A rural intelligent power distribution network energy management system supporting multi-energy coordination

Abstract

This invention discloses an energy management system for rural smart distribution networks supporting multi-energy collaboration, belonging to the field of smart distribution network and distributed energy management technology. The system adopts a three-layer collaborative architecture of cloud, edge, and terminal. The edge side deploys a lightweight adaptive optimization algorithm with dual-center collaboration and an adaptive search radius adjustment mechanism, possessing edge-side autonomy. The cloud side only executes non-real-time tasks, while the terminal side completes data acquisition and command execution. The algorithm achieves scheduling through data acquisition initialization, dual-point dynamic collaborative optimization, edge-terminal real-time control, equipment health management, and local autonomous deployment. It also designs mechanisms for local voltage rapid response and cost-related assessment of energy storage health. This invention achieves controllable computing resource requirements, simplified parameter configuration, adaptive search adjustment, adaptability to rural edge hardware, improved scheduling economy, effective control of energy storage losses, and ensures the voltage qualification rate of the distribution network, possessing practical application potential.

Description

Technical Field

[0001] This invention belongs to the field of smart distribution network and distributed energy management technology, specifically relating to a rural smart distribution network energy management system that supports multi-energy synergy. Background Technology

[0002] With the large-scale integration of distributed photovoltaic (PV) systems in rural areas, the operation of power distribution networks faces numerous challenges. While current multi-layered collaborative technologies for urban renewable energy have made initial explorations in areas such as multi-energy complementarity optimization and improved energy utilization, their adaptability to the actual operating scenarios of vast rural power distribution networks still has significant technical shortcomings.

[0003] For example, the invention patent with publication number CN117574209A, "A Multi-layer Collaborative Real-time Classification System and Method for Urban-scale Renewable Energy", constructs a multi-layer collaborative architecture that includes a data acquisition unit, a data classification unit, a data analysis unit, and a scheduling unit. By collecting various renewable energy power generation data and urban electricity load data, it realizes multi-energy complementarity and real-time monitoring and optimized control of renewable energy, in order to improve the stability of biogas energy production, the utilization rate of renewable energy, and the sustainability of power generation.

[0004] However, this technical solution has significant limitations and falls short of the actual operational needs of rural power distribution networks: First, the classification logic lacks scenario adaptability, relying solely on static classification based on energy sources and conversion methods, without considering dynamic conditions such as transient load fluctuations and the randomness of renewable energy output in rural power distribution networks; Second, the scheduling mechanism design is imperfect, failing to introduce lightweight scheduling optimization algorithms, placing high demands on the computing power of rural edge scheduling equipment, and making it difficult to adapt to deployment scenarios of low- to mid-range edge gateways; Third, real-time response performance is insufficient, with a coordination delay between data classification and scheduling command issuance, making it unable to quickly respond to sudden fluctuations in rural electricity load and failing to guarantee the voltage stability and power supply reliability of the power distribution network.

[0005] Therefore, it is necessary to design an energy collaborative management scheme with a simple structure, controllable computational load, and strong adaptability to better adapt to the physical and resource characteristics of rural power distribution networks. Summary of the Invention

[0006] This invention aims to provide an energy management system for rural smart power distribution networks that supports multi-energy synergy. Specific objectives include: reducing computational resource requirements: designing a computationally controllable optimization method that can run on edge computing devices with limited computing power, thus lowering the hardware deployment threshold. Simplifying parameter configuration: reducing algorithm parameters requiring manual debugging, improving the system's deployability in areas with limited technical resources. Establishing an adaptive search mechanism: through a dynamically adjusted search strategy, enabling the algorithm to automatically adjust the focus of exploration and development based on the iteration process, without external intervention.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A smart energy management system for rural power distribution networks that supports multi-energy collaboration adopts a three-layer collaborative architecture of cloud, edge, and terminal, including a terminal-side intelligent data acquisition and execution unit, an edge-side edge computing control unit, and a cloud-side non-real-time management unit.

