Distribution network operation mode optimization method and system for distributed photovoltaic consumption

Abstract

The present invention discloses a method and system for optimizing the operation mode of a distribution network for distributed photovoltaic consumption. The method comprises: collecting system data, including system power data, system structure data, and meteorological data associated by timestamps; using a scene classification model to identify the scene mode corresponding to the current photovoltaic power, and calling a photovoltaic power prediction model corresponding to the identified scene mode to obtain a photovoltaic power prediction value; inputting the system power data, system structure data, and photovoltaic power prediction value of the current period as the current system state into a decision model based on reinforcement learning to obtain a distribution network topology reconstruction strategy under the current system state; monitoring the distribution network topology reconstruction strategy, and revising the distribution network topology reconstruction strategy that violates system safety constraints until the system safety constraints are met, thereby obtaining an optimized distribution network topology reconstruction strategy. The present invention can effectively improve the safety and economic efficiency of distribution network system operation.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation optimization, and in particular to a distribution network operation mode optimization method and system for distributed photovoltaic consumption. Background Art

[0002] With the explosive growth of distributed photovoltaic power generation, distribution networks have become a crucial channel for grid connection, with a large number of distributed photovoltaic plants being connected to the end of distribution lines. During periods of peak photovoltaic output, sudden increases in PV output can raise the voltage at the grid connection point, directly impacting the power quality of surrounding users and even causing distributed photovoltaic power to disconnect from the grid due to excessively high voltage at the grid connection point. Furthermore, because distributed photovoltaic power generation in high-penetration areas is significantly affected by weather, intermittent and large fluctuations can also cause significant jumps in the voltage of low-voltage distribution buses. The overall absorption capacity of the distribution network is insufficient, necessitating the construction of a large number of new energy storage and other regulating resources. However, the high economic cost hinders the sustainable development of photovoltaic power on a large scale. Distribution network operational reconfiguration technology is one key measure to address this issue. Generally, to ensure power quality and safety, distribution network architectures are designed to accommodate different operating modes. Therefore, without adding additional energy storage equipment, it is possible to improve the absorption capacity of photovoltaic power by modifying the grid architecture by operating line backup switches.

[0003] Compared to traditional distribution network systems, new power systems are more prone to problems such as line overloads and voltage over-limits during topology reconstruction. Furthermore, while adjusting the grid through line switches can directly control the system's power flow distribution and node voltages, it also increases the complexity of solving grid dispatching and operation problems. Furthermore, compared to day-ahead dispatching problems, distribution network operation reconstruction places higher demands on computational timeliness. Renewable energy output, represented by photovoltaics, is volatile and random. The impact of the system grid on operational indicators is closely related to the renewable energy and photovoltaic output at the time of operation. Promptly and quickly generating reconstruction strategies during the corresponding time period can significantly improve the economic efficiency of system operation. However, most current technologies generally struggle to achieve the required computational speed without consuming a large amount of computing power. Summary of the Invention

[0004] To address the problem that traditional algorithms have low computational efficiency and cannot perform real-time optimization solutions, the present invention provides a distribution network operation mode optimization method and system for distributed photovoltaic absorption, aiming to provide an efficient distribution network operation scheduling strategy with the goal of promoting distributed photovoltaic absorption.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution of the present invention is as follows:

[0006] According to a first aspect of the present invention, a method for optimizing the operation mode of a distribution network for distributed photovoltaic consumption is proposed. The method comprises the following steps:

[0007] Step 1: Collect system data, including power data with timestamps, system structure data, and meteorological data; the system power data includes photovoltaic power data and load data of each node;

[0008] Step 2: Using a scene classification model to identify the scene mode corresponding to the current photovoltaic power, and calling a photovoltaic power prediction model corresponding to the identified scene mode to obtain a photovoltaic power prediction value; the scene classification model and the photovoltaic power prediction model are trained by dividing the system power data using meteorological data based on clustering technology;

[0009] Step 3: Input the current system power data, system structure data, and PV power forecast value into the reinforcement learning-based decision model as the current system state to obtain the distribution network topology reconstruction strategy under the current system state;

[0010] Step 4: Monitor the distribution network topology reconstruction strategy and iteratively correct the strategy that violates the system security constraint until the system security constraint strategy is satisfied.

[0011] Furthermore, in step 1, the meteorological data includes temperature data, humidity data and air pressure data of the substation at each time. The system structure data includes the topology of the power system at each time, the status of the line switches and the impedance data of each line.

[0012] Furthermore, in step 2, the scene classification model and the photovoltaic power prediction model are trained in the following manner:

[0013] Step 21, dividing the photovoltaic power data at each moment into corresponding seasons according to the timestamp;

[0014] Step 22: select humidity, air pressure, and temperature from the meteorological dataset as characteristic variables for cluster analysis. Use the K-Means clustering algorithm to further classify the photovoltaic power data for each season into three weather modes: sunny, cloudy, and rainy. This divides the photovoltaic power data into sub-power datasets under 12 scenario modes.

[0015] Step 23 : Based on the division result, the system data is used as a data source to train the scene classification model and the photovoltaic power prediction model under each scene mode.

