Intelligent digital twin collaborative scheduling method

Abstract

This invention discloses an intelligent digital twin collaborative scheduling method, relating to the fields of intelligent distribution network scheduling control and digital twins, comprising: S1: collecting multi-source data of the distribution network and constructing a full-element digital twin mirror model of the distribution network; S2: based on the digital twin mirror model, performing multi-scenario simulation and deduction of the operating status of the distribution network, outputting key operating indicators to characterize the operating status of the power grid, and providing safety constraint boundaries for scheduling optimization based on the simulation results; S3: constructing a multi-timescale collaborative scheduling framework and, within the constraint boundaries, performing multi-objective collaborative optimization of the distribution network scheduling plan to generate an optimized scheduling scheme; S4: executing hierarchical collaborative control based on the optimized scheduling scheme; S5: performing closed-loop feedback and iterative optimization of the digital twin mirror model and scheduling optimization strategy based on the actual operating data collected after the execution of scheduling instructions. This achieves multi-objective collaborative optimization of new energy consumption, low-carbon operation, and line loss optimization.

Description

Technical Field

[0001] This invention relates to the field of intelligent power distribution network dispatch control and digital twin technology, specifically to an intelligent digital twin collaborative dispatch method that adapts to the large-scale access needs of distributed wind power, photovoltaic and other new energy rural scenarios. Background Technology

[0002] Currently, county and rural 10kV and below distribution networks have become the core carriers for the large-scale access of distributed renewable energy and the low-carbon transformation of rural energy. However, the current dispatch and operation of rural distribution networks face many technical bottlenecks:

[0003] (1) The rural power distribution network has a weak foundation, many points and wide distribution, and scattered equipment. Distributed household photovoltaic and distributed wind power output has strong randomness and volatility. After being connected, it is easy to cause problems such as power flow reversal, voltage over-limit, line overload, and light / overload of distribution transformers, resulting in prominent risks to the safe and stable operation of the power grid.

[0004] (2) Existing power distribution network modeling is mostly static ledger-style modeling. The application of digital twin technology is not deep enough. There is a lack of real-time two-way mapping and high-precision state simulation between the physical power grid and the virtual model. It cannot adapt to the dynamic changes in the operating conditions of rural power distribution networks, and the scheduling decision lacks accurate digital mirror support.

[0005] (3) Traditional scheduling algorithms are designed for urban power grids but do not take into account the characteristics of rural distribution networks, such as low bandwidth, limited edge computing power, and dispersed sources and loads. They have problems such as slow convergence speed, easy to get trapped in local optima, and insufficient multi-objective coordination ability. They cannot take into account multiple needs such as new energy consumption, power grid security, low-carbon operation, and line loss optimization.

[0006] (4) The rural distribution network lacks a coordinated mechanism for adjustable resources (household energy storage, agricultural flexible loads, and distributed reactive power compensation devices). The existing dispatching methods have not achieved coordinated optimization of the entire process of source-grid-load-storage, resulting in low renewable energy consumption rate, high distribution network line loss rate, and insufficient carbon emission control capabilities, making it difficult to support the construction goals of the county-level rural low-carbon distribution network. Summary of the Invention

[0007] This invention aims to overcome the shortcomings of existing technologies and provide an intelligent digital twin collaborative scheduling method. Specifically, the invention aims to: construct a full-element digital twin mirror model of a 10kV and below distribution network in a county or rural area, achieving real-time bidirectional mapping and high-precision state simulation between the physical power grid and the virtual model, thus solving the problems of low modeling accuracy, disconnect between the virtual and physical systems, and inconsistent data quality in rural distribution networks; and propose a basketball team optimization algorithm. The Optimization Algorithm (BTOA) serves as the core scheduling optimization solver, addressing the slow convergence and susceptibility to local optima issues of traditional algorithms. Under the premise of satisfying the hard constraints of power grid security, it achieves multi-objective coordinated optimization of renewable energy consumption, low-carbon operation, and line loss optimization. It constructs a full-process scheduling framework of "digital twin simulation - multi-timescale optimization - hierarchical collaborative control - closed-loop feedback iteration," resolving core pain points in rural distribution networks such as voltage fluctuations, power flow imbalances, and poor source-grid-load-storage coordination after renewable energy integration. It develops engineering solutions adapted to the low-bandwidth, highly dispersed scenarios of rural distribution networks. While core components utilize traditional algorithms commonly used in the power system industry, the proposed BTOA algorithm is used only in the optimization solution stage, balancing practicality and innovation to support the implementation of overall solutions for low-carbon intelligent digital distribution network systems in counties and rural areas. This invention covers the entire chain of digital twin modeling of all elements of the distribution network, multi-scenario operation simulation, multi-objective optimization scheduling, and hierarchical collaborative control, providing core technical support for overall solutions for low-carbon intelligent digital distribution network systems in counties and rural areas.

[0008] In a first aspect, embodiments of the present invention provide an intelligent digital twin collaborative scheduling method, comprising the following steps:

[0009] S1: Collect multi-source data of the distribution network, and construct a full-element digital twin mirror model of the distribution network based on the multi-source data; the full-element digital twin mirror model serves as a virtual model of the distribution network in the digital space, and is used to realize bidirectional mapping and state inference between the physical power grid and the virtual model;

[0010] S2: Based on the digital twin mirror model, perform multi-scenario simulation and deduction of the operating status of the distribution network, output key operating indicators to characterize the operating status of the distribution network, and generate safety constraint boundaries for step S3 based on the simulation results.

[0011] S3: Construct a multi-time-scale collaborative scheduling framework, and within the security constraint boundary, use optimization algorithms to perform multi-objective collaborative optimization of the scheduling plan of the distribution network to generate an optimized scheduling scheme;

[0012] S4: Based on the optimized scheduling scheme, perform hierarchical collaborative control and send scheduling instructions to the physical power grid for execution;

[0013] S5: Collect real-time operational data and perform closed-loop feedback and iterative optimization on the multi-objective collaborative optimization of the digital twin mirror model and scheduling plan.

[0014] As a preferred implementation, in step S1, the multi-source data includes: power grid basic ledger data, real-time operation data, distributed new energy data, load and energy storage data, and environmental and external data; an edge computing and cloud-based collaborative acquisition architecture is adopted, and abnormal data is detected and cleaned in real time at the edge.

[0015] As a preferred implementation, the full-element digital twin mirror model constructed in step S1 includes a geometric and topological twin model, a device-level mechanism twin model, a source-load time-series prediction model, and a virtual-real interaction model.

[0016] The geometric and topological twin model is constructed based on the fusion modeling technology of building information modeling and geographic information system and graph theory search algorithm;

[0017] The device-level mechanism twin model is constructed based on classical equivalent circuit theory and parameter identification algorithm;

[0018] The source load time-series prediction model is constructed using a backpropagation neural network.

[0019] The virtual-real interaction model realizes data interaction between the physical power grid and the digital twin mirror model through the OPC unified architecture and the message queue telemetry transmission dual protocol communication interface.

[0020] In a preferred embodiment, step S2 specifically includes:

[0021] Multiple typical operating conditions are set up, and the Newton-Raphson method and time-domain simulation method are used to conduct multi-scenario full-time-domain simulation and deduction of the distribution network. The voltage of each node, line load rate, distribution transformer overload status and new energy absorption capacity are output as key operating indicators.

[0022] Based on the simulation results, the threshold over-limit discrimination method is used to classify and warn of line overload, voltage over-limit, and equipment failure risks into four levels: red, orange, yellow, and blue.

[0023] An entropy weight method-analytic hierarchy process coupled algorithm is adopted to construct an evaluation index system from multiple dimensions such as equipment health, grid redundancy, load importance, and new energy penetration rate. This system is used to assess the vulnerability of the distribution network, output the identification results of weak nodes and weak lines, and provide priority guidance for scheduling optimization in step S3.

[0024] In a preferred embodiment, step S3 includes: constructing a comprehensive objective function as the optimization objective of the multi-objective collaborative optimization, which is used to generate an optimized scheduling scheme under the premise of satisfying the hard constraints of power grid security;

[0025] The comprehensive objective function includes at least the objectives of minimizing line loss, minimizing carbon emissions, minimizing voltage deviation, and maximizing the absorption of new energy sources, and is weighted and summed by assigning dynamic weight coefficients to each sub-objective; the dynamic weight coefficients are adjusted based on the graded early warning results and vulnerability assessment results output in step S2.

[0026] In a preferred embodiment, the hard constraints on grid security include at least power balance constraints, node voltage constraints, equipment load constraints, energy storage system constraints, flexible load constraints, and grid interaction constraints; all optimization solutions must satisfy the hard constraints on grid security.

[0027] As a preferred implementation, the optimization algorithm in step S3 adopts the basketball team optimization algorithm, which constructs a search mechanism by simulating the four core behaviors of a basketball team: high-intensity training, fast-break strategy, dynamic positioning strategy, and boundary control. Its solution process includes:

[0028] Initialization steps: Define the scheduling scheme that satisfies the constraints as the players in the population, and set the population size, iteration parameters, and algorithm core parameters;

[0029] Iterative search steps: Players are divided into elite players and ordinary players based on fitness; a local search strategy simulating high-intensity training is executed for elite players; for ordinary players, a fast convergence search simulating fast break strategy or a global exploration search simulating dynamic positioning strategy is dynamically selected based on adaptive factors; the dynamic positioning strategy further includes a random position strategy and a diagonal position strategy.

[0030] Boundary control steps: Implement a boundary control strategy that simulates a basketball going out of bounds and being replayed for all updated players to correct solutions that exceed the constraint boundaries;

[0031] Termination judgment step: Iteratively execute the iterative search step and boundary control step until the termination condition is met, and output the globally optimal player position as the optimized scheduling scheme.

