A power distribution system service management and control method and device, control system and medium

Abstract

This application discloses a method, device, control system, and medium for business management of a power distribution system, belonging to the field of power distribution network operation and management. The method involves: obtaining a judgment matrix formed by pairwise comparisons of multiple business objectives by the user; generating a priority weight vector through analytic hierarchy process (AHP); receiving measurement data collected by edge devices; calibrating a preset virtual entity model; and constructing a digital twin simulation environment synchronized with the physical power distribution network; iteratively training a strategy network based on the priority weight vector within the environment to obtain a target strategy network; and distributing the target strategy network to each edge device, enabling it to input its assigned measurement data and generate control commands to perform local control of the power distribution system. Therefore, by implementing this application, the problem of poor control optimization performance in power distribution systems under changing operating scenarios can be solved.

Description

Technical Field

[0001] This application relates to the field of power distribution network operation and management, and in particular to a method, device, control system and medium for power distribution system business control. Background Technology

[0002] With the increasing penetration of distributed energy in power distribution systems, power distribution networks face complex operating conditions such as bidirectional power flow and voltage fluctuations. Traditional manual dispatching methods are no longer sufficient to meet real-time control requirements. Therefore, it is necessary to introduce automatic control methods to dynamically regulate the power distribution system's operations in order to ensure the safe, stable, and economical operation of the power grid.

[0003] Existing automatic control methods for power distribution systems typically rely on preset fixed rules or targets for decision-making and control. When faced with different operating scenarios, such as drastic fluctuations in renewable energy output and rapid changes in load peaks and valleys, these control strategies lack flexibility. For example, they prioritize economic absorption when renewable energy output is abundant, while shifting their focus to power supply reliability when extreme weather events are imminent. Therefore, existing fixed-target or fixed-rule control methods struggle to dynamically adjust to different business priorities based on the real-time operational needs of the power grid, resulting in rigid control responses and difficulty in achieving appropriate regulation effects under changing operating conditions. Summary of the Invention

[0004] This application provides a method, device, control system, and medium for business management of a power distribution system, which can solve the problem that the existing control optimization effect is poor under the changing operating scenarios of the power distribution system.

[0005] This application provides a power distribution system service management method in some embodiments, applied to a central controller, wherein the central controller is communicatively connected to multiple edge devices in the power distribution system; the power distribution system service management method includes: The judgment matrix formed by the user pairwise comparison of multiple business objectives of the power distribution system operation is obtained. The judgment matrix is ​​then processed by the analytic hierarchy process to generate a priority weight vector representing the relative importance of each business objective. Receive first measurement data collected by each of the edge devices, calibrate the preset virtual entity model using the first measurement data, and construct a digital twin simulation environment that is synchronized with the physical operating state of the power distribution system. In the digital twin simulation environment, a preset reward function is initialized based on the priority weight vector, and the preset policy network is iteratively trained through the reward function to obtain the target policy network; The target policy network is distributed to each edge device, so that the edge device inputs the corresponding first measurement data into the target policy network, generates the corresponding control command, and performs local control on the power distribution system through the control command.

[0006] Compared with existing technologies, the above embodiments have the following beneficial effects: By obtaining the judgment matrix formed by the user's pairwise comparison of multiple business objectives, and using the analytic hierarchy process (AHP) to generate a priority weight vector, the user can quantify the dynamically adjusted business focus into a calculable weight expression according to changes in the power grid's operating status. On this basis, in a digital twin simulation environment synchronized with the physical distribution network, the policy network is iteratively trained based on this priority weight vector, so that the trained target policy network can adapt to the user's currently declared business priority. This overcomes the shortcomings of existing technologies where fixed target or fixed rule control methods are difficult to dynamically adjust the control focus, improves the control flexibility of the power distribution system under different operating scenarios, and thus improves the control optimization effect of the power distribution system under changing operating scenarios.

[0007] Furthermore, the step of processing the judgment matrix using the analytic hierarchy process (AHP) to generate a priority weight vector representing the relative importance of each of the business objectives includes: Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector; The feature vector is normalized to obtain an initial priority weight vector; The judgment matrix is ​​subjected to consistency verification based on the largest eigenvalue, and after the consistency verification is passed, the initial priority weight vector is used as the priority weight vector.

[0008] Compared with the prior art, the above embodiments have the following beneficial effects: by calculating the largest eigenvalue of the judgment matrix and its corresponding eigenvector, the eigenvector is normalized to obtain the initial priority weight vector, and consistency verification is performed based on the largest eigenvalue. After the verification is passed, the initial priority weight vector is used as the final priority weight vector, so that the priority weight vector generated by the analytic hierarchy process can accurately reflect the user's true preference relationship among multiple business objectives, and avoid the distortion of the weight vector due to inconsistent user judgment logic, thereby providing a reliable optimization direction for the training of the target policy network.

[0009] Further, the calibration of the preset virtual entity model using each of the first measurement data includes: Using a preset state estimator, the first actual operating state of the power distribution system at the current moment is determined based on the first measurement data; Read the second measurement data collected by each of the edge devices at the previous moment, and use the state estimator to determine the second real operating state of the power distribution system at the previous moment based on the second measurement data; The second real operating state is input into the virtual entity model, and the virtual entity model performs deduction to output the first simulated operating state of the power distribution system at the next moment. The first simulation running state is compared with the first real running state to obtain the comparison result, and the virtual entity model is calibrated based on the comparison result.

[0010] Compared with the prior art, the above embodiments have the following beneficial effects: by using a preset state estimator, a first real operating state is determined based on the first measurement data at the current time, and a second real operating state is determined based on the second measurement data at the previous time. The second real operating state is input into the virtual entity model for deduction, and a first simulated operating state is output. The first simulated operating state is compared with the first real operating state to calibrate the virtual entity model. This makes the calibration process based on real measurement data and the model deduction capability as the correction object, which improves the synchronization accuracy between the virtual entity model and the physical operating state of the power distribution system, thereby providing a high-fidelity digital twin simulation environment for the iterative training of the policy network.

[0011] Further, in the digital twin simulation environment, initializing a preset reward function based on the priority weight vector, and iteratively training a preset policy network using the reward function to obtain the target policy network includes: For each training time step, in the digital twin simulation environment, the control action output by the policy network at the current training time step is executed to obtain the estimated value of each business objective, and the reward value of the reward function is obtained by weighting each estimated value through the priority weight vector, and the policy network is updated based on the reward value; For each training time step, the first policy network obtained by updating according to the training time step is updated, and the pre-built policy model set is updated by the non-dominated solution selection method. When the preset iterative training termination condition is met, the corresponding optimal policy network is selected from the policy model set according to the priority weight vector and used as the target policy network.

