset voltage amplitude range, it is considered that there is a voltage over-limit, and the current moment is used as the corresponding photovoltaic control strategy sample acquisition moment, immediately triggering subsequent global control optimization to obtain the corresponding photovoltaic control strategy sample. Otherwise, the current operating state is determined to be safe, and the simulation proceeds directly to the next time step.

[0036] In this embodiment, the photovoltaic control strategy sample can be understood as the global voltage collaborative governance optimization strategy data executed on all photovoltaic nodes when the target distribution network experiences voltage overruns. This includes the control strategies of each photovoltaic node, and each control strategy includes electrical characteristic data and the corresponding optimal power setpoint. To ensure that the obtained photovoltaic control strategy sample represents the system's global collaborative governance strategy for distribution network voltage overruns, and thus provides reliable data samples for subsequent training to construct a photovoltaic local control model capable of learning the characteristics of a centralized optimization global collaborative strategy, specifically, the step of executing global control optimization according to the voltage control optimization model and collecting the corresponding photovoltaic control strategy sample each time a node voltage overrun is detected includes: The node load data, photovoltaic base output data, and photovoltaic node electrical characteristic data at the moment when the voltage exceeds the limit are obtained respectively. Among them, the node load data includes the load value of all nodes in the target distribution network; the photovoltaic base output data includes the upper limit of the power output of all photovoltaic nodes in the target distribution network under the current natural conditions; the photovoltaic node electrical characteristic data includes the actual voltage amplitude, actual injected active power, and actual injected reactive power of each photovoltaic node, that is, the local electrical measurement data of each photovoltaic node in the target distribution network before optimization. It should be noted that the node load data, photovoltaic base output data, and photovoltaic node electrical characteristic data are all obtained through the distribution network simulation model.

[0037] Based on the node load data and the photovoltaic foundation output data, the voltage control optimization model is solved to obtain the optimal power setting value of each photovoltaic node at the time when the corresponding node voltage exceeds the limit. The voltage control optimization model can be solved using existing related optimization solvers, and the corresponding optimal power setting value includes the optimal active power setting value and the optimal reactive power setting value. The optimal power setting value of each photovoltaic node and the electrical characteristic data of the photovoltaic node at the time when the voltage of each node exceeds the limit are summarized to obtain the photovoltaic control strategy sample.

[0038] This embodiment uses a linearized power flow model to perform multi-scenario time-series simulations of the target distribution network. It obtains a voltage limit response sample acquisition mechanism for photovoltaic control strategies that implement global voltage collaborative control optimization strategies when node voltage exceeds limits. This mechanism can efficiently acquire high-quality voltage control optimization samples that are representative, practical, and can improve the grid's ability to cope with fluctuating and intermittent renewable energy sources under multiple scenarios and states, while fully considering the temporal changes and uncertainties of load and photovoltaic output.

[0039] S13. Group all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets; wherein, the photovoltaic node control strategy training set can be understood as a dataset constructed for training a photovoltaic local control model applicable to a single photovoltaic node; specifically, the step of grouping all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets includes: Based on the principle of grouping control strategies for the same photovoltaic node into a group, all photovoltaic control strategy samples are classified to obtain multiple photovoltaic node control strategy sample sets; that is, each photovoltaic node control strategy sample set includes control strategies for multiple scenarios and multiple states of only one photovoltaic node.

[0040] The electrical feature data of each control strategy sample in the control strategy sample set of each photovoltaic node are spliced ​​together to obtain the corresponding electrical feature vector sample, and the corresponding optimal power setting value is used to generate the label vector of the electrical feature vector sample; wherein, the label vector can be obtained by sequentially splicing the optimal active power setting value and the optimal reactive power setting value in the optimal power setting value, which will not be described in detail here.

[0041] The corresponding photovoltaic node control strategy training set is generated by summarizing the electrical feature vector samples and label vectors corresponding to each photovoltaic node control strategy sample set.