[0009] The edge computing control unit is equipped with a lightweight adaptive optimization algorithm based on dual-point dynamic collaboration. The algorithm adopts a dual-center collaboration and search radius adaptive adjustment mechanism. The edge computing control unit has edge autonomy capabilities and can complete multi-energy collaborative scheduling of the power distribution network based on local cached data when communication is interrupted.

[0010] The cloud-side non-real-time management unit sends the day-ahead electricity price and daily energy storage health estimation model to the edge computing control unit according to a preset non-real-time cycle. It receives and stores historical data uploaded by the edge computing control unit and analyzes it. It only triggers data interaction and model updates during the non-real-time cycle and does not participate in millisecond / second-level real-time scheduling. It only performs non-real-time management tasks.

[0011] The end-side intelligent acquisition and execution unit completes the acquisition of multi-energy and operation data of the power distribution network and executes the control commands issued by the edge computing control unit.

[0012] Preferably, the end-side intelligent data acquisition and execution unit includes a photovoltaic inverter, an energy storage converter, and a smart meter;

[0013] The non-real-time management tasks performed by the cloud-side non-real-time management unit include the issuance of day-ahead electricity prices, updating of the daily energy storage health estimation model, and storage and analysis of historical operation data.

[0014] Preferably, the execution flow of the lightweight adaptive optimization algorithm based on two-point dynamic collaboration includes data acquisition and state initialization, two-point dynamic collaborative optimization, real-time control and command issuance of edge-end collaboration, equipment health management and long-term economic optimization, and local autonomous deployment of the system.

[0015] Preferably, the data acquisition and state initialization includes normalizing the acquired multi-energy, operational data and cloud-based data, initializing two core state points for the scheduling strategy, namely development point A and exploration point B, and binding independent search radii to the two core state points and completing initial value assignment. The dimensions of development point A and exploration point B are consistent with the number of distribution network scheduling decision variables.

[0016] Preferably, the two-point dynamic collaborative optimization first evaluates the scheduling strategies corresponding to development point A and exploration point B by constructing a fitness function. The fitness function formula is:

[0017] ;

[0018] In the formula, The fitness function value; For electricity purchase costs; The cost of punitive measures for curtailing renewable energy; Cost of energy storage equipment per charge and discharge cycle.

[0019] Preferably, after completing the fitness evaluation, the dual-point dynamic collaborative optimization performs a point exchange operation to always ensure that development point A is the optimal scheduling strategy point obtained by the current optimization and exploration point B is the exploratory scheduling strategy point. Then, candidate solutions are generated based on development point A, exploration point B and their bound search radii. The generation of candidate solutions adopts one of two modes: single-dimensional perturbation and multi-dimensional perturbation.

[0020] Preferably, after the candidate solutions are generated, after inverse normalization, decision variable boundary truncation and dynamic constraint verification of energy storage charge state, the candidate solutions with the best fitness are selected from the candidate solutions corresponding to development point A and exploration point B respectively, and development point A and exploration point B are updated.

[0021] Preferably, in the dual-point dynamic collaborative optimization, the search radius is dynamically adjusted according to a preset iteration interval. This includes updating the core state points of the archives based on local archives with a fixed capacity, and updating the search radius through a preset coefficient, so that the search radius of development point A gradually shrinks to a preset lower limit and the search radius of exploration point B gradually expands to a preset upper limit.

[0022] Preferably, the real-time control and command issuance of the edge-end collaboration includes scheduling strategy decomposition, local rapid response and communication deployment. The local rapid response means that when the grid connection point voltage exceeds a preset threshold, the photovoltaic inverter is called to provide reactive power support for voltage. When the reactive power regulation capability is insufficient, the energy storage is called to perform active / reactive power coordinated regulation.