[0012] Compared with existing technologies, the above embodiments have the following beneficial effects: In each training time step, by executing the control actions output by the policy network in the digital twin simulation environment, the estimated values ​​of each business objective are obtained and the reward values ​​are calculated by weighting them with priority weight vectors. Then, the policy network is updated based on the reward values, so that the update direction of the policy network is always consistent with the business priority declared by the user. At the same time, the policy model set is maintained by the non-dominated solution screening method, and the optimal policy network corresponding to the current priority weight vector is selected as the target policy network. This allows the target policy network to select the one that best matches the current business priority from multiple candidate policies, further improving the matching degree between the control policy and the user's intent.

[0013] Further, in the digital twin simulation environment, executing the control action output by the policy network at the current training time step, obtaining the estimated value of each of the business objectives, and calculating the reward value of the reward function by weighting each of the estimated values ​​through the priority weight vector, and updating the policy network based on the reward value, includes: Obtain the second simulation operating state of the power distribution system output by the virtual entity model at the current training time step, input the second simulation operating state into the policy network, and obtain the control action output by the policy network; The control actions are simulated and executed using the virtual entity model, and the estimated values ​​of each business objective are obtained based on the execution results of the control actions. The reward value is obtained by weighting each estimated value using the priority weight vector, and the model parameters of the policy network are updated by gradient optimization based on the reward value.

[0014] Compared with existing technologies, the above embodiments have the following beneficial effects: by obtaining the second simulation running state output by the virtual entity model, inputting it into the policy network to obtain control actions, and then having the virtual entity model simulate the execution of the control actions and obtain the estimated values ​​of each business objective, the reward value is finally calculated by weighting the priority weight vector and the model parameters of the policy network are updated by gradient optimization. This allows the policy network to complete iterative optimization in the digital twin simulation environment through secure interaction with the virtual entity model, avoiding the operational risks brought about by direct trial and error on the physical power grid, while ensuring a high degree of coupling between the training process and user business priorities.

[0015] Further, the policy model set includes: historical policy networks corresponding to several historical training time steps; the step of updating the pre-constructed policy model set using the first policy network updated according to the training time steps and filtering it using a non-dominated solution method includes: The estimated values ​​of each business objective obtained according to the first policy network are used as the first multidimensional objective value, and the estimated values ​​of each business objective obtained according to the historical policy network are used as the second multidimensional objective value. If the first multidimensional objective value is not dominated by any of the second multidimensional objective values, then the first policy network is added to the policy model set.

[0016] Compared with the prior art, the above embodiments have the following beneficial effects: the estimated values ​​of each business objective corresponding to the first strategy network are used as the first multi-dimensional objective values, and the estimated values ​​of each business objective corresponding to each historical strategy network in the strategy model set are used as the second multi-dimensional objective values. When the first multi-dimensional objective value is not dominated by any second multi-dimensional objective value, the first strategy network is added to the strategy model set, so that the strategy model set retains strategy networks that do not dominate each other in the multi-objective dimensions, avoiding the occupation of storage resources by inferior strategy networks, and providing a diverse candidate strategy basis for subsequent selection of target strategy networks based on different priority weight vectors.

[0017] Further, the step of selecting the corresponding optimal policy network from the policy model set based on the priority weight vector as the target policy network includes: By using the priority weight vector, each of the second multidimensional target values ​​is weighted and calculated to obtain the historical reward value corresponding to each of the historical strategy networks; The historical strategy network with the largest historical reward value is selected as the target strategy network.

[0018] Compared with the prior art, the above embodiments have the following beneficial effects: by weighting the second multidimensional target value corresponding to each historical strategy network through the priority weight vector, the historical reward value corresponding to each historical strategy network is obtained, and the historical strategy network with the largest historical reward value is taken as the target strategy network. This enables the target strategy network selected from the strategy model set to accurately match the business priority declared by the current user through the analytic hierarchy process. It realizes the accurate retrieval of the strategy network that best matches the current preference from the multi-target candidate strategy set, and overcomes the technical defect of the prior art that a single fixed strategy is difficult to adapt to the changing business needs.

[0019] Another embodiment of this application provides a power distribution system business management and control device, applied to a central controller, wherein the central controller is communicatively connected to multiple edge devices in the power distribution system; the power distribution system business management and control device includes: a data acquisition module, a calibration module, a training module, and an instruction generation module; The data acquisition module is used to acquire a judgment matrix formed by the user after comparing multiple business objectives of the power distribution system operation in pairs, and to process the judgment matrix by the analytic hierarchy process to generate a priority weight vector representing the relative importance between the business objectives. The calibration module is used to receive first measurement data collected by each of the edge devices, calibrate the preset virtual entity model through each of the first measurement data, and construct a digital twin simulation environment that is synchronized with the physical operating state of the power distribution system. The training module is used to initialize a preset reward function based on the priority weight vector in the digital twin simulation environment, and iteratively train the preset policy network through the reward function to obtain the target policy network. The instruction generation module is used to send the target policy network to each edge device, so that the edge device inputs the corresponding first measurement data into the target policy network, generates the corresponding control instruction, and performs local control on the power distribution system through the control instruction.

[0020] Another embodiment of this application provides a power distribution system business management system, including a central controller. The central controller is communicatively connected to multiple edge devices in the power distribution system, and the central controller implements the steps of the power distribution system business management method of this application.

[0021] Another embodiment of this application also provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the power distribution system business control method of this application. Attached Figure Description

[0022] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a power distribution system service management method provided in some embodiments of this application; Figure 2 This is a schematic diagram of the structure of a power distribution system business management device provided in some embodiments of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0026] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0029] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0030] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0031] Existing automatic control methods for power distribution systems typically rely on preset fixed rules or targets for decision-making and control. When faced with different operating scenarios, such as drastic fluctuations in renewable energy output and rapid changes in load peaks and valleys, these control strategies lack flexibility. For example, they prioritize economic absorption when renewable energy output is abundant, while shifting their focus to power supply reliability when extreme weather events are imminent. Therefore, existing fixed-target or fixed-rule control methods struggle to dynamically adjust to different business priorities based on the real-time operational needs of the power grid, resulting in rigid control responses and difficulty in achieving appropriate regulation effects under changing operating conditions.