[0032] In a preferred embodiment, step S3 includes the multi-time-scale collaborative scheduling framework comprising:

[0033] The day-ahead optimization scheduling is performed with a time scale of 24 hours and a step size of 1 hour. Based on the day-ahead prediction results output by the source load time series prediction model, the basketball team optimization algorithm is used for global optimization to generate the day-ahead baseline scheduling plan.

[0034] The intraday rolling revision schedule has a time scale of 4 hours and a step size of 15 minutes. Based on the intraday rolling prediction results output by the source load time series prediction model, the basketball team optimization algorithm is used to roll revision of the previous day baseline scheduling plan to generate the intraday revised scheduling plan.

[0035] Real-time optimization control, with a time scale of 5 minutes and a step size of 1 minute, is based on real-time mapping of operating data from a digital twin model. It employs a proportional-integral closed-loop control method and uses the intraday revised scheduling plan as a benchmark to quickly respond to random fluctuations in source load.

[0036] In a preferred embodiment, step S4 includes constructing a hierarchical collaborative control system, specifically including:

[0037] The local control layer of the equipment consists of feeder terminal units, distribution transformer terminal units, energy storage converter controllers, new energy inverters and reactive power compensation controllers. It is used to receive upper-level instructions and execute fast closed-loop control at the equipment level, with a response time of no more than 100 milliseconds.

[0038] The regional collaborative control layer, with the distribution network digital twin platform as its core, is deployed on edge computing nodes and cloud servers to coordinate the collaborative operation of the entire process of source-grid-load-storage within the region.

[0039] The power grid interaction layer employs a coordinated control strategy of automatic generation control and automatic voltage control to interact with the upper-level power grid and receive and execute emergency control commands issued by the upper-level power grid.

[0040] In a preferred embodiment, step S5 specifically includes:

[0041] Establish a scheduling performance evaluation index system that includes renewable energy absorption rate, comprehensive line loss rate, carbon emission intensity, node voltage qualification rate, equipment fault early warning accuracy rate, and digital twin model fitting degree; among which, the scheduling performance evaluation index is determined based on the key operation indicators output in step S2.

[0042] The actual values ​​of each evaluation indicator in the scheduling effect evaluation index system are statistically analyzed at a preset period, and the actual values ​​are compared with the preset target values.

[0043] When the deviation between the actual value and the target value of any evaluation index exceeds the preset range, at least one of the following optimization measures shall be executed according to the direction and magnitude of the deviation: re-identify and correct the device parameters of the digital twin mirror model, adjust the dynamic weight coefficients in the comprehensive objective function, perform incremental training on the source load time series prediction model, and verify and optimize the core parameters of the basketball team optimization algorithm.

[0044] Secondly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0045] One or more processors;

[0046] Storage device for storing one or more programs;

[0047] When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent digital twin collaborative scheduling method described in any embodiment of the present invention.

[0048] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent digital twin collaborative scheduling method described in any embodiment of the present invention.

[0049] Compared with existing technologies, the present invention achieves the following beneficial effects:

[0050] (1) The BTOA proposed in this invention constructs a hierarchical adaptive search mechanism by simulating the four core behaviors of a basketball team. It has a strong ability to balance global exploration and local development, which can effectively solve the problem that traditional power grid scheduling algorithms are prone to getting stuck in local optima and slow convergence. It is suitable for the high-dimensional, multi-constraint, and nonlinear scheduling optimization scenario of rural power grids.

[0051] (2) Significant comprehensive benefits, fully supporting the construction of low-carbon distribution networks: Simulation results show that after adopting this method, the annual average consumption rate of new energy can reach more than 95%, the comprehensive line loss rate of distribution networks can be reduced by more than 15%, and the carbon emission intensity can be controlled within 50gCO2 / kWh; Theoretical analysis shows that this method can control the power fluctuation at the grid connection point within ±5%, the monthly qualification rate of node voltage can reach more than 99.5%, and the accuracy rate of equipment fault early warning can reach more than 95%, significantly improving the safe operation level, new energy consumption capacity and low-carbon operation benefits of rural distribution networks;

[0052] (3) Strong technical compatibility and adaptability to existing power grid hardware conditions: The technical solution of the present invention does not require large-scale transformation of existing equipment in rural power distribution networks. It is compatible with existing hardware devices such as FTU / TTU terminals, smart meters, and energy storage inverters. The digital twin mirror model can be quickly constructed based on existing power grid ledger data. BTOA can run stably on industrial-grade edge computing nodes and has strong scenario adaptability and industrialization promotion value. Attached Figure Description

[0053] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0054] Figure 1 This is a flowchart of a digital twin-driven intelligent management method for lifetime optimization of hybrid energy storage systems provided in an embodiment of the present invention;

[0055] Figure 2 This is a system architecture diagram provided in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of the digital twin mirror model structure in an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of the basketball team optimization algorithm (BTOA) in an embodiment of the present invention;

[0058] Figure 5 This is a schematic diagram of a multi-time-scale collaborative scheduling framework according to an embodiment of the present invention;

[0059] Figure 6 This is a schematic diagram of the hierarchical collaborative control system according to an embodiment of the present invention;

[0060] Figure 7 This is a schematic diagram of the closed-loop feedback iterative optimization process in an embodiment of the present invention;

[0061] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0062] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0063] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations (or steps) may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0064] Example 1

[0065] Figure 1 This is a flowchart of an intelligent digital twin collaborative scheduling method provided in Embodiment 1 of the present invention; Figure 2 This is a system architecture diagram provided in an embodiment of the present invention. (For example...) Figure 1 and Figure 2As shown, the present invention provides an intelligent digital twin collaborative scheduling method for county and rural 10kV and below distribution networks. Based on digital twin technology, it achieves low-carbon intelligent collaborative scheduling of distribution networks through five core steps: full-element mirror modeling, multi-scenario operation simulation, BTOA multi-objective optimization scheduling, hierarchical collaborative control, and closed-loop iterative optimization. Specifically, it includes the following steps:

[0066] S1: Collect multi-source data of the distribution network, and construct a full-element digital twin mirror model of the distribution network based on the multi-source data; the digital twin mirror model serves as a virtual model of the distribution network in the digital space, and is used to realize bidirectional mapping and state inference between the physical power grid and the virtual model;

[0067] Step S1 is used to realize multi-source data acquisition and full-element digital twin mirror modeling of county and rural power distribution networks.

[0068] Step S1 constructs a digital twin foundation based on "edge acquisition - cloud modeling - virtual-real interaction," achieving a high-precision digital mirror of all elements and operating conditions of the power distribution network, providing support for dispatching decisions. Specifically, it includes the following steps:

[0069] S11: Construction of a Multi-Source Heterogeneous Data Acquisition System

[0070] To address the challenges of numerous and geographically dispersed rural power distribution networks with limited network bandwidth, a traditional data acquisition architecture combining edge computing and cloud collaboration is adopted. Data preprocessing and compression transmission are performed at the edge, while data aggregation and modeling are conducted in the cloud. The collected data covers five main categories: basic power grid ledger data, real-time operational data, distributed renewable energy data, load and energy storage data, and environmental and external data.

[0071] (1) Basic power grid ledger data: rated parameters of equipment such as 10kV and below lines, distribution transformers, switchgear, and reactive power compensation devices; geographic information of towers and line routes in Geographic Information System (GIS); and power grid topology connection data.

[0072] (2) Real-time operating data: voltage, current, active / reactive power, power factor, frequency, transformer load rate, line loss, switch status, etc. of each node, with a collection frequency of ≥1 time / minute;

[0073] (3) Distributed new energy data: irradiance, wind speed, component temperature, real-time output, and inverter operating status data of household / village photovoltaic and distributed wind power;

[0074] (4) Load and energy storage data: real-time power of residential load, agricultural production load, and rural industrial and commercial load; State of Charge (SOC) value, charging and discharging power of household energy storage; and operating parameters of adjustable loads (irrigation equipment, electric heating, and charging piles).

[0075] (5) Environmental and external data: regional meteorological data, geographical environment data, time-of-use electricity price of the upper-level power grid, marginal carbon emission factor, and power grid dispatch instruction data.

[0076] Data sources include smart meters, feeder terminal units (FTUs), transformer terminal units (TTUs), new energy inverter monitoring systems, household energy storage management systems (EMS), regional weather stations, and GIS systems. Historical data collection spans at least one year and is used for model training and parameter calibration.

[0077] At the edge, the traditional 3σ criterion plus moving average filtering method is used to complete the real-time detection and cleaning of abnormal data, eliminating jump values, missing values ​​and noisy data in the collection process. The data cleaning accuracy is ≥99%, providing high-quality input data for subsequent source-load prediction models and solving the problem of inconsistent data quality in rural power distribution networks.

[0078] In this embodiment of the invention, the distribution network refers to the 10kV and below power grid in counties and rural areas. The distribution network is an abstract concept encompassing its physical entity, topology, equipment parameters, etc., and differs slightly from the physical power grid in step S4. The physical power grid refers to a real-world, physical power network, as opposed to a virtual model; it is a concrete entity emphasizing its physical existence. The distribution network is a specific instance of the physical power grid at a particular voltage level and scenario; the physical power grid is the physical entity of the distribution network.