[0023] Preferably, the system is encapsulated using containerization technology, adapted to edge gateway hardware with advanced reduced instruction set machine architecture, and control commands are issued via power line carrier or wireless communication with a fixed data interaction cycle.

[0024] The present invention provides a rural smart power distribution network energy management system that supports multi-energy synergy, achieving several technical advantages:

[0025] 1. Controllable computational resource requirements: The method proposed in this invention only needs to evaluate 4D candidate policy points per generation, generating 2D points for point A and 2D points for point B. The number of evaluations is comparable to that of the standard particle swarm optimization algorithm, i.e., the population size is 4D. However, since there is no need to maintain population sorting and global optimal tracking, the overall algorithm implementation is simpler and suitable for running on edge devices with limited computing power.

[0026] 2. Simplified parameter configuration: The core parameter of the algorithm is only the iteration interval. and radius variation coefficient And the value is within a certain range, specifically: , It has no significant impact on the final optimization results, reducing the difficulty of on-site debugging.

[0027] 3. Adaptive search capability: By developing a mechanism that allows the radius of the exploration point to shrink successively and the radius of the exploration point to expand successively, the algorithm can focus on global search in the early stage of iteration and local fine-tuning in the later stage without the need for external parameter control.

[0028] 4. Excellent system integration: Utilizing containerized encapsulation and standard communication protocols, it is compatible with mainstream edge gateway hardware. The local fast response mechanism can complete the initial adjustment of voltage fluctuations within 100ms, meeting the basic requirements of rural power distribution networks for fast response. Attached Figure Description

[0029] Figure 1 This is a block diagram of the cloud, edge, and terminal three-layer architecture of the present invention.

[0030] Figure 2 This is a flowchart illustrating the two-point dynamic collaborative optimization process of the lightweight adaptive optimization algorithm based on two-point dynamic collaboration according to the present invention. Detailed Implementation

[0031] The following detailed implementation description of a rural smart power distribution network energy management system supporting multi-energy coordination, with reference to specific embodiments, is provided below. These embodiments are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0032] Example 1: Implementation of a rural smart power distribution network energy management system that supports multi-energy synergy.

[0033] Combined with appendix Figures 1-2 As shown, the present invention provides an energy management system for rural smart power distribution networks that supports multi-energy synergy.

[0034] like Figure 1 As shown, Figure 1 This is a block diagram of the cloud, edge, and terminal three-layer architecture of the present invention.

[0035] This invention provides an energy management system for rural smart distribution networks that supports multi-energy synergy. Its core is an optimization method based on two-point dynamic collaboration. This method draws on the idea of ​​generating candidate solutions by perturbing the current point in pattern search, and introduces a dual-center collaboration and adaptive search radius adjustment mechanism on this basis.

[0036] This method, centered on the classic pattern search method, fully utilizes its basic logic of generating candidate solutions by applying targeted perturbations to the current decision point and iteratively selecting the best solution from these candidates. It also specifically overcomes the inherent shortcomings of traditional pattern search methods, such as a single optimization center, fixed and rigid search strategies, and insufficient adaptability. Considering the practical characteristics of rural smart distribution networks—limited edge computing power, high real-time scheduling requirements, and complex multi-energy collaborative scenarios involving photovoltaics, energy storage, and loads—this method innovatively introduces two core mechanisms—dual-center collaboration and adaptive adjustment of the search radius—on top of the basic perturbation optimization framework, thus reconstructing a lightweight, adaptive, and highly adaptable new optimization model.

