Power distribution network voltage control method based on operation state estimation

Abstract

The invention relates to a power distribution network voltage control method based on operation state estimation, and belongs to the field of power distribution network voltage control. The method comprises the following steps: firstly, establishing an optimization control model aiming at minimizing voltage deviation and reactive power consumption, and defining a state space and an action space of a system; the core of the method is that a graph attention network strategy is introduced, and attention weights of topological nodes of the power distribution network are dynamically calculated, so that an electrical coupling relationship between the nodes is captured. And then, inputting the weight information and the current and predicted states of the power grid into a pipeline model prediction controller together, carrying out rolling optimization, and outputting an optimal reactive power control instruction. According to the invention, through combination of a graph attention mechanism and model prediction control, the controller can preferentially coordinate nodes with short electrical distance, explicit input of network topology priori knowledge is not needed, and the voltage control efficiency and robustness in a photovoltaic power generation and load rapid change scene are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of distribution network voltage control, and specifically relates to a distribution network voltage control method based on operating state estimation. Background Technology

[0002] With the large-scale integration of photovoltaic (PV) power generation, its instantaneous, spatially intermittent, and non-uniform characteristics have brought unprecedented challenges to voltage regulation, resulting in severe and rapid voltage fluctuations that traditional control schemes cannot effectively address.

[0003] While centralized optimization methods can theoretically achieve global optimization, they face challenges in practical applications, such as changes in distribution network topology and insufficient response to dynamic load and generation changes. Summary of the Invention

[0004] The purpose of this invention is to overcome the problems existing in the prior art and provide a distribution network voltage control method based on operating state estimation. By combining graph attention mechanism and model predictive control, the controller can prioritize coordinating nodes with close electrical distances without explicitly inputting prior knowledge of network topology. This significantly improves the voltage control efficiency and robustness in scenarios of photovoltaic power generation and rapid load changes. This invention is applicable to distribution network voltage optimization control in scenarios of large-scale distributed photovoltaic access.

[0005] To achieve the above objectives, the technical solution of the present invention is: a distribution network voltage control method based on operating state estimation, comprising:

[0006] A voltage optimization control model for the distribution network is established, whose objective function is used to regulate reactive power to maintain the voltage within a safe range and minimize voltage deviation.

[0007] Analyze and define the state space and action space for safe operation of the distribution network;

[0008] Based on the graph attention network strategy, the dynamic attention weights of the distribution network topology nodes are updated in real time under the current state.

[0009] The dynamic attention weights, the current operating state of the distribution network, and the predicted state at the next moment are input into the pipeline model predictive controller, which outputs reactive power control quantities.

[0010] Furthermore, the objective function of the power distribution network voltage optimization control model adopts an optimization method based on a barrier function to penalize voltage deviations that deviate from the safe range.

[0011] Furthermore, the objective function is expressed as follows:

[0012]

[0013] in, Represents a node per-unit voltage, This is a differentiable barrier function used to penalize voltage deviations that deviate from the per-unit range of [0.95, 1.05]. For generator reactive power output, For reactive power penalty weight, For the number of nodes, The number of controllable generators;

[0014] The barrier function can take the form of L1 norm, L2 norm, Courant-Beltrami function, bowl function or bump function, and the choice is made according to the smoothness requirements and penalty characteristics;

[0015] Voltage control drives the system to converge to a safe operating state by optimizing the objective function, as shown in the following equation:

[0016] .

[0017] Furthermore, the state space at time t aggregates the multi-dimensional power system measurements s(t) including photovoltaic power generation, active and reactive power demand of the load, reactive power output of the generator, voltage amplitude, and phase angle, as follows:

[0018]

[0019] in, Let be the active power generated by the photovoltaic system at time t. Let be the active power demand at node t. Let t be the reactive power demand at node t. Let be the reactive power generated by the generator at time t; Indicates the next state. Represents the external power injection vector. Represents the nonlinear AC power flow equation;

[0020] The control actions in the action space are normalized to a reactive power setpoint under inverter capacity constraints, specifically through an action parameter in the range of [-1, 1]. Defined as the product of the inverter's remaining capacity, specifically as follows:

[0021]

[0022] in, The rated apparent power capacity of the inverter. This represents the active power output of the j-th photovoltaic power generation unit at time t. This represents the maximum available active power output of the j-th photovoltaic power generation unit under the current operating conditions. Its value is determined by the photovoltaic installed capacity, operating conditions, and inverter rated capacity.