[0042] This embodiment, based on the data decoupling requirement of decentralized control, decomposes the photovoltaic node control strategies in each globally optimal photovoltaic control strategy sample to obtain a strategy training set construction mechanism with global collaborative optimization characteristics and suitable for local autonomous control of photovoltaic nodes. This mechanism can reasonably convert the centralized globally optimal strategy obtained through offline learning into local autonomous control strategies for each photovoltaic node, providing reliable training data support for the subsequent construction of a photovoltaic local control model that can obtain a near-globally optimal local autonomous control strategy based solely on local measurement information.

[0043] S14. Based on the training set of each photovoltaic node control strategy, the preset neural network model is trained to obtain the corresponding photovoltaic local control model. In principle, the preset neural network model can be a neural network that can obtain the corresponding optimal power set value based on the electrical characteristic data analysis of the photovoltaic node. In order to ensure that the photovoltaic local control model can efficiently and accurately capture the global optimization features and achieve high-precision fitting of the photovoltaic node local control strategy to the global optimal control strategy, this embodiment preferably sets the preset neural network model as a deep residual network.

[0044] The topology of the deep residual network (ResNet) in this embodiment is as follows: Figure 2 The diagram shows a deep network architecture comprising an input feature extraction layer, several stacked residual learning modules (Residual Blocks), and an output decision layer. The input feature extraction layer receives preprocessed local electrical feature vectors (including voltage amplitude, active power, and reactive power) and maps these low-dimensional features to a high-dimensional feature space through linear transformations and activation functions. The stacked residual learning modules contain multiple cascaded residual blocks, each consisting of a main path and a shortcut connection. The main path contains two sequentially connected fully connected layers and a non-linear activation function (ReLU). The shortcut connection directly adds the input of the residual block to the output. The corresponding mathematical expression is as follows: in, For the first Layer input, and These represent the corresponding weights and biases; this structure allows gradients to flow through deep networks without loss during backpropagation, effectively avoiding the gradient vanishing problem when fitting complex optimal power flow strategies.

[0045] The output decision layer maps the output of the last residual block back to the control dimension, and outputs the optimal power setpoint (including the optimal active power setpoint and the optimal reactive power setpoint) for the photovoltaic inverter.

[0046] The local control models for each photovoltaic node are trained based on the same preset neural network model. To obtain a response control strategy capable of quickly suppressing voltage fluctuations and preventing voltage exceedances, this embodiment preferably uses a sample set of control strategies for each photovoltaic node, employing methods such as... Figure 3 The refined training strategy, which includes an adaptive optimization strategy, performs refined supervised training and optimization on a pre-defined neural network model. The specific steps are as follows: 1) Data standardization preprocessing: Z-Score standardization is performed on the electrical feature vector samples in the photovoltaic node control strategy sample set to eliminate the influence of data with different dimensions on the gradient and accelerate network convergence; 2) Parameter initialization: The weight parameters in the network are initialized using the He normal distribution initialization method to adapt to the ReLU activation function and prevent neuron death in the early stage of training; 3) Constructing the regularized loss function: Construct a composite loss function that includes a prediction error term and a regularization term. The prediction error term uses mean squared error (MSE) to measure the Euclidean distance between the prediction strategy and the globally optimal label; the regularization term uses the L2 norm (Weight Decay) to penalize excessively large weight parameters and prevent overfitting of the model with limited samples; the corresponding regularization loss function is expressed as: in, The regularization coefficient is used. M For training batch size; The first in the photovoltaic node control strategy sample set j Each control strategy sample is based on the output strategy predicted by a pre-set neural network. For the globally optimal target strategy / true label, The squared L2 norm of the weight parameters.

[0047] 4) Adaptive Iterative Optimization: The AdamW optimizer is used instead of the traditional stochastic gradient descent. During training, a cosine annealing learning rate scheduling strategy is introduced. A large learning rate is used to rapidly decrease the learning rate in the early stage of training, and the learning rate is automatically decayed in the later stage to finely search for the optimal solution. The gradient of the loss function with respect to the network parameters is calculated by the backpropagation algorithm, and the weights of all residual blocks and fully connected layers are updated iteratively in batches (Mini-batch) until the loss value on the validation set no longer decreases or reaches the preset number of iterations (Epoch), thus completing the training of the photovoltaic local control model.