[0016] Furthermore, step 22 includes:

[0017] Step 221 randomly selects three initial centroids, continuously updates the centroids, and assigns the meteorological data points corresponding to each photovoltaic power data point to the nearest centroid according to the distance difference from each centroid, thus completing the clustering of the power data. Each centroid is a three-dimensional vector consisting of the average values ​​of humidity, air pressure, and temperature. The centroids and distances are calculated as follows:

[0018]

[0019] in, For meteorological data points i and centroid j The Euclidean distance between H i 、 P i and T i Represents the data points i Humidity, air pressure and temperature values; k For meteorological data points i The number of the category cluster to which it belongs, N k For the k The number of meteorological data points in each category cluster, is the new centroid coordinate; Cluster k Indicates the k The set of meteorological data points corresponding to each photovoltaic power data in the category cluster;

[0020] Step 222 : Analyze the characteristics of the meteorological data in each category cluster, and determine and mark the weather pattern corresponding to each category cluster based on the analysis results.

[0021] Furthermore, step 23 includes:

[0022] Step 231, construct the loss functions of the scene classification model and the photovoltaic power prediction model respectively, which are expressed as:

[0023]

[0024] in, N 1 and N 2 is the number of samples in the training sample set corresponding to the scene classification sub-model and the classification prediction sub-model, k is the sample category; is the loss function of the scene classification sub-model, is the first i The true category labels of samples, is the first i samples belong to the categoryk The probability of is the loss function of the photovoltaic power prediction model, and They are the first i The actual value of photovoltaic power and the predicted value of photovoltaic power corresponding to the sample.

[0025] Step 232, to minimize and As the goal, the Adam optimizer is called to update the parameters of the scene classification model and the photovoltaic power prediction model respectively using cross entropy loss and mean absolute error through gradient back propagation to realize the training of the scene classification sub-model and the classification prediction sub-model.

[0026] Furthermore, in step 3, the decision model based on reinforcement learning is trained in the following manner:

[0027] Step 31: Establish a power system virtual environment model, wherein the power system virtual environment model is used to simulate the steady state of the distribution network through distributed power flow solution based on the distribution network topology reconstruction strategy output by the decision model, and generate simulated system power data and system structure data in combination with the set photovoltaic power prediction value;

[0028] Step 32: The power system virtual environment model and the decision model are interacted, and a reinforcement learning-based neural network is trained with the goal of maximizing a reward function to obtain the decision model; the reward function is related to the photovoltaic absorption ratio index, network loss index, switching loss index, whether the power flow solution result converges, and whether the current distribution network topology reconstruction strategy meets the system safety constraints.

[0029] Furthermore, step 31 includes:

[0030] Step 311: Perform distributed power flow calculation using the following formula:

[0031]

[0032] in, 、 Branch ij The active power and reactive power, 、 Branch jk The active power and reactive power, 、 Node i The active and reactive power, and Node i and jThe square of the voltage amplitude on It is a branch road ij The square of the upper current amplitude, 、 denote the set of nodes and branches respectively, Representation node i and j Intermediate branch ij , 、 Branch Road ij The resistance and reactance, and Represent the upper and lower bounds of the variable respectively;

[0033] Step 312: Determine whether the distributed power flow solution result converges, and feedback the first negative penalty if it is determined that the power flow solution result does not converge. r pen1 ;

[0034] Step 313: Determine whether the current distribution network topology reconstruction strategy meets the system safety operation conditions, and feedback the second negative value penalty if the system safety operation conditions are violated. r pen2 .

[0035] Furthermore, in step 32, the reward function is expressed as:

[0036]

[0037] in, is the absorption ratio of distributed photovoltaics in the system, 、 is the photovoltaic power before and after optimization, is the line loss of the network, is the number of photovoltaic nodes, is the total number of system branches, is the basic reward of the reinforcement learning model, is the number of switch changes under the current strategy, 、 、 The proportional coefficients are used to adjust the weights of the photovoltaic absorption ratio index, network loss index and switching loss index in the reward. A small amount used to prevent the denominator from being too small and causing reward overflow or division by zero errors; is the total negative penalty, and r pen = r pen1+ r pen2 ; is the total reward function of the model.

[0038] Furthermore, in step 32, the reinforcement learning-based neural network is updated using a policy gradient algorithm and includes an actor network and a critic network. The actor network inputs the system state data and outputs the current distribution network reconstruction strategy. The critic network inputs the system current data and the reconstruction strategy given by the actor network and outputs the state action value to guide the training parameters. The power system executes the optimization strategy given by the actor network, returns the reward parameter, and updates its own system state. The reinforcement learning-based neural network is expressed as:

[0039]

[0040] in, For the j The switching action of each branch, For the The distribution network topology reconstruction action at each moment, is the topological state of the system, For the The total state of the system at time is the node load demand, is the predicted value of future photovoltaic power; S represents the state space, with a dimension of 1× N branch ; The loss function of the critic network is used to update the parameters of the critic network; is the system state at the next moment; 、 They are the main critic network and the main actor network respectively; For the target critic network, For the target actor network, and 、 and The structure is exactly the same. and The parameters of are obtained by performing soft parameter updates after several interactions; s and a They are the system status and distribution network topology reconstruction actions at each moment; The parameters of the main actor network are obtained by gradient ascent of the state-action value of the main critic network. is the parameter change, n The total number of samples collected and used for training in the current training batch.