[0032] Please refer to Figure 1 To address the problem of poor control optimization performance in existing power distribution systems under varying operating scenarios, this application provides a power distribution system business management method applied to a central controller, which is communicatively connected to multiple edge devices in the power distribution system.

[0033] Preferably, in some embodiments of this application, the central controller adopts server-grade hardware and has the ability to run large-scale simulations and parallel machine learning training, such as being configured with a multi-socket central processing unit (CPU), multiple high-memory graphics processing units (GPUs), terabyte (TB) level random access memory (RAM) and high-speed storage array.

[0034] Preferably, in some embodiments of this application, the edge device employs industrial-grade, rugged hardware with low power consumption, a wide operating temperature range, and sufficient processing power to perform its tasks (e.g., NVIDIA Jetson series, ARM Cortex-A processor with ML accelerator).

[0035] Furthermore, the power distribution system business control method includes the following steps S101 to S104: S101: Obtain the judgment matrix formed by the user after comparing multiple business objectives of the power distribution system operation in pairs, and process the judgment matrix by the analytic hierarchy process to generate a priority weight vector representing the relative importance of each business objective.

[0036] Preferably, in some embodiments of this application, the formation process of the judgment matrix is ​​as follows: the user compares multiple business objectives of the power distribution system operation in pairs through the graphical user interface provided by the central controller. Specifically, the graphical user interface displays any combination of two business objectives among the multiple business objectives to the user and prompts the user to judge the importance of each relative to the overall operation objective, for example, prompting the user to answer "compared to the optimal operation strategy, which is more important, minimizing cost or maximizing reliability?" The user uses a preset scale to quantify the importance, with the scale being an integer scale from 1 to 9, where 1 indicates that the two business objectives are equally important, 9 indicates that one business objective is extremely important relative to the other, and each intermediate value represents the importance between adjacent judgments. After the user completes the comparison and scoring of all pairs of business objectives, all scoring results are arranged according to the comparison correspondence to form a judgment matrix. The element in the i-th row and j-th column of the judgment matrix represents the importance scale value of the i-th business objective relative to the j-th business objective, and satisfies the reciprocity constraint that all elements on the main diagonal are 1 and elements symmetrical about the main diagonal are reciprocals of each other.

[0037] Preferably, in some embodiments of this application, the business objectives include: economic cost objectives, power supply reliability objectives, clean energy consumption objectives, and grid security objectives. Specifically, the economic cost objective corresponds to the comprehensive economic expenditures during the operation of the distribution system, including electricity purchase costs, network loss costs, and equipment operation and maintenance costs; the power supply reliability objective corresponds to the degree to which the voltage of each node in the distribution system deviates from the rated range and the overload situation of each branch; the clean energy consumption objective corresponds to the proportion of distributed renewable energy output in the total load of the distribution system; and the grid security objective corresponds to the number of components in the distribution system that can still maintain safe and stable operation under N-1 anticipated fault scenarios. These four business objectives reflect the quality of distribution system operation from different perspectives, and there are interdependent relationships between them. For example, excessively pursuing the minimization of economic costs may sacrifice some power supply reliability. Therefore, it is necessary to use the analytic hierarchy process (AHP) to quantitatively weigh and express the relative importance of the above business objectives.

[0038] Furthermore, in some embodiments of this application, the step of processing the judgment matrix using the analytic hierarchy process to generate a priority weight vector representing the relative importance among the various business objectives includes: Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector; The feature vector is normalized to obtain an initial priority weight vector; The judgment matrix is ​​subjected to consistency verification based on the largest eigenvalue, and after the consistency verification is passed, the initial priority weight vector is used as the priority weight vector.

[0039] Specifically, in some embodiments of this application, after obtaining the judgment matrix constructed by the user, the eigenvector method is used to calculate the maximum eigenvalue and its corresponding eigenvector of the judgment matrix. Specifically, the characteristic equation of the judgment matrix is ​​solved to obtain all eigenvalues ​​of the judgment matrix. The eigenvalue with the largest value is selected as the maximum eigenvalue, and the eigenvector corresponding to the maximum eigenvalue is extracted. Each component in the eigenvector corresponds to the original weight information of each business objective among multiple business objectives. Further, the eigenvector is normalized by dividing each component of the eigenvector by the sum of all components, so that the sum of all components in the normalized initial priority weight vector equals 1, and each component is a positive number. The i-th component in the initial priority weight vector represents the relative importance of the i-th business objective among the multiple business objectives. Finally, to ensure the logical consistency of the pairwise comparison judgments given by the user when constructing the judgment matrix, the judgment matrix is ​​checked for consistency based on the maximum eigenvalue. Specifically, the consistency is checked according to the order of the judgment matrix. Calculate the consistency index based on the largest eigenvalue and the order of the eigenvalue. The corresponding random consistency index is used to calculate the consistency ratio. When the consistency ratio is less than a preset consistency threshold, the judgment matrix is ​​determined to have passed the consistency check, indicating that there is no significant logical contradiction between the user's pairwise comparisons. In this case, the initial priority weight vector is output as the priority weight vector. If the consistency ratio is greater than or equal to the preset consistency threshold, the judgment matrix is ​​determined to have failed the consistency check. In this case, the user is prompted to re-compare the multiple business objectives pairwise, adjust the values ​​of the elements in the judgment matrix, and re-execute the feature vector calculation, normalization processing, and consistency check steps until the consistency check is passed.

[0040] Preferably, in some embodiments of this application, it is assumed that the largest eigenvalue is The consistency index is calculated using the following formula: .when When the value is 4, the random consistency index can be obtained by looking up a table. Furthermore, calculate the consistency ratio. ,when If the judgment matrix passes the consistency check, the user is advised to readjust the comparison judgment.

[0041] This application calculates the largest eigenvalue of the judgment matrix and its corresponding eigenvector, normalizes the eigenvector to obtain the initial priority weight vector, and performs consistency verification based on the largest eigenvalue. After the verification passes, the initial priority weight vector is used as the final priority weight vector. This ensures that the priority weight vector generated by the analytic hierarchy process can accurately reflect the user's true preference relationship among multiple business objectives, avoiding weight vector distortion due to inconsistent user judgment logic, thereby providing a reliable optimization direction for the training of the target policy network.