[0037] The dual-center collaborative mechanism constructs two core optimization carriers: a development point and an exploration point. The development point focuses on local fine-tuning, continuously exploring the optimization potential of the current optimal strategy, while the exploration point focuses on global exploration, proactively expanding the optimization boundary to avoid the algorithm getting trapped in local optima. The two points are dynamically evaluated and switch roles as needed, achieving a dynamic balance between local development and global exploration. The adaptive search radius adjustment mechanism abandons the traditional fixed search radius mode, automatically adjusting the search range of the two points according to the iteration process. The radius of the development point gradually shrinks to achieve precise optimization, while the radius of the exploration point gradually expands to strengthen the global search. It can autonomously switch optimization priorities without manual intervention or complex parameter tuning. The overall improvement retains the advantages of the pattern search method, such as simple calculation and low computing power requirements, while significantly improving optimization efficiency and robustness through the dual-center and adaptive mechanisms, perfectly adapting to the computing power conditions of rural power distribution network edge hardware and the actual needs of multi-energy collaborative scheduling.

[0038] The system adopts a three-layer collaborative architecture of cloud, edge, and terminal, including a terminal-side intelligent data acquisition and execution unit, an edge-side edge computing control unit, and a cloud-side non-real-time management unit.

[0039] The edge computing control unit is equipped with a lightweight adaptive optimization algorithm based on dual-point dynamic collaboration. The algorithm adopts a dual-center collaboration and search radius adaptive adjustment mechanism. The edge computing control unit has edge autonomy capabilities and can complete multi-energy collaborative scheduling of the power distribution network based on local cached data when communication is interrupted.

[0040] The cloud-side non-real-time management unit sends the day-ahead electricity price and daily energy storage health estimation model to the edge computing control unit according to a preset non-real-time cycle. It receives and stores historical data uploaded by the edge computing control unit and analyzes it. It only triggers data interaction and model updates during the non-real-time cycle and does not participate in millisecond / second-level real-time scheduling. It only performs non-real-time management tasks.

[0041] The end-side intelligent acquisition and execution unit completes the acquisition of multi-energy and operation data of the power distribution network and executes the control commands issued by the edge computing control unit.

[0042] In this system, the end-side intelligent acquisition and execution unit includes a photovoltaic inverter, an energy storage converter, and a smart meter; the cloud-side non-real-time management unit performs non-real-time management tasks including day-ahead electricity price issuance, daily energy storage health estimation model updates, and historical operation data storage and analysis.

[0043] In this system, the execution flow of the lightweight adaptive optimization algorithm based on two-point dynamic collaboration includes data acquisition and state initialization, two-point dynamic collaborative optimization, real-time control and command issuance of edge-end collaboration, equipment health management and long-term economic optimization, and local autonomous deployment of the system.

[0044] In this system, data acquisition and state initialization include normalizing the acquired multi-energy, operational, and cloud-based data, and initializing two core state points for the scheduling strategy: development point A and exploration point B. Independent search radii are bound to the two core state points and initial values ​​are assigned. The dimensions of development point A and exploration point B are consistent with the number of variables in the distribution network scheduling decision.

[0045] In this system, after completing the fitness evaluation, the two-point dynamic collaborative optimization performs a point exchange operation to ensure that the development point A is the optimal scheduling strategy point obtained by the current optimization and the exploration point B is the exploratory scheduling strategy point. Then, candidate solutions are generated based on the development point A, the exploration point B and their bound search radii. The generation of candidate solutions adopts one of two modes: single-dimensional perturbation and multi-dimensional perturbation.

[0046] In this system, after candidate solutions are generated, they are inversely normalized, the decision variable boundary is truncated, and the energy storage charge state is dynamically constrained and verified. Then, the candidate solutions with the best fitness are selected from the candidate solutions corresponding to development point A and exploration point B, respectively, and development point A and exploration point B are updated.

[0047] In this system, during the two-point dynamic collaborative optimization, the search radius is dynamically adjusted according to a preset iteration interval. This includes updating the core state points of the archives based on local archives with a fixed capacity, and updating the search radius through preset coefficients, so that the search radius of development point A gradually shrinks to the preset lower limit and the search radius of exploration point B gradually expands to the preset upper limit.