[0023] Furthermore, the node weight update strategy based on graph attention network includes: performing a linear transformation encoding on local observations with layer normalization to obtain initial node features; calculating and normalizing the attention coefficients between a node and its neighboring nodes; aggregating neighboring features using normalized attention weights; concatenating the outputs of all attention heads and generating an action through a final transformation.

[0024] Furthermore, the pipeline model predictive controller takes minimizing the overall voltage deviation of the distribution network as its optimization objective and performs rolling optimization.

[0025] Furthermore, the rolling optimization needs to satisfy the power regulation constraints, response time constraints, ramp rate constraints, duration constraints, and regulation cost constraints of distributed power sources, as well as the voltage safety constraints and branch power constraints of distribution network nodes.

[0026] Furthermore, the graph attention network is a multi-head graph attention mechanism that can aggregate information from adjacent distributed power sources based on the physical topology of the power network.

[0027] Furthermore, the method also includes: after the pipeline model predictive controller obtains the optimal control sequence of the adjustable device, it issues the first instruction in the sequence and performs a smooth transition on a real-time scale.

[0028] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the steps of the method described above.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] (1) The method of the present invention combines graph attention network with pipeline model predictive control algorithm. This enables pipeline model predictive control to learn dynamic attention weights based on the physical topology of power network and prioritize coordination with electrically connected neighbors. The network topology and photovoltaic location are learned automatically through the pipeline using scene metadata, without explicit input, which reduces the complexity of actual deployment.

[0031] (2) The method of this invention proposes a model predictive control algorithm, which uses the current distribution network topology node weight as a feedback quantity when solving optimization instructions, thereby improving optimization efficiency. Attached Figure Description

[0032] Figure 1This is a schematic diagram of the specific control flow of the rolling optimization model of the present invention. Detailed Implementation

[0033] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0034] This invention provides a distribution network voltage control method based on operating state estimation, comprising:

[0035] A voltage optimization control model for the distribution network is established, whose objective function is used to regulate reactive power to maintain the voltage within a safe range and minimize voltage deviation.

[0036] Analyze and define the state space and action space for safe operation of the distribution network;

[0037] Based on the graph attention network strategy, the dynamic attention weights of the distribution network topology nodes are updated in real time under the current state.

[0038] The dynamic attention weights, the current operating state of the distribution network, and the predicted state at the next moment are input into the pipeline model predictive controller, which outputs reactive power control quantities.

[0039] The following is a detailed implementation process of the present invention.

[0040] A distribution network voltage control method based on operating state estimation includes:

[0041] 1. Establish a voltage optimization control model for the distribution network.

[0042] The core objective of voltage control is to maintain the voltage within a safe range while minimizing voltage deviation under conditions of high photovoltaic penetration and load variation by adjusting reactive power. An optimization method based on a barrier function is adopted, and the objective function is shown in equation (1):

[0043]

[0044] in, Represents a node per-unit voltage, This is a differentiable barrier function used to penalize voltage deviations that deviate from the per-unit range of [0.95, 1.05]. For generator reactive power output, For reactive power penalty weight, For the number of nodes, The number of controllable generators is given. The obstacle function can take the form of l1 norm, l2 norm, Courant-Beltrami function, bowl function, or bump function, and is selected according to the smoothness requirements and penalty characteristics. Voltage control drives the system to converge to a safe operating state by optimizing the above objective function, as shown in equation (2):

[0045]

[0046] 2. Analyze the state space and action space of the distribution network for safe operation.