[0048] The photovoltaic local control model trained based on the above methods and steps, after having its weight parameters and bias parameters fixed, can be downloaded and deployed to the local embedded control unit (such as a DSP or ARM controller) of the corresponding distributed photovoltaic node in the target distribution network, for executing the corresponding autonomous control loop during the real-time operation phase of the target distribution network.

[0049] This embodiment uses a deep residual network to train and construct a photovoltaic local control model, which can more effectively extract the complex nonlinear features between local electrical feature data and global control strategy. With the regularization training strategy, it can effectively avoid the overfitting problem of model training, so that the photovoltaic local control model has strong robustness and generalization ability, and can still maintain stable control performance under unseen complex operating conditions.

[0050] S15. Obtain the corresponding power control command in real time according to the photovoltaic local control model of each photovoltaic node, and control the power output of the corresponding photovoltaic inverter according to the power control command; wherein, the power control command includes active power control command and reactive power control command; specifically, the step of obtaining the corresponding power control command in real time according to the photovoltaic local control model of each photovoltaic node includes: The local measured electrical characteristic data of each photovoltaic node is acquired, and the local measured electrical characteristic data is spliced ​​to obtain the corresponding real-time electrical characteristic vector. The local measured electrical characteristic data includes voltage amplitude measurement value, actual photovoltaic active power output value and actual photovoltaic reactive power output value, that is, the local electrical measurement data before control optimization is performed, which is collected in real time by relevant measurement equipment deployed in the actual target distribution network at a preset sampling frequency.

[0051] The real-time electrical feature vectors of each photovoltaic node are input into the corresponding photovoltaic local control model to predict the output power and obtain the power control command; the power control command includes the target active power setpoint and the target reactive power setpoint.

[0052] In practical applications, the local embedded control units of each photovoltaic node input the real-time electrical feature vectors collected in real time into the locally deployed photovoltaic local control model for forward inference. Even when communication with the distribution network central management system and other neighboring nodes is cut off, the residual mapping mechanism within the network can be used to quickly calculate the power control command that approximates the global optimum at the current moment. Finally, the underlying actuator of the photovoltaic inverter adjusts the duty cycle of the power electronic switching devices according to the power control command to quickly adjust the output power. Through the independent autonomous adjustment of all photovoltaic nodes in the target distribution network, the coordinated management of voltage exceedances in the distribution network and the maximization of photovoltaic absorption are achieved on a macro level.

[0053] This invention provides a linearized power flow model for constructing a target distribution network. Based on this model, a voltage control optimization model is constructed with the optimization objectives of maximizing photovoltaic (PV) absorption and minimizing reactive power regulation costs. Multi-scenario time-series simulations are performed on the target distribution network using the linearized power flow model. Each time a node voltage exceeds its limit, global control optimization is performed according to the voltage control optimization model to collect corresponding PV control strategy samples. All PV control strategy samples are grouped according to PV nodes to obtain multiple PV node control strategy training sets. Based on each PV node control strategy sample set, a preset neural network model is trained to obtain the corresponding PV local control model. Finally, based on the PV node's... The photovoltaic local control model acquires corresponding power control commands in real time and controls the power output of the corresponding photovoltaic inverters according to the power control commands. By constructing a sample set of local autonomous control strategies for each photovoltaic node through offline learning based on optimal power flow data, the photovoltaic local control model is trained to achieve local online autonomous control of the operating photovoltaic nodes. It can achieve near-global optimal voltage efficiency and photovoltaic absorption effect by using only distributed photovoltaic local measurement information without relying on real-time communication networks. This effectively improves the safety, stability and economy of high-proportion new energy distribution network operation, while also reducing the construction and maintenance costs of communication networks.