[0079] S12: Digital Twin Mirror Modeling of All Elements in Distribution Network

[0080] The full-element digital twin mirror model includes a geometric and topological twin model, an equipment-level mechanistic twin model, a source-load time-series prediction model, and a virtual-physical interaction model. The geometric and topological twin model is constructed based on the fusion modeling technology of Building Information Modeling (BIM) and Geographic Information System (GIS) and graph theory search algorithms. The equipment-level mechanistic twin model is constructed based on classical equivalent circuit theory and parameter identification algorithms. The source-load time-series prediction model is constructed using a backpropagation neural network. The virtual-physical interaction model realizes data interaction between the physical power grid and the digital twin mirror model through a unified OPC architecture and a dual-protocol communication interface for message queue telemetry transmission. Four core twin sub-models are constructed, with traditional power system engineering algorithms as the core, and only the source-load prediction uses a basic 3-layer BP neural network.

[0081] In this embodiment of the invention, "full elements" refers to the completeness of all aspects of the distribution network covered by the digital twin mirror model. The digital twin model not only includes a single dimension (such as geometry alone or only ledgers), but also covers multiple dimensions of the distribution network, including geometry, topology, equipment mechanisms, operational sequence, and virtual-physical interaction, forming a complete, high-precision digital mirror. Specifically, it includes:

[0082] (1) Geometric elements: 1:1 geometric mirror of the equipment (based on BIM+GIS modeling);

[0083] (2) Topological elements: electrical connection relationships, switch states, etc. (based on graph theory DFS algorithm);

[0084] (3) Equipment mechanism elements: equivalent circuit models and parameters of equipment such as transformers and lines;

[0085] (4) Source-load time series elements: Time series prediction model of distributed new energy output and load changes;

[0086] (5) Virtual-real interaction elements: a two-way data interaction mechanism between the physical power grid and the virtual model.

[0087] like Figure 3 The diagram shown is a schematic representation of the digital twin mirror model structure provided in an embodiment of the present invention. The digital twin mirror model is as follows:

[0088] S121: Geometric and Topological Twin Model

[0089] Based on GIS and equipment ledger data, a 1:1 geometric mirror of 10kV lines, distribution transformers, low-voltage distribution areas, and user-side equipment is constructed using Building Information Modeling (BIM) + GIS fusion modeling technology. The electrical topology twin model is constructed using the traditional graph theory depth-first search (DFS) algorithm to map the switching status and topology connection changes of the physical power grid in real time, with a topology update delay of ≤1s.

[0090] S122: Device-level Mechanism Twin Model

[0091] For core equipment such as distribution transformers, transmission lines, reactive power compensation devices, and inverters, a mechanistic model is constructed based on the classical equivalent circuit theory of power systems. The traditional least squares (LS) method is used to identify and dynamically correct equipment parameters, solving the problems of aging equipment parameters and deviations in measured data in rural power distribution networks. The model output has a good fit with the measured data of physical equipment of ≥98%, and the model update frequency is ≥1 time / day.

[0092] S123: Source Load Time Series Prediction Model

[0093] The most common 3-layer back propagation (BP) neural network in the power system industry is used to complete the time-series forecasting of distributed renewable energy and load, providing basic input for scheduling optimization. The model structure is extremely simple: input layer - single hidden layer - output layer, without complex network structure.

[0094] (1) Input characteristics: historical meteorological data, historical output / load data, date type, agricultural production cycle;

[0095] (2) Output content: 24-hour forecast value, 15-minute rolling forecast value, and 5-minute ultra-short-term forecast value;

[0096] (3) Model training details: Batch gradient descent is used to complete the weight iteration, with mean square error (MSE) as the loss function, the learning rate is set to 0.01, and an L2 regularization coefficient of 0.001 is added to prevent overfitting. The batch size is set to 32, and the maximum number of iterations is 50 to complete the model convergence training; Alternatively, the default training parameters of the feedforwardnet network in the MATLAB Deep Learning Toolbox can be used directly to complete the model construction.

[0097] (4) Accuracy description: Under ideal working conditions where the data quality meets the standards after cleaning, the relative error of the day-ahead prediction is ≤15%, the relative error of the intraday rolling prediction is ≤8%, and the relative error of the ultra-short-term prediction is ≤5%, which meets the needs of rural power distribution network dispatching projects.

[0098] S124: Virtual-Real Interaction and Data Cleaning Model

[0099] A dual-protocol communication interface of OPC Unified Architecture (OPC UA) and Message Queuing Telemetry Transport (MQTT) is constructed to realize bidirectional data interaction between the physical power grid and the twin model, with an interaction latency of ≤500ms. The traditional 3σ criterion + moving average filtering method is used to complete the secondary anomaly detection and correction of real-time operating data to ensure the validity of the data input to the twin model.

[0100] S2: Based on the digital twin mirror model, perform multi-scenario simulation and deduction of the operation status of the distribution network, output key operation indicators to characterize the operation status of the power grid, and generate safety constraint boundaries for step S3 based on the simulation results.

[0101] Step S2 is used to realize the simulation of the operation status of the distribution network and the early warning of safety risks based on digital twins, providing constraint boundaries and risk prediction for scheduling optimization. Further, step S2 includes: setting multiple typical operating conditions, using the Newton-Raphson method and time-domain simulation method to perform multi-scenario full-time-domain simulation of the distribution network, outputting the voltage of each node, line load rate, transformer overload status, and renewable energy absorption capacity as key operating indicators; based on the simulation results, using the threshold over-limit discrimination method to classify line overload, voltage over-limit, and equipment failure risks into four levels: red, orange, yellow, and blue; using the entropy weight method-analytic hierarchy process coupled algorithm, constructing an evaluation index system from multiple dimensions such as equipment health, network redundancy, load importance, and renewable energy penetration rate, to conduct vulnerability assessment of the distribution network, outputting the identification results of weak nodes and weak lines, and providing priority guidance for scheduling optimization in step S3. Specifically, it includes the following steps:

[0102] S21: Digital Twin Simulation of Multi-Scenario Operation Status

[0103] Based on the constructed full-element digital twin mirror model, five typical working conditions are set up for full-time domain simulation and deduction:

[0104] (1) New energy generation scenarios (full photovoltaic power generation at midday, low load);

[0105] (2) Peak load scenario (peak residential load in the evening, and zero output of new energy sources);

[0106] (3) Extreme weather scenarios (strong winds, heavy rain, low temperature freezing damage);

[0107] (4) Equipment failure scenarios (line breakage, transformer overload);

[0108] (5) Power grid maintenance scenario.

[0109] The system employs the classic Newton-Raphson method for power system steady-state power flow calculations and the traditional time-domain simulation method for transient simulation deductions. It outputs key operating indicators such as node voltage, line load rate, transformer overload, line loss rate, and renewable energy absorption capacity. The simulation step size is configurable (15 min to 24 h), providing constraint boundaries for dispatch optimization.

[0110] S22: Security Risk Classification, Early Warning, and Vulnerability Assessment

[0111] S221: Safety Risk Classification and Early Warning

[0112] Based on the State Grid's "Distribution Network Operation Regulations", equipment operation thresholds are set, and the traditional threshold over-limit judgment method is used to complete the safety risk classification and early warning. It can provide red, orange, yellow and blue four-level early warnings for risks such as line overload, voltage over-limit, equipment failure and power flow imbalance 15 minutes to 24 hours in advance.

[0113] S222: Vulnerability Assessment of Distribution Networks

[0114] A coupled algorithm of traditional entropy weight method and analytic hierarchy process (AHP) is used to complete the vulnerability assessment of distribution networks. An assessment index system is constructed from four dimensions: equipment health, network redundancy, load importance, and new energy penetration rate. This accurately identifies weak nodes and weak lines in rural distribution networks and provides priority guidance for dispatch optimization.

[0115] S3: Construct a multi-time-scale collaborative scheduling framework, and within the security constraint boundary, use optimization algorithms to perform multi-objective collaborative optimization of the scheduling plan of the distribution network to generate an optimized scheduling scheme;

[0116] Step S3 is used to implement multi-timescale, multi-objective cooperative scheduling optimization based on the BTOA proposed in this invention.

[0117] This invention proposes a Basketball Team Optimization Algorithm (BTOA), which simulates four core behaviors of a basketball team: high-intensity training, fast-break strategy, dynamic positioning strategy, and boundary reentry. It constructs a three-level multi-timescale scheduling framework of "day-ahead optimization - intraday rolling correction - real-time optimization control" to achieve multi-objective collaborative optimal scheduling of the distribution network under the premise of meeting the hard constraints of power grid security.

[0118] S31: Constructing a multi-objective optimization comprehensive objective function

[0119] The core principles are clear: the requirements related to power grid safety, such as node voltage, equipment load, and power balance, are hard constraints that cannot be broken, and are rigidly limited by the constraints in step S32; the objective function is a multi-dimensional soft optimization that satisfies the hard constraints, and the optimization priority under different operating conditions is adjusted by weighting coefficients.

[0120] With low-carbon operation of rural power distribution networks as the core, and taking into account economic operation, renewable energy consumption, and voltage quality optimization, a multi-objective optimization function is constructed. The optimization variables are: energy storage charging and discharging power, flexible load adjustment, reactive power compensation device switching capacity, and renewable energy active / reactive power output adjustment in each time period.