[0047] At any moment System status It aggregates multi-dimensional power system measurements, including photovoltaic power generation, load demand, reactive power output, voltage amplitude, and phase angle, as shown in equation (3):

[0048]

[0049] in, The active power generated by the photovoltaic system, For the active power requirements of the node, For node reactive power requirements, The reactive power generated by the generator. The control action is normalized to the power setpoint under the inverter capacity constraint, as shown in equation (4):

[0050]

[0051] in, The inverter's rated apparent power capacity is given, with a safety factor of 1.2. Unlike the linear control model, the voltage dynamics of the distribution grid follow a nonlinear AC power flow equation, and the next state calculation is shown in equation (5):

[0052]

[0053] in, Represents the external power injection vector. This represents the nonlinear AC power flow equation.

[0054] 3. Based on a graph attention strategy, update the weights of distribution network topology nodes in real time under the current state.

[0055] The distribution network update strategy employs a multi-head graph attention mechanism, aggregating information from neighboring distributed generation sources based on the power network topology. This process comprises four phases:

[0056] 1) Local observations are encoded into initial node features through a linear transformation with layer normalization, as shown in equation (6):

[0057]

[0058] ReLU represents the rectified linear activation function, used to introduce a non-linear mapping; LayerNorm represents the layer normalization operation, used to normalize features to improve training stability; W enc b represents the learnable weight matrix during the encoding phase; enc This represents the bias vector corresponding to the weight matrix; o i This represents the local observation input vector of node i; the subscript i indicates the node number, and enc indicates the coding layer.

[0059] 2) For each attention head compute nodes with his neighbors The attention coefficients between them are shown in equations (7) and (8):

[0060]

[0061]

[0062] Where LeakyReLU represents a linear rectified activation function with a leakage slope; h i h represents the feature vector of node i. j e represents the feature vector of its neighbor node j; ij (k) Let represent the unnormalized attention coefficients between node i and its neighbor j under the k-th attention head, where the subscripts i and j represent the target node and neighbor node numbers, respectively, and the superscript (k) indicates the k-th attention head. For learnable weight matrix, For attention vectors, This indicates a splicing operation, where the neighbor set is determined from the network topology adjacency matrix A extracted from the power system line connections. .

[0063] 3) Aggregate neighbor features using normalized attention weights.

[0064] 4) Concatenate the outputs of all attention heads and generate the action through the final transformation.

[0065] 4. Input the node weight information, the current operating status of the distribution network, and the status at the next moment into the pipeline model predictive controller, and output the reactive power control quantity.

[0066] The pipeline model predictive controller set in this invention takes the overall voltage deviation of the distribution network as the optimization target, as shown in equation (9):

[0067] (9)

[0068] In the formula, V bais This is the voltage deviation value; V i V i,ref Here are the node voltages and their reference values.

[0069] This optimization objective requires rolling optimization. The model must satisfy constraints on the direction and magnitude of active and reactive power regulation, response time, ramp rate, duration, and distributed power source regulation cost during distributed generation operation. Simultaneously, each node in the distribution network must meet voltage safety constraints and branch power constraints, etc. Figure 1 As shown, the specific control flow of the rolling optimization model is as follows:

[0070] Step 1: Initialize the parameters of each control device in the distribution network, and import the daytime distributed voltage output scenario library and the intraday distribution gateway node voltage into the optimization model as constraints.

[0071] Step 2: With the goal of minimizing the overall voltage deviation of the distribution network, and with constraints on the direction and magnitude of active and reactive power regulation of distributed power sources within the distribution network, response time constraints, ramp rate constraints, duration constraints, and distributed power source regulation costs, establish a rolling optimization model, and solve for the optimal control sequence of adjustable equipment at the current moment based on the node weights of the distribution network topology at the current moment.

[0072] Step 3: Issue the first instruction in the optimal control sequence of the adjustable device.

[0073] Step 4: Issue intraday resource scheduling reference instructions for the distribution network, and smoothly transition between intraday instructions on a real-time scale.

[0074] Step 5: Use the real-time measurement data from the measurement system as the initial state for a new round of rolling optimization, return to Step 2, and execute a new round of optimization.