[0054] It should be noted that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise explicitly stated in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0055] In one embodiment, such as Figure 4 As shown, a distributed photovoltaic autonomous control system for a power distribution network is provided, the system comprising: Model building module 1 is used to build a linearized power flow model of the target distribution network, and based on the linearized power flow model, to build a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs; Sample acquisition module 2 is used to perform multi-scenario time-series simulation of the target distribution network according to the linear power flow model, and to perform global control optimization according to the voltage control optimization model each time a node voltage over-limit is detected, and to acquire the corresponding photovoltaic control strategy sample. Training set construction module 3 is used to group all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets; Model training module 4 is used to train the preset neural network model according to the training set of each photovoltaic node control strategy to obtain the corresponding photovoltaic local control model. The photovoltaic control module 5 is used to obtain the corresponding power control command in real time according to the photovoltaic local control model of each photovoltaic node, and to control the power output of the corresponding photovoltaic inverter according to the power control command.

[0056] Specific limitations regarding the distributed photovoltaic (PV) autonomous control system for distribution networks can be found in the limitations of the distributed PV autonomous control method for distribution networks described above; the corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned distributed PV autonomous control system for distribution networks can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0057] Figure 5 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 5 As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it can implement a distributed photovoltaic autonomous control method for power distribution networks. The display screen can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0058] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0059] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0060] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0061] In summary, the distributed photovoltaic autonomous control method, system, equipment, and medium provided by the embodiments of the present invention construct a sample set of local autonomous control strategies for each photovoltaic node by using a centralized global collaborative control strategy obtained through offline learning driven by optimal power flow data to train the photovoltaic local control model. This enables local online autonomous control of operating photovoltaic nodes, achieving near-global optimal voltage efficiency and photovoltaic absorption effect by using only distributed photovoltaic local measurement information without relying on real-time communication networks. This effectively improves the safety, stability, and economy of high-proportion renewable energy distribution networks while reducing the construction and maintenance costs of communication networks.

[0062] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0063] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for autonomous control of distributed photovoltaic power in a distribution network, characterized in that, The method includes: A linearized power flow model of the target distribution network is constructed, and a voltage control optimization model is constructed based on the linearized power flow model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs. The target distribution network is simulated in multiple scenarios based on the linear power flow model. When a node voltage exceeds the limit each time it is detected, global control optimization is performed based on the voltage control optimization model, and corresponding photovoltaic control strategy samples are collected. All the photovoltaic control strategy samples are grouped according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets; Based on the training set of each photovoltaic node control strategy, the preset neural network model is trained to obtain the corresponding photovoltaic local control model; The corresponding power control commands are obtained in real time according to the photovoltaic local control model of each photovoltaic node, and the power output of the corresponding photovoltaic inverter is controlled according to the power control commands.

2. The autonomous control method for distributed photovoltaic power grids as described in claim 1, characterized in that, The steps for constructing a linearized power flow model of the target distribution network include: Based on the topology of the target distribution network, a corresponding distribution network power flow model is constructed based on node power balance analysis; the distribution network power flow model includes photovoltaic node power balance equations, load node power balance equations, and node voltage amplitude constraints; Based on a preset distribution network benchmark operating point, the sensitivity coefficient of the voltage amplitude of each node in the target distribution network relative to the injected power of each photovoltaic node is obtained. Based on each sensitivity coefficient, the power flow model of the distribution network is linearized according to Taylor series expansion to obtain the linearized power flow model. The linearized power flow model is a linear model that describes the relationship between the node voltage change and the photovoltaic power adjustment.

3. The autonomous control method for distributed photovoltaic power grids as described in claim 1, characterized in that, The steps for constructing a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs include: Based on the absolute values ​​of active and reactive power of each photovoltaic node, an optimization objective function is constructed with the goal of maximizing photovoltaic absorption and minimizing reactive power regulation costs. The voltage control optimization model is obtained based on the optimization objective function and the preset optimization constraints; the preset optimization constraints include photovoltaic inverter capacity limit constraints, voltage safety constraints, and photovoltaic power adjustable range constraints.