[0041] Furthermore, in step 4, the correction includes:

[0042] Step 41: Determine the photovoltaic power coefficient of all nodes using the following formula:

[0043]

[0044] in, For nodes The photovoltaic power coefficient;

[0045] Step 42: Prioritize reduction of the node with the highest photovoltaic power coefficient using the following formula:

[0046]

[0047] in, The reduction coefficient is used to regulate the reduction ratio; and They are the nodes with the highest photovoltaic power coefficient PV power before and after curtailment;

[0048] Step 43: After the reduction, the state is updated and the power flow is recalculated. If there is still a violation, the photovoltaic power coefficient is recalculated and reduced again. The cycle is iterated until the distribution network topology reconstruction strategy given by the model meets the system safety constraints.

[0049] According to a second aspect of the present invention, a distribution network operation mode optimization system for distributed photovoltaic consumption using the method described in the first aspect of the present invention is provided. The system comprises:

[0050] A data acquisition module is used to collect system data, including system power data with timestamps, system structure data, and meteorological data; the system power data includes photovoltaic power data and load data of each node;

[0051] A power prediction module is configured to identify the current photovoltaic power scenario using a scene classification model and, based on the identified scenario, call a corresponding photovoltaic power prediction model to obtain a photovoltaic power prediction value; the scene classification model and the photovoltaic power prediction model are trained by dividing system power data using meteorological data based on clustering technology;

[0052] The distribution network reconstruction optimization module is used to input the current system power data, system structure data, and photovoltaic power forecast value into the decision model based on reinforcement learning as the system state, and obtain the distribution network topology reconstruction strategy under the current system state;

[0053] The decision correction module is used to monitor the distribution network topology reconstruction strategy and correct the distribution network topology reconstruction strategy that violates the system safety constraints until the system safety constraints are met, thereby obtaining an optimized distribution network topology reconstruction strategy.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. The present invention introduces a classification model to call different prediction models for photovoltaic power in different scene modes to perform targeted predictions, providing a data basis for the optimization model;

[0056] 2. Based on reinforcement learning theory, the present invention constructs a virtual environment training decision model for the distribution network system, and realizes real-time and high-speed reconstruction and optimization of the distribution network topology.

[0057] 3. This invention builds an additional layer of security to detect and correct decisions that violate system safety constraints. This reinforcement learning-based decision model can dynamically adjust grid operation, further increasing the distribution system's photovoltaic absorption capacity and improving system safety and economic efficiency.

[0058] In summary, this paper integrates classification, prediction, and reinforcement learning techniques to identify and solve the physical mechanisms and power trends of distribution networks. It then proposes an efficient distribution network operation and scheduling strategy aimed at promoting the integration of distributed photovoltaic power. To address the potential for violations of limits imposed by neural network strategies, a safety layer is constructed to ensure that all system safety constraints are met, thereby improving the economic efficiency and safety of distribution networks containing distributed photovoltaic power. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic flow chart of a method for optimizing the operation mode of a distribution network for distributed photovoltaic consumption according to the present invention;

[0060] Figure 2 This is a schematic diagram of a specific model structure of a distribution network operation mode optimization method for distributed photovoltaic consumption according to the present invention;

[0061] Figure 3 Schematic diagram of the topological structure of the optimized distribution network at different time sections;

[0062] Figure 4 The present invention is a schematic structural diagram of a distribution network operation mode optimization system for distributed photovoltaic consumption in one embodiment of the present invention.

[0063] Figures 5(a) to 5(c) are the clustering results obtained by using K-means clustering, AP clustering, and hierarchical clustering methods, respectively. DETAILED DESCRIPTION

[0064] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0065] In the first aspect of the present invention, a method for optimizing the operation mode of a distribution network for distributed photovoltaic consumption is designed. Figure 1 and 2 As shown, in one embodiment, the method includes the following steps:

[0066] Step 1: Collect system data, including system power data with time stamps, system structure data, and meteorological data; the system power data includes photovoltaic power data and load data of each node.

[0067] In this step, meteorological data includes temperature, humidity, and air pressure data at each moment; system structure data includes the current distribution network topology, the status of circuit breakers, and the impedance data of each line. The meteorological data timestamp corresponds to the power data and is used to subsequently cluster the power data into different scenarios. The load data and system structure data serve as decision-making information for power system scheduling. For power system operation, no specific distinction is made between user load and distributed photovoltaic power; therefore, the difference between the two, or net load data, is used as the power data input for the power system. Grid structure data, with the exception of circuit breaker status, are inherent system parameters and therefore do not change with changes in the scheduling strategy.

[0068] Step 2: Use the scene classification model to identify the scene pattern corresponding to the current photovoltaic power, and call the photovoltaic power prediction model corresponding to the identified scene pattern to obtain the photovoltaic power prediction value. The scene classification model and the photovoltaic power prediction model are trained by using clustering technology to divide the system power data using meteorological data.