[0121] The comprehensive objective function is used to optimize the operation performance of the distribution network while meeting the hard constraints of power grid security. The comprehensive objective function includes at least the objectives of minimizing line losses, minimizing carbon emissions, minimizing voltage deviation, and maximizing renewable energy absorption. These objectives are weighted and summed by assigning dynamic weight coefficients to each sub-objective. The dynamic weight coefficients are adjusted based on the graded early warning results and vulnerability assessment results output in step S2. as follows:

[0122]

[0123] in, , , , For dynamic weighting coefficients, satisfying The operational status and risk level are dynamically adjusted based on digital twin simulations. The adjustment rules are as follows:

[0124] 1. If the renewable energy consumption rate is lower than ,but The remaining weights are equally distributed, with priority given to improving the level of new energy consumption;

[0125] 2. If the carbon emission intensity exceeds the standard, then The remaining weights are equally distributed, with priority given to ensuring low-carbon operation goals;

[0126] 3. If a voltage over-limit or line overload warning occurs (without exceeding the hard constraint boundary), then The remaining weights are evenly distributed, with priority given to optimizing voltage quality and reducing equipment overload risk;

[0127] 4. Under normal operating conditions, , , , Prioritize low-carbon operation.

[0128] The specific definitions of each sub-objective function are as follows:

[0129] 1. Minimize line loss :

[0130]

[0131] In the formula, The total time step of the scheduling cycle (unit: h). Total number of lines (unit: lines) for Time of the first Current of the line (unit: A). For the first Resistance of the line (unit: ), The scheduling time step (unit: h);

[0132] 2. Carbon emission minimization target :

[0133] The calculation boundary is the carbon emissions corresponding to the net purchase of electricity from the upstream power grid by the county and rural distribution networks. Distributed renewable energy self-consumption is not included in carbon emissions. Formula:

[0134]

[0135] In the formula, for Net power purchased by the distribution network from the upper-level grid at any given time (unit: kW, negative values ​​are counted as 0 for grid connection). for Marginal carbon emission factor of upstream power grid at any time (unit: (Data taken from provincial carbon trading platforms)

[0136] 3. Voltage deviation minimization objective :

[0137]

[0138] In the formula, Total number of distribution network nodes (unit: nodes). for Time of the first Voltage of each node (unit: kV). Rated voltage of the power grid (unit: );

[0139] 4. Maximizing the absorption of new energy sources :

[0140] By incorporating the negative value into the minimization objective, the absorption rate can be maximized.

[0141]

[0142] In the formula, for Local absorption capacity of new energy sources at any time (unit: kW) for Total output power of new energy sources at any time (unit: kW).

[0143] S32: Scheduling Constraints

[0144] To align with the actual operation of 10kV and below power distribution networks in rural counties, a full-dimensional hard constraint boundary is set. All optimization solutions must satisfy the following constraints; otherwise, they are considered invalid solutions:

[0145] 1. Power balance constraint: In the formula, Power exchanged with the upstream power grid (unit: kW). Energy storage discharge power (unit: kW). Energy storage charging power (unit: kW). Total load power (unit: kW). Total network loss power (unit: kW);

[0146] 2. Node voltage constraints: (Complies with the requirements of GB / T 12325-2008 national standard);

[0147] 3. Equipment load constraints: Line load rate ≤ 100%, distribution transformer load rate ≤ 80% (with safety margin reserved);

[0148] 4. Constraints of energy storage systems: The charging and discharging power shall not exceed the rated maximum value, and the number of cycles per day shall be ≤2.

[0149] 5. Flexible load constraints: The duration of a single adjustment is ≥30 minutes, and the total daily adjustment is ≤20% of the total daily load, ensuring the basic electricity needs of residents and agricultural production;

[0150] 6. Power grid interaction constraints: The grid connection capacity is set according to the approved grid connection capacity of the superior power grid (unit: kW) to avoid large power fluctuations.

[0151] S33: Multi-objective optimization solution based on BTOA

[0152] The BTOA proposed in this invention constructs an efficient search mechanism by simulating the four core behaviors of a basketball team, thereby achieving rapid optimization of the global optimal solution.

[0153] The solution process of the basketball team optimization algorithm BTOA includes: Initialization step: defining the scheduling scheme that satisfies the constraints as the players in the population, and setting the population size, iteration parameters, and core algorithm parameters; Iterative search step: dividing players into elite players and ordinary players according to fitness; performing a local search strategy simulating high-intensity training for elite players; and dynamically selecting between a fast convergence search simulating a fast break strategy or a global exploration search simulating a dynamic positioning strategy based on an adaptive factor for ordinary players; the dynamic positioning strategy further includes a random position strategy and a diagonal position strategy; Boundary control step: performing a boundary control strategy simulating a basketball going out of bounds and replaying for all updated players to correct solutions that exceed the constraint boundaries; Termination judgment step: iteratively executing the iterative search step and boundary control step until the termination condition is met, and outputting the globally optimal player position as the optimized scheduling scheme. Figure 4 This is a flowchart illustrating the Basketball Team Optimization Algorithm (BTOA) in an embodiment of the present invention. The specific solution process is as follows:

[0154] S331: BTOA Algorithm Initialization

[0155] (1) Population definition: Each group of scheduling optimization variables (energy storage charging and discharging plan, load adjustment, etc.) that satisfy the hard constraints of step S32 is defined as a "player", and the population size is TeamSize. (Adapted to edge computing power of rural power distribution networks);

[0156] (2) Iteration parameter: maximum number of iterations Set by time scale: 200 times for daily optimization, 80 times for intraday rolling optimization, and 30 times for real-time optimization;

[0157] (3) Optimization dimension: variable dimension ,in Number of energy storage units (unit: units). The number of flexible loads (unit: units). Number of reactive power compensation devices (unit: units);

[0158] (4) Boundary setting: Based on the constraints in step S32, set the upper and lower bounds for each optimization variable. , (The units should be consistent with the corresponding optimization variables);

[0159] (5) Fitness function: based on the comprehensive objective function of step S31 As a fitness function, the smaller the fitness value, the better the scheduling scheme corresponding to that "player";

[0160] (6) Core parameters: (Adaptive factor amplification coefficient) (Adaptive factor decay coefficient), strongnum TeamSize (Number of players in high-intensity training, unit: players).

[0161] S332: BTOA Core Search Strategy and Mathematical Model

[0162] During the algorithm iteration process, players are ranked according to their fitness and divided into elite players and ordinary players, and different search strategies are implemented for each. The core mathematical model is as follows:

[0163] High-intensity training strategy

[0164] This algorithm only applies to elite players with the highest fitness ranking (strongnum), simulating the high-intensity training process of elite basketball players using the best player as a benchmark. This achieves localized, refined optimization of the scheduling scheme. Mathematical model:

[0165] ;

[0166] In the formula: For the players Current position (current scheduling scheme, dimension) ), For the globally optimal player positions (globally optimal scheduling scheme, dimension) ), and For two randomly selected player positions (dimensions) );

[0167] The fatigue factor (dimensionless) dynamically decreases with the number of iterations, balancing exploration and development. The formula is:

[0168] ;

[0169] in This represents the current iteration number (in units of iterations). Total number of iterations (unit: iterations). The value decreased from 0.9 to 0.1 with each iteration.

[0170] The interference elasticity coefficient (dimensionless) is linked to player fitness rankings, and the formula is:

[0171] ;

[0172] in Players are ranked by their adaptability (unit: number). The higher the ranking, the better. The smaller the value, the stronger the anti-interference ability;

[0173] For binary variable attribute masks (dimensions) ), controls the adjustable variable dimensions of the player, determined by the maximum variable attribute coefficient. Generate, formula:

[0174] ;

[0175] in, To round up. CreateBinaryVector The function generates a length of ,Include A binary vector of 1s; the lower the player's ranking, the more dimensions can be adjusted, and the stronger their global exploration ability.

[0176] (2) Fast Break Strategy

[0177] For ordinary players, through adaptive factors Determining whether to execute a fast break strategy involves simulating the behavior of rapidly approaching the basket during a basketball fast break, and achieving rapid convergence of the scheduling scheme towards the optimal solution. Mathematical model:

[0178] ;

[0179] In the formula: for Random number vector of an interval (dimension) ), introducing search randomness; To adjust the dimensionless factor, which dynamically changes with the number of iterations to simulate improved shooting accuracy, the formula is:

[0180]

[0181] The value decreases from 2 to 1 with each iteration, and the later stages of the iteration focus on local development. The dimensionless item simulates fast attack defense interference avoidance. The interference item has a larger weight in the early stage of iteration, which enhances global exploration.

[0182] (3) Dynamic Positioning Strategy

[0183] For ordinary players, adaptive factor Execution is performed in real time, and the global space is fully explored by guiding the algorithm through a diagonal structure, which solves the problem that traditional algorithms are prone to getting trapped in local optima. It is divided into random position strategy and diagonal position strategy:

[0184] ① Random Position Strategy: Randomly select a player and dimension, projecting them onto the diagonal of the search space to generate candidate solutions. Suppose a player is randomly selected from the population. Randomly select one dimension from the total dimension D. Calculate the player's projected position on the diagonal of the search space:

[0185] ;in, ; ;

[0186] based on and the global optimal solution Generate intermediate solutions Finally, candidate solutions are generated: ;

[0187] In the formula, The projected position of a random player on the diagonal (dimensions) ), For candidate solutions (dimensions) ), A random binary vector of 0-1 (dimension) ), introduce random perturbations; The scheduling scheme (position) corresponding to a randomly selected "player" from the population, and the total dimension of the optimization variables. Consistent; RandomPlayer(TeamSize): Random player selection function, the input is the population size TeamSize, the function is to randomly select 1 "player" from all "players" and output the corresponding scheduling scheme (position) of the "player". : Randomly selected optimization variable dimensions, i.e., from the total dimensions RandomDimension is a single dimension randomly selected from the data. : Random dimension selection function, the input is the total dimension of the optimization variables. The function is from Randomly select one dimension from the optimization dimensions and output the index of that dimension; The projection position of a random player on the diagonal of the search space, with dimension 1. , is an intermediate variable used to generate candidate solutions; The lower bound of the optimization variable, with units consistent with the corresponding optimization variable (energy storage charging and discharging power, flexible load adjustment, etc.); : The upper bound of the optimization variable, with the same unit as the corresponding optimization variable; Random players In random dimension The specific numerical values ​​are consistent with the units of the optimization variables corresponding to that dimension. : Optimization variables in the random dimension The lower bound of the above, with units consistent with the optimization variables corresponding to that dimension; : Optimization variables in the random dimension The upper bound of the dimension is consistent with the unit of the optimization variable corresponding to that dimension. Based on random player projection positions and the global optimal solution The generated intermediate solution has a dimension of rand: A random number vector of intervals, with dimension 1. This is used to introduce randomness into the update of the solution; The globally optimal player position, i.e., the optimal scheduling scheme in the current iteration, has the following dimensions: ; Candidate scheduling schemes (candidate solutions) generated by the random location strategy, with dimensions of ; : A random binary vector of 0-1, with dimension Used for candidate solutions The generation introduces random perturbations; : The ordinary player currently implementing the dynamic positioning strategy Location (current scheduling scheme), dimension is .