[0042] S102: Receive the first measurement data collected by each of the edge devices, calibrate the preset virtual entity model using the first measurement data, and construct a digital twin simulation environment that is synchronized with the physical operating state of the power distribution system.

[0043] Furthermore, in some embodiments of this application, the calibration of the preset virtual entity model using each of the first measurement data includes: Using a preset state estimator, the first actual operating state of the power distribution system at the current moment is determined based on the first measurement data; Read the second measurement data collected by each of the edge devices at the previous moment, and use the state estimator to determine the second real operating state of the power distribution system at the previous moment based on the second measurement data; The second real operating state is input into the virtual entity model, and the virtual entity model performs deduction to output the first simulated operating state of the power distribution system at the next moment. The first simulation running state is compared with the first real running state to obtain the comparison result, and the virtual entity model is calibrated based on the comparison result.

[0044] Preferably, in some embodiments of this application, the virtual entity model includes: physical entities, virtual entities, connections between the two, twin data, and services. It accurately models the topology of the power distribution system, the physical characteristics of equipment (transformers, lines), and the dynamic behavior of loads and distributed energy sources. Data collected through edge devices is used to calibrate and synchronize the virtual entity model with the actual state of the physical power grid in real time.

[0045] Preferably, in some embodiments of this application, after the central controller receives the first measurement data collected by each edge device, the first measurement data is first preprocessed. The preprocessing process includes bad data detection and missing value interpolation (using time-series interpolation based on Kalman filtering), wherein bad data detection is used to identify and remove outliers in the measurement data, and missing value interpolation is used to reasonably fill in measurement gaps caused by communication interruption or equipment failure, and high-quality first measurement data is formed after preprocessing.

[0046] Preferably, in some embodiments of this application, after the first measurement data is preprocessed, a preset state estimator is used to solve for the actual operating state of the power distribution system at the current moment using the weighted least squares method based on the preprocessed first measurement data, thereby obtaining the first true operating state. The first true operating state solved by the state estimator includes key operating parameters such as the voltage amplitude and phase angle of each node in the power distribution system, and the distribution of active and reactive power in each branch. This first true operating state reflects the actual physical condition of the power distribution system at the current moment.

[0047] Preferably, in some embodiments of this application, after reading the second measurement data collected and uploaded by each edge device at the previous moment, and after undergoing the same preprocessing process as the first measurement data, the second actual operating state is obtained by using a state estimator based on the second measurement data. The second actual operating state characterizes the actual operating condition of the power distribution system at the previous moment. Using the second actual operating state as the initial input condition of the virtual entity model, the virtual entity model simulates the operational evolution of the power distribution system from the previous moment to the current moment based on its internally preset power grid topology, equipment physical characteristics, and load and generation dynamic models, and outputs the first simulated operating state of the power distribution system at the current moment. The first simulated operating state is the current moment's state prediction value obtained by the virtual entity model based on the correct starting point of the second actual operating state and according to its own model logic. By comparing the first simulated operating state with the first actual operating state item by item, the deviation between the two in each operating parameter is calculated.

[0048] Preferably, in some embodiments of this application, the deviation can be determined by calculating the root mean square error of voltage and the absolute percentage error of power.

[0049] Preferably, in some embodiments of this application, when each deviation exceeds the corresponding preset threshold (e.g., the root mean square error of voltage is less than 0.005 pu, and the absolute percentage error of power is less than 2%), key equipment parameters such as transformer short-circuit impedance and line resistance reactance are identified online based on the first measurement data obtained from the edge device, and the internal parameters corresponding to the virtual entity model are further corrected. Through the above comparison and correction process, the virtual entity model's ability to predict the evolution of the power distribution system gradually approaches the real behavioral characteristics of the physical power distribution system.

[0050] This application utilizes a preset state estimator to determine a first true operating state based on the first measurement data at the current moment, and simultaneously determines a second true operating state based on the second measurement data at the previous moment. The second true operating state is then input into a virtual entity model for deduction, outputting a first simulated operating state. The first simulated operating state is then compared with the first true operating state to calibrate the virtual entity model. This calibration process uses real measurement data as a benchmark and the model's deduction capability as the correction target, improving the synchronization accuracy between the virtual entity model and the physical operating state of the power distribution system. Consequently, it provides a high-fidelity digital twin simulation environment for the iterative training of the policy network.

[0051] S103: In the digital twin simulation environment, a preset reward function is initialized based on the priority weight vector, and the preset policy network is iteratively trained through the reward function to obtain the target policy network.

[0052] Preferably, in some embodiments of this application, the policy network is part of a reinforcement learning model, which employs a fully connected neural network with two hidden layers, each with 256 neurons, and uses the ReLU activation function. The reinforcement learning model also includes a value network, which has the same structure as the policy network.

[0053] Furthermore, in some embodiments of this application, the step of initializing a preset reward function based on the priority weight vector in the digital twin simulation environment, and iteratively training a preset policy network through the reward function to obtain a target policy network, includes the following steps S1031 to S1033: S1031: For each training time step, in the digital twin simulation environment, execute the control action output by the policy network at the current training time step, obtain the estimated value of each business objective, and calculate the reward value of the reward function by weighting each estimated value through the priority weight vector, and update the policy network based on the reward value.

[0054] Further, in some embodiments of this application, the step of executing the control action output by the policy network at the current training time step in the digital twin simulation environment, obtaining the estimated value of each of the business objectives, and calculating the reward value of the reward function by weighting each of the estimated values ​​through the priority weight vector, and updating the policy network based on the reward value, includes: Obtain the second simulation operating state of the power distribution system output by the virtual entity model at the current training time step, input the second simulation operating state into the policy network, and obtain the control action output by the policy network; The control actions are simulated and executed using the virtual entity model, and the estimated values ​​of each business objective are obtained based on the execution results of the control actions. The reward value is obtained by weighting each estimated value using the priority weight vector, and the model parameters of the policy network are updated by gradient optimization based on the reward value.

[0055] Preferably, in some embodiments of this application, in each training time step, the virtual entity model, based on the control action execution results of the previous time step and its internal dynamic model, extrapolates and outputs the second simulated operating state of the power distribution system at the current training time step. The second simulated operating state includes operating status information such as the voltage amplitude of each node in the power distribution system, the active and reactive power distribution of each branch, the load rate of each line, the real-time output level of distributed energy resources, and equipment health indices. This status information collectively constitutes the current observation input required for the policy network to make decisions. The second simulated operating state is input into the policy network, which performs forward propagation calculations on the input simulated operating state based on the model parameters of the current training time step and outputs the corresponding control action.