[0069] Specifically, in step 2, the scene classification model and photovoltaic power prediction model are trained in the following way:

[0070] Step 21: Divide the photovoltaic power data at each moment into the corresponding season according to the timestamp.

[0071] In this step, since the power data is timestamped when it is collected, the power data can be divided into four seasons: spring, summer, autumn, and winter according to the timestamp.

[0072] In step 22, humidity, air pressure, and temperature are selected from the meteorological dataset as characteristic variables for cluster analysis. The photovoltaic power data of each season is further divided into three weather modes: sunny, cloudy, and rainy, using the K-Means clustering algorithm. Thus, the photovoltaic power data is divided into sub-power datasets under 12 scenario modes.

[0073] Specifically, step 22 includes:

[0074] In step 221, three initial centroids are randomly selected. The power data is clustered by continuously updating the centroids and assigning the meteorological data points corresponding to each photovoltaic power data point to the nearest centroid according to the distance difference from each centroid. Each centroid is a three-dimensional vector composed of the average values ​​of humidity, air pressure, and temperature. The calculation process of the centroid and distance is as follows:

[0075]

[0076] in, For meteorological data points i and centroid j The Euclidean distance between H i 、 P i and T i Represents the data points i Humidity, air pressure and temperature values; k is the number of the category cluster to which the meteorological data point belongs, N k For the k The number of meteorological data points in a class cluster, is the new centroid coordinate; Cluster k Indicates the k The set of meteorological data points corresponding to each photovoltaic power data in the category cluster.

[0077] It should be noted that when clustering, each centroid is set to a three-dimensional vector consisting of the average values ​​of humidity, air pressure and temperature, which is carefully considered. Specifically, for photovoltaic prediction, although irradiance is key data that directly affects the power generation of photovoltaic power stations, traditional algorithms also use this data as clustering features. However, high-precision irradiance measurement equipment is expensive, requires regular maintenance and calibration, and can only represent data within a certain range, with limited coverage. In contrast, it is easier to obtain humidity, air pressure and temperature data, and measurement equipment is relatively popular. Meteorological data can be obtained through various public channels such as weather forecasts and online meteorological platforms. The data of a single measurement point can reflect the meteorological characteristics of an area and has a wide coverage. Therefore, unlike traditional feature selection, the present invention trains a classification model through macroscopic meteorological data (i.e., humidity, air pressure and temperature, etc.), thereby reducing data acquisition costs and improving prediction efficiency while ensuring prediction accuracy.

[0078] Step 222 : Analyze the characteristics of the meteorological data in each category cluster, and determine and mark the weather pattern corresponding to each category cluster based on the analysis results.

[0079] The weather pattern corresponding to each cluster can be determined based on the characteristic values ​​of the cluster center, combined with meteorological knowledge and experience. For example, if a cluster is characterized by high humidity, low temperature, and low pressure, it can be defined as a "rainy weather pattern"; if a cluster is characterized by low humidity, high temperature, and high pressure, it can be defined as a "sunny weather pattern"; and if a cluster is characterized by high humidity, low temperature, and low pressure, it can be defined as a "cloudy weather pattern." It should be noted that the humidity, temperature, and pressure characteristics of rainy, sunny, and cloudy conditions have certain universal characteristics and regularities, but they can also vary due to various factors. Therefore, in practical applications, comprehensive analysis and judgment based on specific meteorological observation data and meteorological principles are necessary.

[0080] Step 23 : Based on the division result, the system data is used as a data source to train the scene classification model and the photovoltaic power prediction model under each scene mode.

[0081] Generally speaking, this step involves training scenario-specific PV power prediction models for different datasets. A power scenario classification model is trained using the entire system dataset and classification label information. The power scenario classification model determines the scenario mode based on the power sequence. Both models—the scenario classification model and the PV power prediction model for each scenario mode—are constructed using neural networks.

[0082] Specifically, step 23 includes:

[0083] Step 231: construct the loss functions of the scene classification model and the photovoltaic power prediction model respectively, expressed as:

[0084]

[0085] in, N 1 and N 2 is the number of samples in the training sample set corresponding to the scene classification sub-model and the classification prediction sub-model, k is the sample category; is the loss function of the scene classification sub-model, is the first i The true category labels of samples, is the first i samples belong to the category k The probability of is the loss function of the photovoltaic power prediction model, and They are the first i The actual value of photovoltaic power and the predicted value of photovoltaic power corresponding to the sample.

[0086] Step 232, to minimize and As the goal, the Adam optimizer is called to update the parameters of the scene classification model and the photovoltaic power prediction model respectively using cross entropy loss and mean absolute error through gradient back propagation to realize the training of the scene classification sub-model and the classification prediction sub-model.

[0087] In this way, the scenario mode of the current power fluctuation can be identified through the scenario classification model according to the photovoltaic power sequence, and the photovoltaic power prediction model under the corresponding scenario mode can be called to obtain the photovoltaic power prediction value of the distributed photovoltaic.

[0088] Furthermore, methods such as cross-validation can be used to evaluate the stability and accuracy of the clustering results, and based on the validation results, characteristic variable parameters can be adjusted to optimize the clustering results.