4. The autonomous control method for distributed photovoltaic power grids as described in claim 1, characterized in that, The steps of performing multi-scenario time-series simulation of the target distribution network based on the linearized power flow model include: The historical runtime sequence data of the target distribution network is obtained, and a time-series simulation environment is constructed using the historical runtime sequence data as time-varying boundary conditions. The historical runtime sequence data includes the net active power injection and net reactive power injection of all nodes at each discrete time point. Based on the timing simulation environment and the linearized power flow model, power flow recursion calculation is performed based on a preset time step, and it is determined whether the voltage amplitude of each node at each time step exceeds the preset voltage amplitude range.

5. The autonomous control method for distributed photovoltaic power grids as described in claim 1, characterized in that, The photovoltaic control strategy sample includes the control strategy for each photovoltaic node; the control strategy includes electrical characteristic data and the corresponding optimal power setpoint. The step of performing global control optimization based on the voltage control optimization model and collecting corresponding photovoltaic control strategy samples each time a node voltage over-limit is detected includes: The node load data, photovoltaic foundation output data, and photovoltaic node electrical characteristic data at the moment when the voltage exceeds the limit are obtained respectively; the photovoltaic node electrical characteristic data includes the actual voltage amplitude, actual injected active power, and actual injected reactive power of each photovoltaic node; Based on the node load data and the photovoltaic foundation output data, the voltage control optimization model is solved to obtain the optimal power setting value for each photovoltaic node at the moment when the corresponding node voltage exceeds the limit; the optimal power setting value includes the optimal active power setting value and the optimal reactive power setting value. The optimal power setting value of each photovoltaic node and the electrical characteristic data of the photovoltaic node at the time when the voltage of each node exceeds the limit are summarized to obtain the photovoltaic control strategy sample.

6. The autonomous control method for distributed photovoltaic power grids as described in claim 5, characterized in that, The step of grouping all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets includes: Based on the principle of grouping control strategies for the same photovoltaic node into a group, all photovoltaic control strategy samples are classified to obtain multiple photovoltaic node control strategy sample sets. The electrical feature data of each control strategy sample in the control strategy sample set of each photovoltaic node are spliced ​​together to obtain the corresponding electrical feature vector sample, and the corresponding optimal power set value is used to generate the label vector of the electrical feature vector sample. The corresponding photovoltaic node control strategy training set is generated by summarizing the electrical feature vector samples and label vectors corresponding to each photovoltaic node control strategy sample set.

7. The autonomous control method for distributed photovoltaic power grids as described in claim 1, characterized in that, The step of obtaining the corresponding power control command in real time based on the photovoltaic local control model of each photovoltaic node includes: Local measured electrical characteristic data of each photovoltaic node are acquired, and the local measured electrical characteristic data are spliced ​​to obtain the corresponding real-time electrical characteristic vector; the local measured electrical characteristic data includes voltage amplitude measurement value, actual photovoltaic active power output value and actual photovoltaic reactive power output value; The real-time electrical feature vectors of each photovoltaic node are input into the corresponding photovoltaic local control model to predict the output power and obtain the power control command; the power control command includes the target active power setpoint and the target reactive power setpoint.

8. A distributed photovoltaic autonomous control system for a power distribution network, characterized in that, The system includes: The model building module is used to build a linearized power flow model of the target distribution network, and based on the linearized power flow model, to build a voltage control optimization model with the optimization objectives of maximizing photovoltaic absorption and minimizing reactive power regulation costs. The sample acquisition module is used to perform multi-scenario time-series simulation of the target distribution network according to the linear power flow model, and to perform global control optimization according to the voltage control optimization model each time a node voltage over-limit is detected, thereby acquiring the corresponding photovoltaic control strategy sample. The training set construction module is used to group all the photovoltaic control strategy samples according to photovoltaic nodes to obtain multiple photovoltaic node control strategy training sets; The model training module is used to train the preset neural network model according to the training set of the control strategy of each photovoltaic node, so as to obtain the corresponding photovoltaic local control model. The photovoltaic control module is used to obtain the corresponding power control commands in real time according to the photovoltaic local control model of each photovoltaic node, and to control the power output of the corresponding photovoltaic inverter according to the power control commands.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images