[0056] Preferably, in some embodiments of this application, the control action can be a discrete command or a continuous setpoint, including but not limited to on-load tap changer tap position adjustment, capacitor bank switching operation, feeder topology reconfiguration switching action, and active or reactive power output scheduling setpoints issued to distributed energy sources. The specific form of the control action output by the strategy network depends on the pre-defined action space definition.

[0057] Preferably, in some embodiments of this application, when the virtual entity model receives a control action, it simulates and executes the control action within its simulation environment. During the simulation, the virtual entity model calculates the changes in the electrical state of each node in the power distribution system after the control action is implemented, based on the physical model, load model, and power generation dynamic model of the power distribution system, and quantitatively evaluates the performance of each business objective at the current training time step based on the execution results. Specifically, based on the node voltage distribution, branch power distribution, number of equipment actions, proportion of distributed energy output, and N-1 security verification results after the simulation, the estimated values ​​corresponding to the economic cost target, power supply reliability target, clean energy consumption target, and grid security target are calculated respectively. In addition, an action penalty term can be introduced to suppress frequent actions of the control device by measuring the change between the current control action and the control action of the previous training time step.

[0058] Preferably, in some embodiments of this application, the economic cost target is calculated using the following formula: ; in, for Real-time economic cost target; for Real-time electricity purchase cost; For network loss costs; The maintenance cost is calculated based on the number of times the equipment is operated. This is the normalized benchmark cost.

[0059] Preferably, in some embodiments of this application, the power supply reliability target is calculated using the following formula: ; in, For power supply reliability objectives; This refers to the number of nodes in the power distribution system. For nodes per-unit voltage; This refers to the number of power distribution lines. For the line exist Load rate at any given time; For the line Maximum allowable load rate; This is the over-limit weighting coefficient.

[0060] Preferably, in some embodiments of this application, the clean energy consumption target is calculated using the following formula: ; in, To achieve the goal of clean energy consumption; Contribute to renewable energy; This represents the total load.

[0061] Preferably, in some embodiments of this application, the power grid security target is calculated using the following formula: ; in, For the purpose of power grid security; Weighting for security penalties; for The number of components that do not meet the N-1 safety check at any given time.

[0062] Preferably, in some embodiments of this application, the action penalty item is: ; in, For action-based penalties; Weighting of action penalties; for The number of times the control device operates at a given time; for The number of times the control device operates at a given time.

[0063] Preferably, in some embodiments of this application, after calculating the estimated values ​​corresponding to each business objective, the estimated values ​​of each business objective are weighted using a priority weight vector. Specifically, the estimated value of the economic cost objective is multiplied by the weight component corresponding to the economic dimension in the priority weight vector; the estimated value of the power supply reliability objective is multiplied by the weight component corresponding to the reliability dimension; the estimated value of the clean energy consumption objective is multiplied by the weight component corresponding to the environmental dimension; and the estimated value of the power grid security objective is multiplied by the weight component corresponding to the security dimension. The weighted results are then summed to obtain the reward value for the current training time step. This reward value incorporates the relative importance preferences among the business objectives declared by the operator through the analytic hierarchy process, ensuring that the evaluation criteria for the current control action are consistent with the operator's current operational intentions.

[0064] Preferably, in some embodiments of this application, gradient optimization updates of the model parameters of the policy network based on reward values ​​include: calculating the advantage estimate of the control action at the current training time step based on the reward value, constructing a policy optimization objective function with the goal of maximizing the cumulative reward value, and updating the model parameters of the policy network using a proximal policy pruning mechanism. The proximal policy pruning mechanism avoids training instability caused by excessively large single-step updates by limiting the update magnitude between the old and new policies. Simultaneously, it combines the mean squared error loss of the value network and the policy entropy regularization term, and optimizes the model parameters of the policy network through gradient backpropagation using the Adam optimizer. After iterative updates over multiple training time steps, the policy network gradually learns control action selection policies that can obtain high reward values ​​in the digital twin simulation environment.

[0065] This application obtains the second simulation running state output by the virtual entity model, inputs it into the policy network to obtain control actions, and then the virtual entity model simulates the execution of the control actions and obtains the estimated values ​​of each business objective. Finally, the reward value is calculated by weighting the priority weight vector and the model parameters of the policy network are updated by gradient optimization. This allows the policy network to complete iterative optimization in the digital twin simulation environment through safe interaction with the virtual entity model, avoiding the operational risks caused by direct trial and error on the physical power grid, while ensuring a high degree of coupling between the training process and user business priorities.

[0066] S1032: For each training time step, update the first policy network obtained according to the training time step, and update the pre-built policy model set through the non-dominated solution screening method.

[0067] Furthermore, in some embodiments of this application, the policy model set includes: a historical policy network corresponding to several historical training time steps; the step of updating the pre-constructed policy model set using a non-dominated solution selection method based on the first policy network updated according to the training time steps includes: The estimated values ​​of each business objective obtained according to the first policy network are used as the first multidimensional objective value, and the estimated values ​​of each business objective obtained according to the historical policy network are used as the second multidimensional objective value. If the first multidimensional objective value is not dominated by any of the second multidimensional objective values, then the first policy network is added to the policy model set.

[0068] Understandably, during training, the policy model set acts as a cover set of the Pareto front, used to store the non-dominated policy networks that emerge during training. For the first policy network updated at the current training time step, it is loaded into the digital twin simulation environment and run a complete evaluation cycle, recording its average performance on the four operational objectives of economic cost, power supply reliability, clean energy consumption, and grid security, forming the first multidimensional objective value. Similarly, for each historical policy network already stored in the policy model set, there is also a corresponding set of second multidimensional objective values ​​evaluated during previous training.