[0089] To further illustrate the advantages of combining K-means clustering for photovoltaic power prediction, Table 1 shows the improvement in prediction accuracy indicators of the prediction model using K-means clustering compared to the ordinary prediction model that does not use K-means clustering and only uses timestamp classification.

[0090] Table 1

[0091]

[0092] Among them, RMSE stands for root mean square error, MAE stands for mean absolute error, and SMAPE stands for symmetric mean absolute percentage error. As can be seen from Table 1, after using K-means clustering, the prediction accuracy of the photovoltaic prediction model is significantly improved.

[0093] Furthermore, it should be noted that in addition to the K-means clustering algorithm, other clustering algorithms can also be used, such as the AP clustering algorithm and the hierarchical clustering algorithm. Figures 5(a) through 5(c) show the results of K-means clustering, AP clustering, and hierarchical clustering, respectively, where the number of AP clustering categories is greater than 3. As shown in Figures 5(a) through 5(c), after merging adjacent categories, the clustering results obtained by the three clustering algorithms differ slightly. However, among these three algorithms, the K-means clustering algorithm is simple and fast, and can efficiently aggregate large amounts of feature data. Therefore, considering all factors, the K-means algorithm is the best choice among the three algorithms.

[0094] Step 3: The system power data, system structure data, and photovoltaic power forecast value of the current period are input into the reinforcement learning-based decision model as the current system state to obtain the distribution network topology reconstruction strategy under the current system state.

[0095] Specifically, step 3 includes:

[0096] Step 31: Establish a power system virtual environment model. The power system virtual environment model is used to simulate the steady state of the distribution network through distributed power flow solution based on the distribution network topology reconstruction strategy output by the decision model, and generate simulated system power data and system structure data in combination with the set photovoltaic power prediction value.

[0097] Among them, by generating simulated system power data and system structure data through the power system virtual environment model, data samples for parameter training of the decision model can be obtained. Specifically, step 31 includes:

[0098] Step 311: Perform distributed power flow calculation using the following formula:

[0099]

[0100] in, 、 Branch ij The active power and reactive power, 、 Branch jk The active power and reactive power, 、 Node i The active and reactive power, and Node i and j The square of the voltage amplitude on For branch ij The square of the upper current amplitude, 、 denote the set of nodes and branches respectively, Representation node i and j Intermediate branch ij , 、 Branch Road ij The resistance and reactance, and They represent the upper and lower bounds of the variable respectively.

[0101] Step 312: Determine whether the distributed power flow solution result converges, and feedback the first negative penalty if it is determined that the power flow solution result does not converge. r pen1 .

[0102] Step 313: Determine whether the current distribution network topology reconstruction strategy meets the system safety operation conditions, and feedback the second negative value penalty if the system safety operation conditions are violated. rpen2 .

[0103] Compared to existing methods that use the Newton-Raphson method to solve AC power flow formulas, this embodiment uses a distributed power flow solution formula to achieve a faster solution speed, thereby quickly generating results. However, reinforcement learning models require extremely high training data volumes, typically interacting with the environment hundreds of thousands to millions of times. Therefore, using a distributed power flow solution method can significantly improve the efficiency of interaction between the reinforcement learning model and the environment.

[0104] Step 32: The power system virtual environment model and the decision model are interacted, and a reinforcement learning-based neural network is trained with the goal of maximizing a reward function to obtain the decision model; the reward function is related to the photovoltaic absorption ratio index, network loss index, switching loss index, whether the power flow solution result converges, and whether the current distribution network topology reconstruction strategy meets the system safety constraints.

[0105] The reward function determines the optimization direction of the decision-making model. In this invention, by setting an appropriate reward function that simultaneously considers PV absorption indicators, economic indicators (including grid loss indicators and switching loss indicators), and safety indicators (i.e., whether the power flow converges and whether the topology reconstruction strategy meets system safety constraints), the trained model can make decisions that maximize PV absorption while finding a topology that meets safety constraints and minimizes grid loss and economic costs. This is of great significance for improving the operational efficiency and economic efficiency of power systems.

[0106] Specifically, the reward function of this embodiment is expressed as:

[0107]

[0108] in, is the absorption ratio of distributed photovoltaics in the system, 、 is the photovoltaic output before and after optimization, is the line loss of the network, is the number of photovoltaic nodes, is the total number of system branches, is the basic reward of the reinforcement learning model, is the number of switch changes under the current strategy, 、 、 The proportional coefficients are used to adjust the weights of the photovoltaic absorption ratio index, network loss index and switching loss index in the reward. A small amount used to prevent reward overflow or division by zero errors due to a small denominator. When the current distribution network topology reconstruction strategy does not meet the system security constraints, an additional negative penalty is introduced. is the total reward function of the model.