[0188] ② Diagonal Position Strategy: Project the global optimal solution onto the diagonal, and generate candidate solutions based on symmetric expansion. Formula:

[0189] ;

[0190] ;

[0191] ;

[0192] In the formula, The dimension of the randomly selected optimal solution (unit: dimension). The projection position (dimension) of the optimal solution onto the diagonal. ), For candidate solutions (dimensions) ); Global optimal solution The projection position on the diagonal of the search space, with dimension [missing information]. It generates candidate solutions. Intermediate variables; The dimension of the randomly selected optimal solution, i.e., from the total dimension. One dimension (unit: dimension) that is randomly selected from the data and is related to the global optimal solution. Global optimal solution In random dimension The specific numerical values ​​are consistent with the units of the optimization variables corresponding to that dimension. : Optimization variables in the random dimension The lower bound of the above, with units consistent with the optimization variables corresponding to that dimension; : Optimization variables in the random dimension The upper bound of the dimension is consistent with the unit of the optimization variable corresponding to that dimension. : Random vectors generated based on the upper and lower bounds of the optimization variables, with dimension . Used to determine candidate solutions The symmetrical expansion direction; Candidate scheduling schemes (candidate solutions) generated by the diagonal position strategy, with dimension . ; Candidate solutions In the The specific numerical value of the dimension, and the unit compared to the first dimension. The optimization variables corresponding to the dimensions are consistent; Projection position In the The specific numerical value of the dimension, and the unit of the dimension. The optimization variables corresponding to the dimensions are consistent; : The optimization variable is in the th The upper bound of the dimension, the unit and the first The optimization variables corresponding to the dimensions are consistent; : The optimization variable is in the th The lower bound of the dimension, the unit and the first The optimization variables corresponding to the dimensions are consistent; Random vector In the The specific numerical value of the dimension.

[0193] ③ Selection of optimal candidate solutions: comparison and Based on the fitness value, select the better solution as the new position:

[0194] ;

[0195] Players The updated new location (new scheduling scheme) has the following dimensions: From candidate solutions and Those with better fitness were selected; Candidate solutions The fitness value is based on the comprehensive objective function in step S31. The smaller the value, the better the scheduling scheme. Candidate solutions The fitness value is based on the comprehensive objective function in step S31. The smaller the value, the better the scheduling scheme. : Optimize the total dimension of the variable, its value is ( The total time step of the scheduling cycle. The number of energy storage units. For the number of flexible loads, (Number of reactive power compensation devices).

[0196] (4) Strategy switching adaptive factor

[0197] Used for dynamic switching between fast-attack and dynamic positioning strategies, balancing global exploration and local exploitation. Formula:

[0198] ;

[0199] : Strategy switching adaptive factor (dimensionless), used to dynamically switch between fast break strategies and dynamic positioning strategies for ordinary players, balancing the algorithm's global exploration and local development capabilities; : Adaptive factor amplification coefficient (dimensionless), a core parameter of the BTOA algorithm, with a default value of 5; : Adaptive factor decay coefficient (dimensionless), a core parameter of the BTOA algorithm, with a default value of 1; Adaptability ranking of ordinary players (unit: rank), the higher the ranking, the better. The higher the probability; TeamSize: the population size of the BTOA algorithm, i.e. the total number of "players", which is 40 when adapted to the edge computing power of rural power distribution networks; : Natural exponential function, used to achieve exponential decay adjustment of adaptive factors; : Logarithmic function, used to introduce randomness and adjust the numerical range of the adaptive factor; rand: Scalar random numbers (dimensionless) within an interval, calculated each time. Regenerated periodically, introducing randomness into policy switching.

[0200] like Execute a fast-attack strategy; if A dynamic positioning strategy is implemented. Under default parameters, players ranked higher receive more favorable positioning. The higher the probability, the more priority should be given to local development; the lower the ranking of the player, The higher the probability, the higher the priority for global exploration.

[0201] (5) Boundary control strategy (Ball Re-entry)

[0202] The rules for re-inbounding a basketball after it goes out of bounds are simulated. Solutions that exceed the boundary are reset, historical information is preserved to avoid oscillations, and the mathematical model is as follows:

[0203] ;

[0204] In the formula: Solution in the first place The numerical value of the dimension (corresponding to the unit of the optimization variable, such as the unit of energy storage charging and discharging power kW). : No. The upper bound of the optimization variable (corresponding to the unit of the optimization variable); : No. The lower bound of the optimization variable (corresponding to the unit of the optimization variable); Solution in the first place Historical values ​​of the dimension (corresponding to the unit of the optimization variable) are used to retain historical information during the iteration process; : Independent scalar random numbers (dimensionless) in an interval, when This is used to reset the solution for linear interpolation, and it is regenerated each time the solution is out of bounds. : Independent scalar random numbers (dimensionless) in an interval, when This is used to reset the solution for linear interpolation, and it is regenerated each time the solution is out of bounds. : Dimension number of the optimization variable (unit: dimension), with a value range of . ( To optimize the total dimension of variables.

[0205] S333: Termination condition of BTOA algorithm iteration

[0206] The iteration terminates and the globally optimal scheduling scheme is output when any of the following conditions are met:

[0207] (1) The number of iterations reaches the set maximum number of iterations, MaxGen;

[0208] (2) The algorithm converges when the change in the global optimal fitness value over 20 consecutive iterations is ≤1e-6.

[0209] In summary, based on the BTOA algorithm principle and mathematical model described in steps S331 to S333 above, the complete optimization solution process of the BTOA algorithm is as follows: First, initialize the player population that meets the constraints and calculate the initial fitness; then, in each iteration, after sorting by fitness, perform a high-intensity training strategy for elite players to conduct a local fine-tuning search, and dynamically select either a fast-attack strategy or a dynamic positioning strategy to update the positions of ordinary players based on the adaptive factor; after all players are updated, a boundary control strategy is executed to correct out-of-bounds solutions, and it is verified whether the hard constraints are met; finally, iterate until the termination condition is met, and output the globally optimal scheduling scheme. This process has verified its effectiveness and engineering applicability in multiple large-scale distribution network examples.

[0210] Step S34: Three-level multi-timescale scheduling framework

[0211] Combining the operational characteristics of rural power distribution networks with edge computing power constraints, a three-level multi-time-scale scheduling framework is constructed to adapt to the solution efficiency of BTOA:

[0212] like Figure 5 The diagram shown is a schematic representation of a multi-time-scale collaborative scheduling framework in an embodiment of the present invention. The multi-time-scale collaborative scheduling framework includes:

[0213] (1) Day-ahead optimization scheduling (time scale 24h, step size 1h): Based on the day-ahead source load prediction results (i.e. day-ahead prediction results), the basketball team optimization algorithm BTOA is used for global optimization. The population size is 40, the maximum number of iterations is 200, and the output includes the day-ahead 24h energy storage charging and discharging, flexible load adjustment, and reactive power compensation switching, so as to achieve the daily cycle global optimum.

[0214] (2) Intraday rolling correction scheduling (time scale 4h, step size 15min): Based on the intraday 15min rolling prediction results (i.e., intraday rolling prediction results), the simplified basketball team optimization algorithm BTOA (population size 20, maximum number of iterations 80) is adopted. Combined with the real-time simulation results of digital twin, the previous day's baseline scheduling plan is rolled and corrected every 15 minutes. The hot start strategy is adopted to accelerate the solution, which can meet the computational timeliness requirements of intraday rolling scheduling and generate intraday corrected scheduling plan. This module is deployed on edge computing nodes with industrial-grade computing power (such as Kunpeng series edge servers, NVIDIA Jetson series edge computing units) to ensure the real-time performance of optimization solution.

[0215] (3) Real-time optimization control (time scale 5min, step size 1min): Based on the real-time mapping of the digital twin, the traditional proportional-integration (PI) closed-loop control is adopted. Based on the intraday revised scheduling plan, it can quickly respond to random fluctuations of source load and correct voltage over-limit and line overload problems. In the industrial Ethernet or 5G communication environment, the end-to-end control delay can be controlled within 200ms.