[0069] Preferably, in some embodiments of this application, the dominance relationship between the first multidimensional target value and the second multidimensional target value is determined by the following process: comparing the first multidimensional target value with the second multidimensional target value of each historical strategy network dimension by dimension. If there exists a historical strategy network whose second multidimensional target value is not higher than the first multidimensional target value in terms of economic cost, not lower than the first multidimensional target value in terms of power supply reliability, not lower than the first multidimensional target value in terms of clean energy consumption, and not lower than the first multidimensional target value in terms of grid security, and is strictly superior to the first multidimensional target value in at least one target dimension, then it is determined that the historical strategy network dominates the first strategy network, indicating that the first strategy network does not provide a better overall performance than existing strategies in multiple objectives. Conversely, if all historical strategy networks in the strategy model set cannot satisfy the above dominance conditions, it indicates that the first strategy network is a non-dominated solution, which may be weaker than existing strategies in some target dimensions, but is superior to all existing strategies in at least one target dimension, or forms a new compromise among multiple objectives.

[0070] Preferably, in some embodiments of this application, after the first policy network is added to the policy model set, it can also be checked in reverse whether the first policy network dominates some historical policy networks in the set. If there are dominated historical policy networks, they are removed from the set, so as to keep the policy model set as a set of non-dominant Pareto front policies, avoid unnecessary expansion of the set size, and ensure the diversity and uniformity of candidate policies in the target space.

[0071] This application uses the estimated values ​​of each business objective corresponding to the first strategy network as the first multi-dimensional objective value, and the estimated values ​​of each business objective corresponding to each historical strategy network in the strategy model set as the second multi-dimensional objective value. When the first multi-dimensional objective value is not dominated by any second multi-dimensional objective value, the first strategy network is added to the strategy model set. This ensures that the strategy model set retains strategy networks that are not mutually dominant in the multi-objective dimensions, thus avoiding the occupation of storage resources by inferior strategy networks. At the same time, it provides a diverse candidate strategy basis for subsequent selection of target strategy networks based on different priority weight vectors.

[0072] S1033: When the preset iterative training termination condition is met, the corresponding optimal policy network is selected from the policy model set according to the priority weight vector and used as the target policy network.

[0073] Furthermore, in some embodiments of this application, the step of selecting the corresponding optimal policy network from the policy model set based on the priority weight vector as the target policy network includes: By using the priority weight vector, each of the second multidimensional target values ​​is weighted and calculated to obtain the historical reward value corresponding to each of the historical strategy networks; The historical policy network with the largest historical reward value is selected as the target policy network.

[0074] Understandably, after the iterative training meets the preset termination condition, the strategy model set has saved all the non-dominated strategy networks that emerged during training. Each historical strategy network corresponds to a set of second multidimensional target values ​​recorded during the training evaluation phase. At this point, it is necessary to select a historical strategy network from the strategy model set that best matches the current operator's business preferences as the final deployment target strategy network. Using the priority weight vector obtained from the analytic hierarchy process (AHP) module, the second multidimensional target values ​​corresponding to each historical strategy network in the strategy model set are scalarized and weighted to obtain the historical reward value for that historical strategy network. This historical reward value comprehensively reflects the weighted overall performance of the historical strategy network on the various business objectives currently of concern to the operator; business objectives with higher weights contribute more to the historical reward value. After traversing all historical strategy networks in the strategy model set and calculating their corresponding historical reward values ​​one by one, the historical strategy network with the largest historical reward value is determined as the optimal strategy network and output as the target strategy network. Because the screening process directly uses the priority weight vector declared by the operator through the analytic hierarchy process, the selected target policy network can best match the operator's business preferences in the current operating scenario, and achieves the accurate retrieval of the control policy that best matches the current priority weight vector from the policy model set.

[0075] This application uses a priority weight vector to weight the second multidimensional target value corresponding to each historical strategy network to obtain the historical reward value corresponding to each historical strategy network. The historical strategy network with the largest historical reward value is then used as the target strategy network. This allows the target strategy network selected from the strategy model set to accurately match the business priority declared by the current user through the analytic hierarchy process. This enables the precise retrieval of the strategy network that best suits the current preference from a multi-target candidate strategy set, overcoming the technical deficiency of existing technologies where a single fixed strategy cannot adapt to changing business needs.

[0076] As can be seen from S1031 to S1033 above, in each training time step, this application obtains the estimated values ​​of each business objective by executing the control actions output by the policy network in the digital twin simulation environment, and calculates the reward value by weighting it with the priority weight vector. Then, the policy network is updated based on the reward value, so that the update direction of the policy network is always consistent with the business priority declared by the user. At the same time, the policy model set is maintained by the non-dominated solution screening method, and the optimal policy network corresponding to the current priority weight vector is selected as the target policy network. This allows the target policy network to select the one that best matches the current business priority from multiple candidate policies, further improving the matching degree between the control policy and the user's intention.

[0077] S104: The target policy network is sent to each of the edge devices, so that the edge devices input the corresponding first measurement data into the target policy network, generate corresponding control commands, and perform local control on the power distribution system through the control commands.

[0078] Preferably, in some embodiments of this application, after the edge device receives the target policy network, it loads it into its local inference engine. During the real-time operation phase, the edge device continuously collects first measurement data of its jurisdiction. The edge device preprocesses the collected raw measurement data, including bad data identification and removal, missing value imputation, and data aggregation, transforming massive amounts of raw data into low-volume, high-value information. The preprocessed first measurement data is input into the target policy network loaded locally by the edge device. The target policy network performs forward inference calculations on the input data and outputs control commands for the local area. The form of the control commands is consistent with the action space defined in the training phase, including on-load tap changer tap position adjustment commands, capacitor bank switching commands, feeder topology reconfiguration switch operation commands, and active or reactive power output scheduling setpoint commands issued to the distributed energy inverter. The edge device sends the generated control commands to the corresponding primary equipment actuators through the local communication interface, driving the power distribution equipment to perform corresponding control actions.

[0079] Understandably, because edge devices can complete the entire reasoning process from measurement data input to control command output locally, without needing to communicate and interact with the central controller in real time, they avoid the communication latency and bandwidth bottlenecks of a purely centralized architecture, enabling near real-time rapid responses to changes in local operating conditions. Simultaneously, new measurement data generated after each edge device executes control actions continues to be uploaded to the central controller periodically, flowing back to the digital twin simulation environment for calibrating the virtual entity model, forming a complete closed-loop operating mechanism encompassing state awareness, policy reasoning, command generation, edge execution, and measurement feedback.