[0109] The reinforcement learning-based neural network uses a policy gradient algorithm for updates, and its data structure adopts an action-state-reward triplet architecture. The reinforcement learning-based neural network consists of two parts: an actor network and a critic network. The actor network inputs the system's state data and outputs the current distribution network reconstruction strategy. The critic network inputs the system's current data and the reconstruction strategy provided by the actor network and outputs the state-action value to guide the training parameters. During this process, the power system executes the optimization strategy provided by the actor network, returns the reward parameters, and updates its own system state. The reinforcement learning-based neural network is represented as:

[0110]

[0111] in, For the j The switching action of each branch, For the The distribution network topology reconstruction action at each moment, is the topological state of the system, For the The total state of the system at time is the node load demand, is the predicted value of future photovoltaic output; S represents the state space, with a dimension of 1× N branch ; is the loss function of the critic network used to update the parameters of the critic network, is the system state at the next moment; 、 They are the main critic network and the main actor network respectively; The target critic network for the model, The target actor network for the model, and 、 and The structure is exactly the same. and The parameters of are obtained by performing soft parameter updates after several interactions. s and a They are the system status and distribution network topology reconstruction actions at each moment; The parameters of the main actor network are obtained by gradient ascent of the state-action value of the main critic network. is the parameter change, n The total number of samples collected and used for training in the current training batch.

[0112] Step 4: Monitor the distribution network topology reconstruction strategy and modify the distribution network topology reconstruction strategy that violates the system safety constraint until the system safety constraint is satisfied, thereby obtaining an optimized distribution network topology reconstruction strategy.

[0113] Specifically, in step 4, the correction includes:

[0114] Step 41: Determine the photovoltaic power coefficient of all nodes using the following formula:

[0115]

[0116] in For the The photovoltaic output coefficient of the node. The higher the proportion, the more serious the photovoltaic output of the node is, and priority reduction is required.

[0117] Step 42: Prioritize reduction of the node with the highest photovoltaic power coefficient using the following formula:

[0118]

[0119] in, The reduction coefficient is used to adjust the reduction ratio. and They are the nodes with the highest photovoltaic power coefficient PV power before and after curtailment.

[0120] Step 43: After the reduction, the state is updated and the power flow is recalculated. If there is still a violation, the output coefficient is recalculated and the reduction is performed again. The cycle is iterated until the given distribution network topology reconstruction strategy meets the system safety constraints.

[0121] To validate the performance of the proposed method, the model's optimization results were evaluated using real-world system data. A case study was conducted on a 100-node distribution network constructed with local data from Lianyungang. The network topology includes three backup lines and several distributed photovoltaic power sources. The distributed power sources are combined with user load demands as system inputs. PV data is collected from the actual PV output in Lianyungang and then normalized and incorporated into the system data. Based on this dataset, the distribution network was reconfigured and optimized in real time. The distribution network system is equipped with 14 tie switches, which are normally closed. To reduce network losses, stabilize voltage, and further enhance the regional system's PV absorption capacity, these switches are opened during runtime to adjust the grid topology. The experimental parameters are configured as follows: a penalty of -0.5 for negative values ​​that violate safety constraints, a penalty of -1 for negative values ​​that fail to converge, and a discounted reward coefficient of 0.99 for DDPG training. Under these parameters, the proposed classification prediction model and decision-making model are trained using real-world meteorological and power data sets.

[0122] Figure 3 Schematic diagrams of the power system distribution network topology at different time intervals are shown. Table 2 shows the various indicators of each model used to optimize the distribution network operation mode. The effectiveness of the method was measured by comparing the number of switching operations within the cycle, the average reward value, and the time required for scheduling decision calculations.

[0123] Table 2 Optimization results of different models

[0124]

[0125] It can be observed that for different load demands and changes in photovoltaic output, the studied method can accurately capture the distance weight between the load and the power source, and distribute the user load by adjusting the interconnecting switches of the controllable branches, thereby improving the operation status of the power grid with the least number of operations. For the reconstruction scheduling results within 1 day, the average network loss is controlled from 3.57MW to 3.34MW, and the total network loss is reduced by about 6%. Under the initial topology, the voltage is concentrated in the range of 0.97~1.02 pu. After reconstruction optimization, the voltage distribution of each node in different time periods is more concentrated near the rated voltage. The voltage fluctuation amplitude is significantly reduced during the optimization period, and the voltage deviation is reduced by about 3%. The local voltage rise during the photovoltaic output peak period is significantly suppressed. This shows that the method proposed in this study can improve the power flow distribution of the distribution network system through small-step topology changes, and further enhance the system's photovoltaic absorption capacity.

[0126] In the second aspect of the present invention, a distribution network operation mode optimization system for distributed photovoltaic consumption is provided, which utilizes the method described in the first aspect of the present invention. Figure 4 , the system comprises:

[0127] A data acquisition module is used to collect system data, including system power data with timestamps, system structure data, and meteorological data; the system power data includes photovoltaic power data and load data of each node;

[0128] A power prediction module is configured to identify the current photovoltaic power scenario using a scene classification model and, based on the identified scenario, call a corresponding photovoltaic power prediction model to obtain a photovoltaic power prediction value; the scene classification model and the photovoltaic power prediction model are trained by dividing system power data using meteorological data based on clustering technology;

[0129] The distribution network reconstruction optimization module is used to input the current system power data, system structure data, and photovoltaic power forecast value into the decision model based on reinforcement learning as the system state, and obtain the distribution network topology reconstruction strategy under the current system state;

[0130] The decision correction module is used to monitor the distribution network topology reconstruction strategy and correct the distribution network topology reconstruction strategy that violates the system safety constraints until the system safety constraints are met, thereby obtaining an optimized distribution network topology reconstruction strategy.