[0075] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the steps of the method described above.

[0076] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A distribution network voltage control method based on operating state estimation, characterized in that, include: A voltage optimization control model for the distribution network is established, whose objective function is used to regulate reactive power to maintain the voltage within a safe range and minimize voltage deviation. Analyze and define the state space and action space for safe operation of the distribution network; Based on the graph attention network strategy, the dynamic attention weights of the distribution network topology nodes are updated in real time under the current state. The dynamic attention weights, the current operating state of the distribution network, and the predicted state at the next moment are input into the pipeline model predictive controller, which outputs reactive power control quantities.

2. The distribution network voltage control method based on operating state estimation according to claim 1, characterized in that, The objective function of the distribution network voltage optimization control model adopts an optimization method based on a barrier function to penalize voltage deviations that deviate from the safe range.

3. A distribution network voltage control method based on operating state estimation according to claim 1 or 2, characterized in that, The objective function is expressed as follows: in, Represents a node per-unit voltage, This is a differentiable barrier function used to penalize voltage deviations from the per-unit range of [0.95, 1.05]. For generator reactive power output, For reactive power penalty weight, For the number of nodes, The number of controllable generators; The barrier function can take the form of L1 norm, L2 norm, Courant-Beltrami function, bowl function or bump function, and the choice is made according to the smoothness requirements and penalty characteristics; Voltage control drives the system to converge to a safe operating state by optimizing the objective function, as shown in the following equation: 。 4. The distribution network voltage control method based on operating state estimation according to claim 1, characterized in that, The state space at time t, s(t), aggregates multi-dimensional power system measurements including photovoltaic power generation, active and reactive power demand of the load, reactive power output of the generator, voltage amplitude, and phase angle. s(t) is represented as follows: in, Let be the active power generated by the photovoltaic system at time t. Let be the active power demand at node t. Let be the reactive power demand at node t. Let be the reactive power generated by the generator at time t; Indicates the next state. Represents the external power injection vector. Represents the nonlinear AC power flow equation; The control actions in the action space are normalized to a reactive power setpoint under inverter capacity constraints, specifically through an action parameter in the range of [-1, 1]. Defined as the product of the inverter's remaining capacity, specifically as follows: in, The rated apparent power capacity of the inverter. This represents the active power output of the j-th photovoltaic power generation unit at time t. This represents the maximum available active power output of the j-th photovoltaic power generation unit under the current operating conditions. Its value is determined by the photovoltaic installed capacity, operating conditions, and inverter rated capacity.

5. The distribution network voltage control method based on operating state estimation according to claim 1, characterized in that, The graph attention network-based strategy for updating node weights includes: performing a linear transformation encoding on local observations with layer normalization to obtain initial node features; calculating and normalizing the attention coefficients between a node and its neighboring nodes; aggregating neighbor features using normalized attention weights; concatenating the outputs of all attention heads and generating an action through a final transformation.

6. The distribution network voltage control method based on operating state estimation according to claim 1, characterized in that, The pipeline model predictive controller aims to minimize the overall voltage deviation of the distribution network and performs rolling optimization.

7. The distribution network voltage control method based on operating state estimation according to claim 6, characterized in that, The rolling optimization needs to satisfy the power regulation constraints, response time constraints, ramp rate constraints, duration constraints, and regulation cost constraints of distributed power sources, as well as the voltage safety constraints and branch power constraints of distribution network nodes.

8. The distribution network voltage control method based on operating state estimation according to claim 1, characterized in that, The graph attention network is a multi-head graph attention mechanism that can aggregate information from adjacent distributed power sources based on the physical topology of the power network.

9. The distribution network voltage control method based on operating state estimation according to claim 1, characterized in that, The method further includes: after the pipeline model predictive controller obtains the optimal control sequence of the adjustable equipment, it issues the first instruction in the sequence and performs a smooth transition on a real-time scale.

10. An electronic 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 program, it implements the steps of the method as described in any one of claims 1 to 9.

Images