[0048] In this system, real-time control and command issuance for edge-end collaboration includes scheduling strategy decomposition, local rapid response, and communication deployment. The local rapid response involves calling the reactive power of the photovoltaic inverter to support the voltage when the grid connection point voltage exceeds a preset threshold, and then calling the energy storage to perform active / reactive power coordinated regulation when the reactive power regulation capability is insufficient.

[0049] In this system, containerization technology is used for encapsulation, adapting to edge gateway hardware with an advanced reduced instruction set machine (RISC) architecture. Control commands are issued via power line carrier or wireless communication and use a fixed data exchange cycle. Details are as follows:

[0050] System Architecture:

[0051] The system adopts a cloud-edge-device collaborative architecture:

[0052] The edge-side intelligent data acquisition and execution unit, i.e. the device layer, consists of intelligent terminals such as photovoltaic inverters, energy storage converters, and smart meters, which are responsible for data acquisition and command execution.

[0053] Edge computing control unit, i.e., the station area layer: deploys edge computing gateways, runs the lightweight adaptive optimization algorithm based on two-point dynamic collaboration proposed in this invention, and is responsible for the coordinated control of multiple devices within the station area.

[0054] The cloud-side non-real-time management unit, i.e. the main station layer, is responsible for non-real-time tasks, including the issuance of day-ahead electricity prices, the daily estimation model update of energy storage health (SOH), and the storage and analysis of historical data. It does not participate in millisecond / second-level real-time scheduling.

[0055] As the core of non-real-time control in the system's cloud-edge-device three-layer architecture, the cloud-side non-real-time management unit strictly adheres to a preset daily non-real-time cycle. It does not participate in the millisecond / second-level real-time scheduling calculations and control command issuance of the distribution network, entrusting key tasks such as real-time collaborative scheduling and rapid voltage response entirely to the edge computing control unit, ensuring low latency and high reliability of rural distribution network scheduling. This unit issues daily time-of-use pricing and daily energy storage health estimation models to the edge computing control unit, providing core parameters and model support for the edge-side lightweight optimization algorithm. Simultaneously, it actively receives all historical data on photovoltaics, energy storage, load, and grid operation uploaded by the edge computing control unit, completing centralized storage, classification, and in-depth analysis of the data, providing data support for long-term system economic optimization, energy storage health management, and scheduling strategy iteration. The cloud-side non-real-time management unit only triggers data interaction and model updates during the agreed non-real-time period, without consuming real-time communication bandwidth and computing resources. Through clear decoupling of cloud and edge responsibilities, it achieves efficient collaboration between non-real-time management and real-time control, adapting to application scenarios where rural power distribution networks have limited edge computing power and diverse communication conditions.

[0056] The execution flow of the lightweight adaptive optimization algorithm based on two-point dynamic collaboration is as follows:

[0057] 1. Data acquisition and status initialization.

[0058] The following data was collected from the edge gateway of the rural power distribution area:

[0059] New Energy Data: Real-time Active Power of Photovoltaic Inverters reactive power .

[0060] Energy storage data: State of charge (SOC), charge / discharge power Rated capacity .

[0061] Load data: Total load of the transformer area .

[0062] Grid data: Voltage at grid connection point Grid connection point power .

[0063] Cloud-based data distribution: next day's time-of-use electricity price Initial value of energy storage health status (SOH).

[0064] Normalize the data. For any physical quantity and its upper and lower limits , Normalized physical quantities The calculation formula is:

[0065] ;

[0066] Normalized candidate solutions generated during the optimization process are denormalized back to physical values ​​before calculating fitness and truncated at the boundary.

[0067] Two core state points are initialized: point A (development point) and point B (exploration point). These two points represent two different scheduling strategy vectors, and the dimension of each vector corresponds to the number of decision variables. Table 1 defines the decision variables and their physical constraints.

[0068] Table 1

[0069]

[0070] in, This represents the maximum charging and discharging power of the energy storage. This represents the maximum reactive power of the photovoltaic inverter. This represents the maximum adjustment amount for flexible loads.