[0131] In summary, compared with the existing technology, the method of the present invention analyzes the operating mechanism of the distributed photovoltaic distribution network, constructs a reinforcement learning model for optimizing the distribution network operation mode for photovoltaic absorption, introduces classification mechanism and prediction technology to provide solid and effective optimization information data for decision-making, and embeds artificial intelligence technology into traditional operation optimization problems by designing a unique reward function and intelligent agent architecture. It can ensure that the scheduling optimization plan fully meets the system safety constraints, thereby ensuring the safety and stability of the active distribution network operation.

[0132] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0133] A computer-readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium can be, for example, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0134] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0135] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, the state information of the computer-readable program instructions is used to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), so that the electronic circuit can execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0136] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the scope of knowledge possessed by ordinary technicians in this field without departing from the spirit of the present invention.

Claims

1. A distribution network operation mode optimization method for distributed photovoltaic consumption, characterized in that: The following steps are involved: Step 1: Collect system data, including system power data with timestamps, system structure data, and meteorological data; the system power data includes photovoltaic power data and load data of each node; Step 2: Use the scene classification model to identify the scene mode corresponding to the current photovoltaic power, and call the photovoltaic power prediction model corresponding to the identified scene mode to obtain the photovoltaic power prediction value; The scene classification model and the photovoltaic power prediction model are obtained by training after dividing the system power data through meteorological data based on clustering technology; Step 3: Input the system power data, system structure data, and PV power forecast value of the current period as the current system state into the reinforcement learning-based decision model to obtain the distribution network topology reconstruction strategy under the current system state; the reinforcement learning-based decision model is trained by the following method: Step 31: Establish a power system virtual environment model, wherein the power system virtual environment model is used to simulate the steady state of the distribution network through distributed power flow solution based on the distribution network topology reconstruction strategy output by the decision model, and generate simulated system power data and system structure data in combination with the set photovoltaic power prediction value; Step 32: interacting the power system virtual environment model with the decision model to train a reinforcement learning-based neural network with the goal of maximizing a reward function, thereby obtaining the decision model; the reward function is related to a photovoltaic absorption ratio indicator, a network loss indicator, a switching loss indicator, whether the power flow solution results converge, and whether the current distribution network topology reconstruction strategy meets system safety constraints. Step 4: Monitor the distribution network topology reconstruction strategy and modify the distribution network topology reconstruction strategy that violates the system safety constraint until the system safety constraint is satisfied, thereby obtaining an optimized distribution network topology reconstruction strategy.

2. The method for optimizing the operation mode of distribution network for distributed photovoltaic consumption according to claim 1, characterized in that: In step 1, the meteorological data includes the temperature data, humidity data and air pressure data of the substation at each moment; the system structure data includes the topology of the power system, the status of the line switches and the impedance data of each line at each moment.

3. The method for optimizing the operation mode of distribution network for distributed photovoltaic consumption according to claim 1, characterized in that: In step 2, the scene classification model and the photovoltaic power prediction model are trained in the following manner: Step 21, dividing the photovoltaic power data at each moment into corresponding seasons according to the timestamp; Step 22: select humidity, air pressure, and temperature from the meteorological dataset as characteristic variables for cluster analysis. Use the K-Means clustering algorithm to further classify the photovoltaic power data for each season into three weather modes: sunny, cloudy, and rainy. This divides the photovoltaic power data into sub-power datasets under 12 scenario modes. Step 23 : Based on the division result, the system data is used as a data source to train the scene classification model and the photovoltaic power prediction model under each scene mode.

4. The method for optimizing the operation mode of a distribution network for distributed photovoltaic consumption according to claim 3, characterized in that: Step 22 includes: In step 221, three initial centroids are randomly selected. The centroids are continuously updated and the meteorological data points corresponding to each photovoltaic power data point are assigned to the nearest centroid according to the distance difference from each centroid, thereby completing the clustering of the power data. Each centroid is a three-dimensional vector consisting of the average values ​​of humidity, air pressure, and temperature. The centroids and distances are calculated as follows: in, For meteorological data points i and centroid j The Euclidean distance between H i 、 P i and T i Represents the data points i Humidity, air pressure and temperature values; k is the number of the category cluster to which the meteorological data point belongs, N k For the k The number of meteorological data points in a class cluster, is the new centroid coordinate; Cluster k Indicates the k The set of meteorological data points corresponding to each photovoltaic power data in the category cluster; Step 222 : Analyze the characteristics of the meteorological data in each category cluster, and determine and mark the weather pattern corresponding to each category cluster based on the analysis results.