[0080] In summary, the power distribution system business management method provided in this application has the following advantages compared with the prior art: by obtaining the judgment matrix formed by the user comparing multiple business objectives pairwise, and using the analytic hierarchy process (AHP) to generate a priority weight vector, the user can quantify the dynamically adjusted business focus into a calculable weight expression according to the changes in the power grid operation status; on this basis, in a digital twin simulation environment synchronized with the physical power distribution network, the policy network is iteratively trained based on the priority weight vector, so that the trained target policy network can adapt to the user's currently declared business priority, overcoming the defect of fixed target or fixed rule control methods in the prior art that are difficult to dynamically adjust the control focus, improving the control flexibility of the power distribution system under different operating scenarios, thereby improving the control optimization effect of the power distribution system under changing operating scenarios.

[0081] like Figure 2 As shown, based on the above method embodiments, one embodiment of this application provides a power distribution system business management and control device, applied to a central controller, wherein the central controller is communicatively connected to multiple edge devices in the power distribution system; the power distribution system business management and control device includes: a data acquisition module 201, a calibration module 202, a training module 203, and an instruction generation module 204; The data acquisition module 201 is used to acquire a judgment matrix formed by the user after comparing multiple business objectives of the power distribution system operation in pairs, and to process the judgment matrix by the analytic hierarchy process to generate a priority weight vector representing the relative importance between the business objectives. The calibration module 202 is used to receive first measurement data collected by each of the edge devices, calibrate the preset virtual entity model through each of the first measurement data, and construct a digital twin simulation environment that is synchronized with the physical operating state of the power distribution system. The training module 203 is used to initialize a preset reward function based on the priority weight vector in the digital twin simulation environment, and iteratively train the preset policy network through the reward function to obtain the target policy network. The instruction generation module 204 is used to send the target policy network to each of the edge devices, so that the edge devices input the corresponding first measurement data into the target policy network, generate corresponding control instructions, and perform local control on the power distribution system through the control instructions.

[0082] Further, in some embodiments of this application, the data acquisition module 201 includes: a calculation unit, a normalization unit, and a verification unit; the data acquisition module 201 is used to process the judgment matrix using the analytic hierarchy process (AHP) to generate a priority weight vector representing the relative importance among the various business objectives, including: The calculation unit is used to calculate the maximum eigenvalue of the judgment matrix and its corresponding eigenvector; The normalization unit is used to normalize the feature vector to obtain an initial priority weight vector. The verification unit is used to perform consistency verification on the judgment matrix based on the largest eigenvalue, and after the consistency verification is passed, the initial priority weight vector is used as the priority weight vector.

[0083] Further, in some embodiments of this application, the calibration module 202 includes: a first estimation unit, a second estimation unit, a simulation unit, and a comparison unit; the calibration module 202 is used to calibrate a preset virtual entity model using each of the first measurement data, including: The first estimation unit is used to determine the first true operating state of the power distribution system at the current moment based on the first measurement data using a preset state estimator. The second estimation unit is used to read the second measurement data collected by each of the edge devices at the previous moment, and use the state estimator to determine the second real operating state of the power distribution system at the previous moment based on the second measurement data; The simulation unit is used to input the second real operating state into the virtual entity model, and the virtual entity model performs deduction to output the first simulated operating state of the power distribution system at the next moment. The comparison unit is used to compare the first simulation running state with the first real running state, obtain the comparison result, and calibrate the virtual entity model based on the comparison result.

[0084] Further, in some embodiments of this application, the training module 203 includes: a first update unit, a second update unit, and a filtering unit; the training module 203 is used to initialize a preset reward function based on the priority weight vector in the digital twin simulation environment, and iteratively train a preset policy network through the reward function to obtain a target policy network, including: The first update unit is configured to, for each training time step, in the digital twin simulation environment, execute the control action output by the policy network at the current training time step, obtain the estimated value of each of the business objectives, and calculate the reward value of the reward function by weighting each of the estimated values ​​through the priority weight vector, and update the policy network based on the reward value; The second update unit is used to update the pre-built policy model set for each training time step by updating the first policy network obtained according to the training time step and using a non-dominated solution filtering method. The filtering unit is used to select the corresponding optimal policy network from the policy model set according to the priority weight vector when the preset iterative training termination condition is met, and use it as the target policy network.

[0085] Further, in some embodiments of this application, the first update unit is configured to execute the control action output by the policy network at the current training time step in the digital twin simulation environment, obtain the estimated value of each of the business objectives, and calculate the reward value of the reward function by weighting each of the estimated values ​​through the priority weight vector, and update the policy network based on the reward value, including: Obtain the second simulation operating state of the power distribution system output by the virtual entity model at the current training time step, input the second simulation operating state into the policy network, and obtain the control action output by the policy network; The control actions are simulated and executed using the virtual entity model, and the estimated values ​​of each business objective are obtained based on the execution results of the control actions. The reward value is obtained by weighting each estimated value using the priority weight vector, and the model parameters of the policy network are updated by gradient optimization based on the reward value.

[0086] Further, in some embodiments of this application, the policy model set includes: a historical policy network corresponding to several historical training time steps; the second update unit is used to update the pre-built policy model set according to the first policy network updated according to the training time steps, using a non-dominated solution filtering method, including: The estimated values ​​of each business objective obtained according to the first policy network are used as the first multidimensional objective value, and the estimated values ​​of each business objective obtained according to the historical policy network are used as the second multidimensional objective value. If the first multidimensional objective value is not dominated by any of the second multidimensional objective values, then the first policy network is added to the policy model set.

[0087] Furthermore, in some embodiments of this application, the filtering unit is used to filter out the corresponding optimal policy network from the policy model set according to the priority weight vector, as the target policy network, including: By using the priority weight vector, each of the second multidimensional target values ​​is weighted and calculated to obtain the historical reward value corresponding to each of the historical strategy networks; The historical policy network with the largest historical reward value is selected as the target policy network.

[0088] It is understood that the above-described device embodiments correspond to the method embodiments of this application, and can implement the power distribution system business management method provided by any of the above-described method embodiments of this application.

[0089] In summary, the power distribution system business management device provided in this application has the following advantages compared with the prior art: by obtaining the judgment matrix formed by the user comparing multiple business objectives pairwise, and using the analytic hierarchy process (AHP) to generate a priority weight vector, the user can quantify the dynamically adjusted business focus into a calculable weight expression according to the changes in the power grid operation status; on this basis, in a digital twin simulation environment synchronized with the physical power distribution network, the strategy network is iteratively trained based on the priority weight vector, so that the trained target strategy network can adapt to the user's currently declared business priority, overcoming the shortcomings of the fixed target or fixed rule control methods in the prior art that are difficult to dynamically adjust the control focus, improving the control flexibility of the power distribution system under different operating scenarios, and thus improving the control optimization effect of the power distribution system under changing operating scenarios.