[0216] S4: Based on the optimized scheduling scheme, perform hierarchical collaborative control and send scheduling instructions to the physical power grid for execution;

[0217] Step S4 is used to execute the "virtual-physical collaboration" hierarchical closed-loop control strategy. By constructing a three-level hierarchical collaborative control system, the precise implementation of dispatching instructions and the safe and stable operation of the physical power grid are achieved. The three-level hierarchical collaborative control system specifically includes: a local equipment control layer, consisting of feeder terminal units, distribution transformer terminal units, energy storage converter controllers, new energy inverters, and reactive power compensation controllers, used to receive upper-level instructions and execute equipment-level fast closed-loop control with a response time of no more than 100ms; a regional collaborative control layer, with the distribution network digital twin platform as the core, deployed on edge computing nodes and cloud servers, used to coordinate the collaborative operation of the entire process of source-grid-load-storage within the region; and a grid interaction layer, which adopts a collaborative control strategy of automatic generation control and automatic voltage control, used to interact with the upper-level power grid and receive and execute emergency control instructions issued by the upper-level power grid. Figure 6 This is a schematic diagram of a hierarchical collaborative control system in an embodiment of the present invention. The specific details of the hierarchical collaborative control system are as follows:

[0218] S41: Device Local Control Layer

[0219] Composed of FTU / TTU terminals, Power Conversion System (PCS) controllers, new energy inverters, and reactive power compensation controllers, it adopts traditional PI / Proportional-Integration-Differentiation (PID) closed-loop control and constant power / constant voltage droop control. It receives upper-level scheduling commands to achieve fast equipment-level closed-loop control with a response time of ≤100ms. At the same time, it collects equipment operation data and uploads it to the digital twin mirror model in real time.

[0220] S42: County / Township Collaborative Control Layer

[0221] With the distribution network digital twin platform as the core, deployed on edge computing nodes and cloud servers, it is responsible for receiving the optimized scheduling plan output by BTOA, coordinating the collaborative operation of the entire process of power generation, grid, load and storage within the jurisdiction, adopting traditional power balance control and reactive power and voltage optimization control strategies, dynamically issuing control commands to the local control layer, and receiving dispatch commands from the upper-level power grid to complete emergency control tasks.

[0222] S43: Upper-level power grid interaction layer

[0223] Following the power grid dispatching procedures, a traditional Automatic Generation Control (AGC) / Automatic Voltage Control (AVC) collaborative control strategy is adopted to upload distribution network operation data to the upper-level power grid in real time (upload frequency 1 time / minute), receive emergency control commands, and quickly adjust energy storage, flexible loads and new energy output through the collaborative control layer to achieve friendly interaction between the distribution network and the upper-level power grid.

[0224] S5: Collect real-time operational data and perform closed-loop feedback and iterative optimization on the multi-objective collaborative optimization of the digital twin mirror model and scheduling plan.

[0225] Step S5 is used to realize closed-loop feedback and iterative optimization of the digital twin scheduling system. Traditional statistical analysis and numerical verification methods are used to achieve closed-loop optimization of the scheduling system. Step S5 includes: establishing a scheduling effect evaluation index system including new energy absorption rate, comprehensive line loss rate, carbon emission intensity, node voltage qualification rate, equipment fault early warning accuracy rate, and digital twin model fitting degree; wherein, the scheduling effect evaluation index is determined based on the key operating indicators output in step S2; statistically analyzing the actual values ​​of each evaluation index in the scheduling effect evaluation index system at a preset period and comparing the actual values ​​with preset target values; when the deviation between the actual value and the target value of any evaluation index exceeds a preset range, at least one of the following optimization measures is executed according to the direction and magnitude of the deviation: re-identifying and correcting the equipment parameters of the digital twin mirror model, adjusting the dynamic weight coefficients in the comprehensive objective function, incrementally training the source-load time-series prediction model, and verifying and optimizing the core parameters of the basketball team optimization algorithm. Figure 7 This is a schematic diagram of the closed-loop feedback iterative optimization process in an embodiment of the present invention, which specifically includes the following steps:

[0226] S51: Evaluation Index System for Scheduling Effectiveness

[0227] Establish core evaluation indicators suitable for low-carbon power distribution networks in counties and rural areas:

[0228] (1) New energy consumption rate (unit: %)

[0229] (2) Comprehensive line loss rate of distribution network (unit: %);

[0230] (3) Carbon emission intensity index of power distribution network (unit: gCO2 / kWh);

[0231] (4) Node voltage qualification rate (unit: %);

[0232] (5) Equipment fault early warning accuracy rate index (unit: %);

[0233] (6) Digital twin model fit index (unit: %).

[0234] S52: Closed-loop iterative optimization

[0235] Daily statistical scheduling performance evaluation metrics are used to complete closed-loop optimization of the entire process based on metric deviations:

[0236] (1) The renewable energy consumption rate does not meet the standard: optimize the target weight and expand the range of flexible load adjustment and the energy storage constraint boundary;

[0237] (2) Exceeding carbon emission intensity: Increase the weight of carbon emission targets, optimize the proportion of self-generated and self-consumed new energy sources, and reduce power purchases from the grid during peak hours;

[0238] (3) Voltage qualification rate is not up to standard: optimize reactive power compensation scheduling strategy and PI control parameters;

[0239] (4) Insufficient model fit: The least squares method was re-used to identify and correct the equipment model parameters;

[0240] (5) Based on actual operation data, the traditional error backpropagation method is used to complete the incremental training of the BP neural network prediction model, and the traditional statistical method is used to complete the verification and optimization of the parameters of the basketball team optimization algorithm BTOA, so as to realize the self-iteration and self-optimization of the scheduling system.

[0241] S6: Performance Verification and Comparison Experiment of Basketball Team Optimization Algorithm

[0242] To verify the performance advantages and versatility of the Basketball Team Optimization Algorithm (BTOA) proposed in this invention in the scenario of rural power distribution network dispatch optimization, a multi-case, multi-dimensional quantitative comparison experiment was designed. From four core dimensions—optimization accuracy, operational stability, convergence speed, and engineering time—a fair comparison was made with mainstream metaheuristic algorithms in the field of power distribution network optimization to ensure the rigor and credibility of the experimental conclusions.

[0243] S61: Experimental Design

[0244] S611: Case Setup

[0245] To avoid the randomness of a single case, three sets of 10kV distribution networks in rural areas of different sizes and with different renewable energy penetration rates were selected as test cases. The topology, equipment parameters, and source-load characteristics of the cases are all consistent with the actual operating characteristics of rural distribution networks. The specific parameters are as follows:

[0246] (1) Example 1 (small scale, low new energy penetration rate): includes 30 grid nodes, connected to distributed photovoltaic (installed capacity 0.5MW), distributed wind power (installed capacity 0.3MW), and household energy storage system (total capacity 200kWh), mainly based on residential basic load, with flexible load accounting for 10%;

[0247] (2) Example 2 (medium scale, typical new energy penetration rate): includes 50 grid nodes, connected to distributed photovoltaic (installed capacity 1.2MW), distributed wind power (installed capacity 0.8MW), and household energy storage system (total capacity 500kWh), covering residential load and flexible agricultural production load (irrigation, electric heating), with flexible load accounting for 20%;

[0248] (3) Example 3 (large scale, high new energy penetration rate): includes 80 grid nodes, connected to distributed photovoltaic (installed capacity 2.0MW), distributed wind power (installed capacity 1.2MW), and household energy storage system (total capacity 800kWh), including rural industrial and commercial loads + agricultural large-scale flexible loads, with flexible loads accounting for 30%.

[0249] S612: Experimental Parameter Settings

[0250] (1) Number of runs: In order to eliminate the interference of random factors on the performance of the algorithm, the BTOA proposed in this invention and each baseline comparison algorithm are run independently 30 times;

[0251] (2) Evaluation period: covering the core time scale of the three-level scheduling framework of this invention, including day-ahead optimization scheduling (24h, step size 1h) and intraday rolling correction scheduling (4h, step size 15min).

[0252] (3) Algorithm parameters: All comparison algorithms adopted the optimal parameter configuration recommended in their original literature, and the BTOA parameters were kept consistent with those in step S331 (TeamSize=40, (200 optimizations were performed recently and 80 optimizations were performed daily). All algorithm convergence criteria and iteration parameters were unified to ensure experimental fairness.

[0253] (4) Iteration termination rule: consistent with step S333, terminated when any of the following conditions are met: ① The number of iterations reaches the maximum number of iterations MaxGen; ② The change in the global optimal fitness value over 20 consecutive iterations is ≤1e-6.

[0254] S613: Hardware Environment

[0255] All algorithms were deployed on a unified edge computing node (NVIDIA Jetson AGX Orin, 8-core ARM CPU, 32GB memory, 200 TOPS computing power), with the operating system being Ubuntu 20.04 LTS and the programming environment being Python 3.9 + NumPy 1.24, completely eliminating the impact of differences in hardware computing power and software environment on the experimental results.

[0256] S62: Evaluation Indicators

[0257] Five core indicators were selected to quantitatively compare the optimization performance and engineering practicality of each algorithm. All indicators were based on the comprehensive objective function F (the smaller the value, the better the scheduling scheme) in step S31 and the definition of algorithm operation characteristics. The calculation rules for all indicators were consistent for all algorithms.

[0258] Optimal objective value: The minimum comprehensive objective function value obtained by the algorithm in 30 independent runs, reflecting the optimization accuracy of the algorithm;

[0259] Average value: The average value of the comprehensive objective function after 30 independent runs, reflecting the overall optimization capability of the algorithm;

[0260] Standard deviation: The standard deviation of the combined objective function value after 30 independent runs. The smaller the value, the stronger the stability of the algorithm.

[0261] Average convergence algebra: The average number of iterations required for the algorithm to reach convergence in 30 independent runs; if the algorithm does not reach convergence within the maximum number of iterations, the convergence algebra of that run is recorded as the maximum number of iterations to ensure consistent statistical rules.

[0262] Average time: The average time (in seconds) for a single solution of the algorithm in 30 independent runs, reflecting the engineering feasibility of the algorithm on an edge computing platform.