[0071] The state of charge (SOC) of the energy storage is not treated as an independent decision variable, but rather as a state variable obtained by integrating the charging and discharging power. Its dynamic constraints are verified after each candidate solution is generated.

[0072] Point A is initialized to its current running state, i.e., the current values ​​of all decision variables, while point B is randomly initialized within the feasible region. Simultaneously, a search radius is initialized for each of the two points. and The initial values ​​are all set to 0.5.

[0073] 2. Two-point dynamic collaborative optimization.

[0074] like Figure 2 As shown, Figure 2 This is a flowchart illustrating the two-point dynamic collaborative optimization process of the lightweight adaptive optimization algorithm based on two-point dynamic collaboration according to the present invention.

[0075] The dual-point dynamic collaborative optimization process is the core of the algorithm, enabling adaptive optimization of distribution network scheduling strategies. In this process, dual-center collaboration uses development point A and exploration point B as two coordination centers for scheduling strategies. A is always responsible for local development as the currently optimized optimal strategy point, while B is responsible for global exploration. After fitness evaluation, the two points are exchanged as needed, each generating candidate solutions and updating them selectively, thus achieving collaborative optimization between development and exploration. The adaptive search radius adjustment mechanism involves updating the two points in the local archive at preset iteration intervals. Then, using a preset formula, the search radius of development point A is gradually shrunk to the lower limit, while the search radius of exploration point B is gradually expanded to the upper limit. This allows the algorithm to adaptively switch between global exploration and local fine-tuning as the iteration progresses. The specific process details are as follows:

[0076] This process achieves adaptive optimization of the scheduling strategy through co-evolution at two points. In each iteration, the following sub-steps are executed:

[0077] ;

[0078] in, The fitness function value represents the overall cost of the scheduling strategy and is used to evaluate the merits of point A and point B. The smaller the fitness value, the better the scheduling strategy. This represents the current electricity purchase cost; The electricity price is real-time and is distributed daily from the cloud. Power at the grid connection point; Penalties for curtailing renewable energy generation The penalty factor is set at 1.2 times the feed-in tariff for photovoltaic power. This refers to the power curtailment of renewable energy sources. This refers to the cost of energy storage equipment per charge-discharge cycle. The depth of SOC change during a single charge-discharge cycle is calculated using the following formula: , To control the cycle, For energy storage charging and discharging power, For the rated capacity of energy storage, The conversion factor is based on the cycle life curve provided by the energy storage manufacturer.

[0079] If point B has a better fitness than point A, then swap points A and B. This operation ensures that point A always represents the best strategy found so far and is responsible for development; while point B is responsible for exploration. Store the current points A and B in the local archive. Radius and It does not switch with points and is always bound to points A and B respectively.

[0080] Candidate solution generation: Based on points A and B and their search radii, new candidate strategy sets are generated. The generation method includes two modes, randomly selected with equal probability:

[0081] Mode 1, Single-Dimensional Perturbation: For a point P, P can be A or B, generate 2D new points, where D is the number of decision variables. Each new point is perturbed in only one dimension, and the perturbation amount is determined by the radius. It is determined by a random number. The formula is:

[0082] ;

[0083] in, Let j represent the j-th dimension of the decision variable at point P. The random numbers are uniformly distributed in [0,1] and are generated by a pseudo-random number generator (the random seed is fixed at 0 during simulation).

[0084] Mode 2, Multidimensional Perturbation: For point P, generate a matrix M containing 2D unique random binary strings, each string having length D. Uniqueness is guaranteed by randomly generating 2D integers within a certain range. If duplicates occur, regenerate until all are unique, then convert to a binary string. Generate 2D new points, and for each new point, perturb each dimension according to the value of the corresponding binary bit:

[0085] ;

[0086] in, ; ; The above are uniformly distributed random numbers in the [0,1] range; Label variables for device operating status, superscript Representing the Each scheduling period, index Representing the One distributed energy / load device.