[0090] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0091] Based on the above embodiments of the power distribution system business management and control method, another embodiment of this application provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the power distribution system business management and control method of any embodiment of this application.

[0092] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete this application. The one or more module units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0093] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0094] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0095] Based on the above-described method embodiments, another embodiment of this application provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the power distribution system business management method described in any of the above-described method embodiments of this application.

[0096] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

Claims

1. A method for business control of a power distribution system, characterized in that, The method is applied to a central controller, which is communicatively connected to multiple edge devices in a power distribution system; the power distribution system service management method includes: The judgment matrix formed by the user pairwise comparison of multiple business objectives of the power distribution system operation is obtained. The judgment matrix is ​​then processed by the analytic hierarchy process to generate a priority weight vector representing the relative importance of each business objective. Receive first measurement data collected by each of the edge devices, calibrate the preset virtual entity model using the first measurement data, and construct a digital twin simulation environment that is synchronized with the physical operating state of the power distribution system. In the digital twin simulation environment, a preset reward function is initialized based on the priority weight vector, and the preset policy network is iteratively trained through the reward function to obtain the target policy network; The target policy network is distributed to each edge device, so that the edge device inputs the corresponding first measurement data into the target policy network, generates the corresponding control command, and performs local control on the power distribution system through the control command.

2. The power distribution system business management and control method as described in claim 1, characterized in that, The step of processing the judgment matrix using the analytic hierarchy process (AHP) to generate a priority weight vector representing the relative importance of each of the business objectives includes: Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector; The feature vector is normalized to obtain an initial priority weight vector; The judgment matrix is ​​subjected to consistency verification based on the largest eigenvalue, and after the consistency verification is passed, the initial priority weight vector is used as the priority weight vector.

3. The power distribution system business management and control method as described in claim 1, characterized in that, The calibration of the preset virtual entity model using each of the first measurement data includes: Using a preset state estimator, the first actual operating state of the power distribution system at the current moment is determined based on the first measurement data; Read the second measurement data collected by each of the edge devices at the previous moment, and use the state estimator to determine the second real operating state of the power distribution system at the previous moment based on the second measurement data; The second real operating state is input into the virtual entity model, and the virtual entity model performs deduction to output the first simulated operating state of the power distribution system at the next moment. The first simulation running state is compared with the first real running state to obtain the comparison result, and the virtual entity model is calibrated based on the comparison result.

4. The power distribution system business management and control method as described in claim 1, characterized in that, In the digital twin simulation environment, a preset reward function is initialized based on the priority weight vector, and a preset policy network is iteratively trained using the reward function to obtain the target policy network, including: For each training time step, in the digital twin simulation environment, the control action output by the policy network at the current training time step is executed to obtain the estimated value of each business objective, and the reward value of the reward function is obtained by weighting each estimated value through the priority weight vector, and the policy network is updated based on the reward value; For each training time step, the first policy network obtained by updating according to the training time step is updated, and the pre-built policy model set is updated by the non-dominated solution selection method. When the preset iterative training termination condition is met, the corresponding optimal policy network is selected from the policy model set according to the priority weight vector and used as the target policy network.

5. The power distribution system business management and control method as described in claim 4, characterized in that, In the digital twin simulation environment, the process of executing the control action output by the policy network at the current training time step, obtaining the estimated values ​​of each business objective, and calculating the reward value of the reward function by weighting each estimated value using the priority weight vector, and updating the policy network based on the reward value, includes: Obtain the second simulation operating state of the power distribution system output by the virtual entity model at the current training time step, input the second simulation operating state into the policy network, and obtain the control action output by the policy network; The control actions are simulated and executed using the virtual entity model, and the estimated values ​​of each business objective are obtained based on the execution results of the control actions. The reward value is obtained by weighting each estimated value using the priority weight vector, and the model parameters of the policy network are updated by gradient optimization based on the reward value.

6. The power distribution system business management and control method as described in claim 4, characterized in that, The policy model set includes: historical policy networks corresponding to several historical training time steps; the first policy network updated according to the training time steps, and the pre-constructed policy model set updated using a non-dominated solution selection method, includes: The estimated values ​​of each business objective obtained according to the first policy network are used as the first multidimensional objective value, and the estimated values ​​of each business objective obtained according to the historical policy network are used as the second multidimensional objective value. If the first multidimensional objective value is not dominated by any of the second multidimensional objective values, then the first policy network is added to the policy model set.

7. The power distribution system business management and control method as described in claim 6, characterized in that, The step of selecting the optimal policy network from the policy model set based on the priority weight vector as the target policy network includes: By using the priority weight vector, each of the second multidimensional target values ​​is weighted and calculated to obtain the historical reward value corresponding to each of the historical strategy networks; The historical strategy network with the largest historical reward value is selected as the target strategy network.

8. A power distribution system business control device, characterized in that, It is applied to a central controller, which is communicatively connected to multiple edge devices in the power distribution system; the power distribution system business management and control device includes: a data acquisition module, a calibration module, a training module, and an instruction generation module; The data acquisition module is used to acquire a judgment matrix formed by the user after comparing multiple business objectives of the power distribution system operation in pairs, and to process the judgment matrix by the analytic hierarchy process to generate a priority weight vector representing the relative importance between the business objectives. The calibration module is used to receive first measurement data collected by each of the edge devices, calibrate the preset virtual entity model through each of the first measurement data, and construct a digital twin simulation environment that is synchronized with the physical operating state of the power distribution system. The training module is used to initialize a preset reward function based on the priority weight vector in the digital twin simulation environment, and iteratively train the preset policy network through the reward function to obtain the target policy network. The instruction generation module is used to send the target policy network to each edge device, so that the edge device inputs the corresponding first measurement data into the target policy network, generates the corresponding control instruction, and performs local control on the power distribution system through the control instruction.

9. A power distribution system business management and control system, characterized in that, The system includes a central controller that is communicatively connected to multiple edge devices in the power distribution system. The central controller implements a power distribution system business management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform a power distribution system business management method as described in any one of claims 1 to 7.

Images