[0263] S63: Experimental Results and Analysis

[0264] S631: Baseline Comparison Algorithm

[0265] Four mainstream metaheuristic algorithms in the field of distribution network optimization were selected as baseline comparison algorithms, all of which are excellent algorithms whose performance has been verified by literature:

[0266] Beluga Whale Optimization (BWO) [1];

[0267] Crested Porcupine Optimizer (CPO) [2];

[0268] Arithmetic Optimization Algorithm (AOA) [3];

[0269] PID-based Search Algorithm (PSA) [4].

[0270] [1] Zhong, C., Li, G. & Meng, Z. Beluga whale optimization: A novel nature-inspired metaheuristic algorithm. Knowledge-Based System S251, 109215 (2022).

[0271] [2] Abdel-Basset, M., Mohamed, R. & Abouhawwash, M. Crested porcupine optimizer: A new nature-inspired metaheuristic. Knowledge-Based System S284, 111257. https: / / doi.org / 10.1016 / j.knosys.2023.111257(2024).

[0272] [3]Abualigah,L.,Diabat,A.,Mirjalili,S.,Abd Elaziz,M.&Gandomi,AHThearithmetic optimization algorithm.Computer methodSin applied mechanicSandengineering 376,113609(2021).

[0273] [4]Gao,Y.Pid-based search algorithm: a novel metaheuristic algorithm based on pid algorithm.Expert SystemSWith ApplicationS232,120886(2023).

[0274] Step S632: Results of the recent optimization scheduling experiment

[0275] The scheduling was recently optimized (24h, step size 1h). In the scenario of (=200), the performance comparison results of each algorithm in the three sets of examples are shown in Tables 1, 2, and 3:

[0276] Table 1. Performance Comparison of Day-ahead Optimized Scheduling Algorithms in Example 1 (30 Nodes)

[0277] algorithm Optimal target value average value Standard deviation Average convergent algebra Average time elapsed (s) BTOA 68.5 70.2 1.1 62 7.8 BWO 73.2 75.6 2 85 9.5 CPO 71.8 74.1 1.7 78 8.9 AOA 75.9 78.3 2.5 98 10.2 PSA 70.9 72.8 1.5 70 8.3

[0278] Table 2. Performance Comparison of Day-ahead Optimized Scheduling Algorithms in Example 2 (50 Nodes)

[0279] algorithm Optimal target value average value Standard deviation Average convergent algebra Average time elapsed (s) BTOA 125.3 128.7 2.1 85 12.4 BWO 132.8 136.2 3.8 112 15.7 CPO 130.5 134.1 3.2 105 14.3 AOA 135.7 139.8 4.5 128 16.8 PSA 129.8 133.5 2.9 98 13.6

[0280] Table 3. Performance Comparison of Day-ahead Optimized Scheduling Algorithms in Example 3 (80 Nodes)

[0281] algorithm Optimal target value average value Standard deviation Average convergent algebra Average time elapsed (s) BTOA 216.7 220.5 3 118 18.6 BWO 228.9 233.4 4.2 155 22.3 CPO 225.6 230.1 3.8 142 20.7 AOA 235.2 240.8 5.1 168 24.5 PSA 223.1 227.8 3.5 130 19.9

[0282] S633: Intraday Rolling Correction Scheduling Experiment Results

[0283] Intraday rolling adjustment scheduling (4h, step size 15min) In the scenario of 80), the performance comparison results of each algorithm in the three sets of examples are shown in Tables 4, 5, and 6:

[0284] Table 4 Performance Comparison of Intraday Rolling Scheduling Algorithm in Example 1 (30 Nodes)

[0285] algorithm Optimal target value average value Standard deviation Average convergent algebra Average time elapsed (s) BTOA 24.3 25.1 0.4 22 2.5 BWO 26.8 27.9 0.8 32 3.6 CPO 26.1 27.2 0.7 29 3.2 AOA 28.5 29.7 1 38 4.1 PSA 25.5 26.4 0.6 25 2.8

[0286] Table 5 Performance Comparison of Intraday Rolling Scheduling Algorithm in Example 2 (50 Nodes)

[0287] algorithm Optimal target value average value Standard deviation Average convergent algebra Average time elapsed (s) BTOA 45.8 47.2 0.8 32 4.1 BWO 49.2 50.9 1.5 45 5.7 CPO 48.5 50.1 1.2 41 5.2 AOA 51.3 53 1.8 52 6.3 PSA 47.9 49.5 1 38 4.8

[0288] Table 6 Performance Comparison of Intraday Rolling Scheduling Algorithm in Example 3 (80 Nodes)

[0289] algorithm Optimal target value average value Standard deviation Average convergent algebra Average time elapsed (s) BTOA 78.9 80.6 1.2 45 6.8 BWO 85.3 87.5 2.1 62 8.9 CPO 83.7 85.9 1.8 56 8.2 AOA 88.9 91.2 2.5 68 9.7 PSA 82.1 84.3 1.5 50 7.5

[0290] S634: Comprehensive Analysis of Experimental Results

[0291] Based on experimental data from three sets of computational examples of different scales and different new energy penetration rates, a comprehensive analysis was conducted from three dimensions: optimization accuracy and stability, convergence speed, and engineering practicality. The conclusions are as follows:

[0292] 1. Higher optimization accuracy and stronger operational stability. In three sets of examples across two scheduling scenarios (daily and intraday), the BTOA proposed in this invention significantly outperforms all baseline comparison algorithms in terms of optimal target value and average value, with the smallest standard deviation among all algorithms (e.g., only 1.2 in the intraday scenario of example 3). This result verifies that BTOA's hierarchical adaptive search mechanism (high-intensity training + fast attack strategy + dynamic positioning strategy) effectively balances global exploration and local development capabilities, avoiding the algorithm from getting trapped in local optima. Even in complex scenarios with 80 nodes, high dimensionality, and high new energy penetration, it maintains excellent optimization performance with no significant fluctuations during operation, demonstrating outstanding stability.

[0293] 2. Faster convergence speed, adaptable to multi-scale scheduling requirements. BTOA's average convergence generation is significantly lower than other comparative algorithms in all examples and scheduling scenarios, and even in large-scale examples, it can still achieve convergence within 60% of the maximum number of iterations (e.g., only 118 generations in the 3-day scenario of example 3). =200). Compared with the optimal baseline algorithm PSA, BTOA's convergence speed is improved by an average of 12.5%~16.3% in the three sets of examples. Its fast convergence characteristics make it suitable for the scheduling optimization needs of rural distribution networks at multiple time scales, including day-ahead and intraday.

[0294] 3. Lower engineering time and excellent adaptability to edge computing power. The average time of BTOA is the lowest among all algorithms, and the time increase shows a linear and moderate increase with the scale of power grid nodes, without exponential growth. The maximum time for intraday rolling scheduling is only 6.8s (80 nodes), which is much less than the 15-minute rolling scheduling cycle, and the maximum time for day-ahead scheduling is only 18.6s, which is much less than the 24-hour day-ahead scheduling cycle, fully meeting the timeliness requirements of engineering. It should be noted that: in this invention, the response time of the real-time control layer ≤200ms is achieved by independent traditional PI closed-loop control, which does not depend on the solution time of the BTOA algorithm. BTOA is only responsible for generating the day-ahead and intraday optimized scheduling plans. Its second-level time does not conflict with the millisecond-level response of real-time control. The two have a clear division of labor and cooperate with each other.

[0295] 4. The algorithm is highly versatile and adaptable to different rural power distribution network scenarios. Whether it is a small-scale, low-penetration case with 30 nodes, a typical-scale case with 50 nodes, or a large-scale, high-penetration case with 80 nodes, BTOA maintains a consistent performance advantage across the five evaluation metrics. Its performance does not degrade with the expansion of the power grid or the increase in renewable energy penetration, verifying the algorithm's strong adaptability and versatility to rural power distribution networks with different characteristics and resolving the issue of randomness in single-case experimental results.

[0296] S64: Experimental Conclusions

[0297] This comparative experiment used three sets of 10kV distribution network examples from counties and villages of different scales and with different renewable energy penetration rates. Under a unified hardware environment, parameter configuration, and evaluation rules, BTOA was quantitatively compared with four mainstream metaheuristic algorithms across all dimensions. The experimental results fully validated:

[0298] (1) In the scenario of multi-objective scheduling optimization of distribution network, the BTOA proposed in this invention is significantly better than the comparative algorithms such as BWO, CPO, AOA, and PSA in terms of optimization accuracy, operation stability, convergence speed and engineering time.

[0299] (2) BTOA has strong versatility and scenario adaptability. It can maintain excellent performance in rural power distribution networks of different scales and different new energy penetration rates without significant performance degradation.

[0300] (3) The solution time of BTOA fully meets the timeliness requirements of the day-ahead and intraday optimization scheduling of rural power distribution networks. It works in collaboration with the PI algorithm of the real-time control layer and can be stably deployed at edge computing nodes, with strong feasibility for engineering implementation.