[0087] After generating candidate solutions, they are inversely normalized to physical values ​​and truncated according to the decision variable boundaries. Simultaneously, the dynamic constraints of the energy storage state of charge (SOC) are verified: based on the candidate... Calculate the predicted SOC value for the next time step. If it exceeds... The range is then adjusted. Until the boundary value that satisfies the constraint.

[0088] From the 2D new points generated for point A and the 2D new points generated for point B, select the points with the best fitness respectively, and update points A and B respectively.

[0089] Dynamic radius adjustment: Every I iterations (I is the preset iteration interval, I=8), a radius adjustment step is executed once.

[0090] File selection: Identify the strategy point with the best fitness from the local files. If it is better than the current point A, replace point A; if it is also better than point B, replace point B. The file size is fixed. That is, 16 records, and old points are eliminated using the first-in-first-out rule.

[0091] Radius Update: The search radius is dynamically adjusted according to the following formula:

[0092] ;

[0093] in, For development point The search radius; To determine the search radius for point B; As a preset coefficient, take ; To prevent the algorithm from stalling when the radius of the development point reaches zero; This prevents excessive disturbance at the exploration point from causing candidate solutions to exceed the device's physical constraints. The formula gradually shrinks the search radius of development point A to its lower limit, guiding it towards a refined local search; while gradually expanding the search radius of exploration point B to its upper limit, enhancing global exploration capabilities.

[0094] Iteration Termination: When the preset maximum number of iterations is reached. When the iteration terminates, output the current point. As the optimal scheduling strategy.

[0095] The dual-point dynamic collaborative optimization mechanism is the core optimization mechanism of this system, precisely addressing the challenges of high computational demands, cumbersome parameter adjustments, and imbalance between exploration and development inherent in traditional optimization algorithms. This mechanism employs a dual-core collaborative architecture of development point A and exploration point B, replacing the large-scale population model of particle swarm optimization and genetic algorithms. Each generation only needs to evaluate 4D candidate solutions, eliminating the need to maintain population sorting and global optimum tracking, significantly reducing computational load and memory consumption. Computational resource requirements are controllable, enabling stable operation even in rural edge gateways with limited computing power. The two points have clearly defined roles and can be dynamically exchanged. Development point A focuses on local fine-tuning and prioritizes optimal scheduling strategies, while exploration point B focuses on global exploration and avoiding local optimum traps, achieving a dynamic balance between exploration and development and significantly improving optimization accuracy and efficiency.

[0096] The adaptive search radius adjustment mechanism requires no manual intervention. It automatically shrinks the search radius of development points and expands the search radius of exploration points according to the iteration process. The algorithm focuses on global search in the early stage and local fine-tuning in the later stage, adaptively switching the optimization strategy. At the same time, the algorithm has very few core parameters and the values ​​are highly robust, eliminating the need for on-site debugging by professional personnel and greatly reducing the difficulty of deployment.

[0097] The optimization process relies on a fitness function that integrates electricity purchase costs, curtailment penalties, and energy storage loss costs, balancing dispatch economics with energy storage health protection, and effectively controlling energy storage charging and discharging losses. After candidate solutions are generated, physical constraints such as energy storage SOC are simultaneously verified to ensure that the dispatch strategy fits the equipment operating boundaries, significantly improving the feasibility and practicality of the strategy.

[0098] 3. Real-time control and command issuance for edge-device collaboration.

[0099] Scheduling strategy decomposition: The edge gateway converts the strategy vector output by the optimization algorithm, including energy storage charging and discharging power, flexible load adjustment amount, etc., into control commands for specific devices, and sends them to the terminal devices in accordance with Modbus RTU or IEC 60870-5-104 protocol.