5. The method for optimizing the operation mode of distribution network for distributed photovoltaic consumption according to claim 3, characterized in that: Step 23 includes: Step 231: construct the loss functions of the scene classification model and the photovoltaic power prediction model respectively, expressed as: in, N 1 and N 2 is the number of samples in the training sample set corresponding to the scene classification sub-model and the classification prediction sub-model, k is the sample category; is the loss function of the scene classification sub-model, is the first i The true category labels of samples, is the first i samples belong to the category k The probability of is the loss function of the photovoltaic power prediction model, and They are the first i The actual value of photovoltaic power and the predicted value of photovoltaic power corresponding to the sample; Step 232, to minimize and As the goal, the Adam optimizer is called to update the parameters of the scene classification model and the photovoltaic power prediction model respectively using cross entropy loss and mean absolute error through gradient back propagation to realize the training of the scene classification sub-model and the classification prediction sub-model.

6. The method for optimizing the distribution network operation mode for distributed photovoltaic consumption according to claim 5, characterized in that: Step 31 includes: Step 311: Perform distributed power flow calculation using the following formula: in, 、 Branch ij The active power and reactive power of 、 Branch jk The active power and reactive power of 、 Node j The active and reactive power, and Node i and j The square of the voltage amplitude on For branch ij The square of the upper current amplitude, 、 denote the set of nodes and branches respectively, Representation node i and j Intermediate branch ij , 、 Branch Road ij The resistance and reactance, and Respectively represent the upper and lower bounds of the variable; Step 312: Determine whether the distributed power flow solution result converges, and feedback the first negative penalty if it is determined that the power flow solution result does not converge. r pen1 ; Step 313: Determine whether the current distribution network topology reconstruction strategy meets the system safety operation conditions, and feedback the second negative value penalty if the system safety operation conditions are violated. r pen2 .

7. The method for optimizing the operation mode of a distribution network for distributed photovoltaic consumption according to claim 6, characterized in that: In step 32, the reward function is expressed as: in, is the absorption ratio of distributed photovoltaics in the system, 、 is the photovoltaic power before and after optimization, is the line loss of the network, is the number of photovoltaic nodes, is the total number of system branches, is the basic reward of the reinforcement learning model, is the number of switch changes under the current strategy, 、 、 The proportional coefficients are used to adjust the weights of the photovoltaic absorption ratio index, network loss index and switching loss index in the reward; A small amount used to prevent the denominator from being too small and causing reward overflow or division by zero errors; is the total negative penalty, and r pen = r pen1+ r pen2 ; is the total reward function of the model.

8. The method for optimizing the distribution network operation mode for distributed photovoltaic consumption according to claim 7, characterized in that: In step 32, the reinforcement learning-based neural network is updated using a policy gradient algorithm and includes an actor network and a critic network. The actor network inputs the system state data and outputs the current distribution network reconstruction strategy. The critic network inputs the system current data and the reconstruction strategy given by the actor network and outputs the state action value to guide the training parameters. The power system executes the optimization strategy given by the actor network, returns the reward parameter, and updates its own system state. The reinforcement learning-based neural network is expressed as: in, For the j The switching action of each branch, For the The distribution network topology reconstruction action at each moment, is the topological state of the system, For the The total state of the system at time is the node load demand, is the predicted value of future photovoltaic power; S represents the state space, with a dimension of 1× N branch ; is the loss function of the critic network, which is used to update the parameters of the critic network; is the system state at the next moment; 、 They are the main critic network and the main actor network respectively; For the target critic network, For the target actor network, and 、 and The structure is exactly the same. and The parameters of are obtained by soft updating after several interactions; s and a They are the system status and distribution network topology reconstruction actions at each moment; is the parameter of the main actor network, obtained by gradient ascent of the state-action value of the main critic network, and is the parameter change. n The total number of samples collected and used for training in the current training batch.

9. The method for optimizing the distribution network operation mode for distributed photovoltaic consumption according to claim 7, characterized in that: In step 4, the correction includes: Step 41: Determine the photovoltaic power coefficient of all nodes using the following formula: in, For nodes The photovoltaic power coefficient; Step 42: Prioritize reduction of the node with the highest photovoltaic power coefficient using the following formula: in, The reduction coefficient is used to regulate the reduction ratio; and They are the nodes with the highest photovoltaic power coefficient PV power before and after curtailment; Step 43: After the reduction, the state is updated and the power flow is recalculated. If there is still a violation, the photovoltaic power coefficient is recalculated and reduced again. The cycle is iterated until the given distribution network topology reconstruction strategy meets the system safety constraints.

10. A distribution network operation mode optimization system for distributed photovoltaic consumption using the method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to collect system data, including system power data with time stamps, system structure data and meteorological data; The system power data includes photovoltaic power data and load data of each node; The power prediction module is used to identify the current photovoltaic power scenario mode using a scenario classification model, and call the corresponding photovoltaic power prediction model according to the identified scenario mode to obtain the photovoltaic power prediction value; The scene classification model and the photovoltaic power prediction model are obtained by training after dividing the system power data through meteorological data based on clustering technology; The distribution network reconstruction optimization module is used to input the current system power data, system structure data, and photovoltaic power forecast value into the decision model based on reinforcement learning as the system state, and obtain the distribution network topology reconstruction strategy under the current system state; The decision correction module is used to monitor the distribution network topology reconstruction strategy and correct the distribution network topology reconstruction strategy that violates the system safety constraints until the system safety constraints are met, thereby obtaining an optimized distribution network topology reconstruction strategy.

Images