[0301] II. Supporting Dispatch and Control System Architecture

[0302] The corresponding collaborative dispatch and control system for low-carbon intelligent digital distribution networks in counties and rural areas, based on this method, includes six core modules and is entirely based on the traditional industrial control architecture:

[0303] (1) Data acquisition module: It consists of edge acquisition terminal, smart meter and sensor network to realize real-time acquisition, preprocessing and edge transmission of multi-source data;

[0304] (2) Digital twin modeling module: Built-in full-element twin sub-model, supporting online identification, dynamic correction and virtual-real interaction of model parameters;

[0305] (3) Status simulation and risk warning module: realizes multi-scenario power flow simulation, security risk classification and early warning and distribution network vulnerability assessment;

[0306] (4) BTOA multi-objective optimization scheduling module: Deploy the BTOA optimization algorithm kernel proposed in this invention to generate three-level scheduling plans: day-ahead, intraday, and real-time;

[0307] (5) Layered collaborative control module: realizes three-level collaboration of local equipment control, regional collaborative control and power grid interactive control, and adopts traditional power control algorithm;

[0308] (6) Communication and security module: It adopts 5G+industrial Ethernet dual-link redundancy, with a main and backup link switching time of ≤100ms. It uses national cryptographic algorithms to achieve encrypted data transmission, which is suitable for the low bandwidth and high security requirements of rural power distribution networks.

[0309] Example 2

[0310] Figure 8 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, and may also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0311] like Figure 8 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0312] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0313] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 executes the intelligent digital twin cooperative scheduling method described above.

[0314] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0315] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. An intelligent digital twin collaborative scheduling method, characterized in that, Includes the following steps: S1: Collect multi-source data of the distribution network, and construct a full-element digital twin mirror model of the distribution network based on the multi-source data; the full-element digital twin mirror model serves as a virtual model of the distribution network in the digital space, and is used to realize bidirectional mapping and state inference between the physical power grid and the virtual model; S2: Based on the digital twin mirror model, perform multi-scenario simulation and deduction of the operating status of the distribution network, output key operating indicators to characterize the operating status of the distribution network, and generate safety constraint boundaries for step S3 based on the simulation results. S3: Construct a multi-time-scale collaborative scheduling framework, and within the security constraint boundary, use optimization algorithms to perform multi-objective collaborative optimization of the scheduling plan of the distribution network to generate an optimized scheduling scheme; S4: Based on the optimized scheduling scheme, perform hierarchical collaborative control and send scheduling instructions to the physical power grid for execution; S5: Collect real-time operational data and perform closed-loop feedback and iterative optimization on the multi-objective collaborative optimization steps of the digital twin mirror model and scheduling plan; In step S1, the multi-source data includes: power grid basic ledger data, real-time operation data, distributed new energy data, load and energy storage data, and environmental and external data; an edge computing and cloud-based collaborative acquisition architecture is adopted, and real-time detection and cleaning of abnormal data are completed at the edge. The full-element digital twin mirror model constructed in step S1 includes a geometric and topological twin model, a device-level mechanism twin model, a source-load time-series prediction model, and a virtual-real interaction model. The geometric and topological twin model is constructed based on the fusion modeling technology of building information modeling and geographic information system and graph theory search algorithm; The device-level mechanism twin model is constructed based on classical equivalent circuit theory and parameter identification algorithm; The source-load time-series prediction model is constructed using a backpropagation neural network. The virtual-real interaction model realizes data interaction between the physical power grid and the digital twin mirror model through the OPC unified architecture and message queue telemetry transmission dual protocol communication interface; Step S2 specifically includes: Multiple typical operating conditions are set up, and the Newton-Raphson method and time-domain simulation method are used to conduct multi-scenario full-time-domain simulation and deduction of the distribution network. The voltage of each node, line load rate, distribution transformer overload status and new energy absorption capacity are output as key operating indicators. Based on the simulation results, a threshold over-limit discrimination method is used to classify and warn of line overload, voltage over-limit, and equipment failure risks into four levels: red, orange, yellow, and blue. An entropy weight method-analytic hierarchy process coupled algorithm is adopted to construct an evaluation index system from multiple dimensions such as equipment health, grid redundancy, load importance and new energy penetration rate. The vulnerability of the distribution network is evaluated, the identification results of weak nodes and weak lines are output, and priority guidance is provided for scheduling optimization in step S3. Step S3 includes: constructing a comprehensive objective function as the optimization objective of the multi-objective collaborative optimization, which is used to generate an optimized scheduling scheme under the premise of satisfying the hard constraints of power grid security; The comprehensive objective function includes at least the objectives of minimizing line loss, minimizing carbon emissions, minimizing voltage deviation, and maximizing the absorption of new energy sources, and is weighted and summed by assigning dynamic weight coefficients to each sub-objective; the dynamic weight coefficients are adjusted based on the graded early warning results and vulnerability assessment results output in step S2; Comprehensive objective function As follows: ； Wherein, , , , Dynamic weight coefficient, satisfy , based on digital twin deduction of running state and risk level dynamic adjustment, adjustment rule: If the renewable energy consumption rate is lower than ,but The remaining weights are equally distributed, with priority given to improving the level of new energy consumption; if carbon emission intensity exceeds the standard, then... The remaining weights are equally distributed, prioritizing the achievement of low-carbon operation targets; if a voltage over-limit or line overload warning occurs, then... The remaining weights are evenly distributed, prioritizing the optimization of voltage quality and reduction of equipment overload risk; under normal operating conditions, , , , Prioritize low-carbon operation; The specific definitions of each sub-objective function are as follows: Minimize line loss : ； In the formula, The total time step of the scheduling cycle, in hours. Total number of lines, unit: lines for Time of the first The current of the line, in amperes (A). For the first Resistance of the line, unit: , The scheduling time step, in hours (h). Carbon emission minimization target : ； In the formula, for Net power purchased by the distribution network from the upstream grid at any given time, in kW. Negative values ​​are recorded as zero for grid connection. for Marginal carbon emission factor of the upstream power grid at any given time, unit: Data is taken from provincial carbon trading platforms. Voltage deviation minimization target : ； In the formula, Total number of distribution network nodes, unit: number. for Time of the first Voltage at each node, unit: kV. Rated voltage of the power grid, unit: ; Maximizing the utilization of new energy sources : ； In the formula, for Local consumption capacity of new energy sources at any time, unit: kW for Total output power of new energy sources at any time, unit: kW; The hard constraints on grid security include at least power balance constraints, node voltage constraints, equipment load constraints, energy storage system constraints, flexible load constraints, and grid interaction constraints; all optimization solutions must satisfy the hard constraints on grid security. In step S3, the multi-time-scale collaborative scheduling framework includes: The day-ahead optimized scheduling has a time scale of 24 hours and a step size of 1 hour. Based on the day-ahead prediction results output by the source-load time series prediction model, the basketball team optimization algorithm is used for global optimization to generate the day-ahead baseline scheduling plan. The intraday rolling correction scheduling has a time scale of 4 hours and a step size of 15 minutes. Based on the intraday rolling prediction results output by the source-load time series prediction model, the basketball team optimization algorithm is used to roll-correct the day-ahead baseline scheduling plan to generate the intraday corrected scheduling plan. Real-time optimization control, with a time scale of 5 minutes and a step size of 1 minute, is based on real-time mapping of operating data from a digital twin model. It employs a proportional-integral closed-loop control method and uses the intraday revised scheduling plan as a benchmark to quickly respond to random fluctuations in source load.

2. The intelligent digital twin collaborative scheduling method according to claim 1, characterized in that, The optimization algorithm in step S3 adopts the basketball team optimization algorithm. It constructs a search mechanism by simulating four core behaviors of a basketball team: high-intensity training, fast-break strategy, dynamic positioning strategy, and boundary control. The solution process includes: Initialization steps: Define the scheduling scheme that satisfies the constraints as the players in the population, and set the population size, iteration parameters, and algorithm core parameters; Iterative search steps: Players are divided into elite players and ordinary players based on fitness; a local search strategy simulating high-intensity training is executed for elite players; for ordinary players, a fast convergence search simulating fast break strategy or a global exploration search simulating dynamic positioning strategy is dynamically selected based on adaptive factors; the dynamic positioning strategy further includes a random position strategy and a diagonal position strategy. Boundary control steps: Implement a boundary control strategy that simulates a basketball going out of bounds and being replayed for all updated players to correct solutions that exceed the constraint boundaries; Termination judgment step: Iteratively execute the iterative search step and boundary control step until the termination condition is met, and output the globally optimal player position as the optimized scheduling scheme.

3. The intelligent digital twin collaborative scheduling method according to claim 1, characterized in that, Step S4 includes constructing a hierarchical collaborative control system, specifically including: The local control layer of the equipment consists of feeder terminal units, distribution transformer terminal units, energy storage converter controllers, new energy inverters and reactive power compensation controllers. It is used to receive upper-level instructions and execute fast closed-loop control at the equipment level, with a response time of no more than 100 milliseconds. The regional collaborative control layer, with the distribution network digital twin platform as its core, is deployed on edge computing nodes and cloud servers to coordinate the collaborative operation of the entire process of source-grid-load-storage within the region. The power grid interaction layer employs a coordinated control strategy of automatic generation control and automatic voltage control to interact with the upper-level power grid and receive and execute emergency control commands issued by the upper-level power grid.

4. The intelligent digital twin collaborative scheduling method according to claim 1, characterized in that, Step S5 specifically includes: Establish a scheduling performance evaluation index system that includes renewable energy absorption rate, comprehensive line loss rate, carbon emission intensity, node voltage qualification rate, equipment fault early warning accuracy rate, and digital twin model fitting degree; among which, the scheduling performance evaluation index is determined based on the key operation indicators output in step S2. The actual values ​​of each evaluation indicator in the scheduling effect evaluation index system are statistically analyzed at a preset period, and the actual values ​​are compared with the preset target values. When the deviation between the actual value and the target value of any evaluation index exceeds the preset range, at least one of the following optimization measures shall be executed according to the direction and magnitude of the deviation: re-identify and correct the device parameters of the digital twin mirror model, adjust the dynamic weight coefficients in the comprehensive objective function, perform incremental training on the source load time series prediction model, and verify and optimize the core parameters of the basketball team optimization algorithm.

Images