[0100] Local fast response: For voltage fluctuations, set the grid connection point voltage threshold. , When the voltage exceeds the threshold, the edge gateway prioritizes using the reactive power of the photovoltaic inverter for support, with the reactive power regulation calculated according to the QV droop curve. If the reactive power regulation capacity is insufficient, energy storage is then used for active / reactive power regulation, without waiting for a new round of optimization. This is the lower limit threshold voltage at the grid connection point; This refers to the upper limit threshold voltage at the grid connection point. This is the rated voltage of the power grid.

[0101] Communication and Deployment: Control commands are issued via high-speed power line carrier (HPLC) or 4G communication, with a data exchange cycle of 5 minutes. The system is containerized using Docker, providing an ARM64 architecture image that is compatible with mainstream edge gateway hardware, such as Rockchip RK3568 and Allwinner H6.

[0102] The edge-end collaborative real-time control and command issuance mechanism adapts to the real-time dispatching needs and communication and hardware conditions of rural power distribution networks, achieving core technical effects such as rapid response, stability, reliability, and ease of deployment. This mechanism adopts an edge-end decoupled architecture, with the edge gateway independently completing real-time control and command issuance, without relying on cloud participation for millisecond / second-level dispatching, significantly reducing communication dependence and system latency. Local fast-response logic is designed to address voltage fluctuations; when voltage exceeds limits, the reactive power support of the photovoltaic inverter is prioritized; when reactive power capacity is insufficient, energy storage is linked for active / reactive power coordinated adjustment, effectively ensuring the grid connection point voltage qualification rate meets standards. The dispatching strategy follows Modbus RTU and IEC 60870-5-104 standard protocol conversion and issuance, ensuring compatibility with end-side photovoltaic inverters, energy storage converters, smart meters, and other equipment. Communication uses high-speed power line carrier or 4G, adapting to diverse rural communication environments, and a fixed data exchange cycle ensures stable control.

[0103] 4. Equipment health management and long-term economic optimization.

[0104] Energy Storage Health (SOH) Estimation: The cloud estimates the SOH daily based on historical charge and discharge data (cumulative charge and discharge volume, number of cycles) reported by the energy storage system, using a linear decay model, and then sends the updated SOH value to the edge gateway. The estimation frequency is once a day.

[0105] Lifecycle cost model: In the fitness function of the above two-point dynamic collaborative optimization process, the single charge-discharge loss cost of the energy storage device is included. Related to the current SOH value: , This is the initial loss coefficient. When SOH is lower than... When this happens, the system issues an operation and maintenance warning and appropriately reduces the priority of energy storage calls.

[0106] Equipment health management and long-term economic optimization enable comprehensive health control and improved system economy throughout the energy storage lifecycle. The cloud-based non-real-time management unit accurately estimates the State of Health (SOH) of energy storage on a daily basis, tracking its degradation status in real time. It dynamically correlates energy storage charging and discharging losses with SOH, integrating this information into the scheduling fitness function to automatically optimize energy storage deployment strategies, effectively controlling charging and discharging depth and frequency, reducing equipment wear, and extending lifespan. When SOH falls below a safe threshold, the system automatically issues a maintenance warning and lowers the priority of energy storage deployment, mitigating equipment operational risks. This mechanism simultaneously considers electricity purchase costs, renewable energy curtailment penalties, and energy storage loss costs, achieving a balance between short-term scheduling benefits and long-term equipment maintenance, significantly improving the overall operational economy and reliability of multi-energy systems in rural power distribution networks.

[0107] 5. Local autonomous deployment of the system.

[0108] The system employs containerization technologies, such as Docker packaging, enabling easy deployment on various ARM (Advanced Reduced Instruction Set Machine) architecture edge gateways. The cloud side is only responsible for lightweight model updates and macro-level policy configuration; the vast majority of computations are performed on the edge, achieving edge autonomy. Even during communication interruptions, the edge can still perform optimized calculations based on locally cached time-of-use pricing, SOH, and other data, maintaining basic scheduling functions.

[0109] The system's local autonomous deployment utilizes containerization technology, enabling rapid adaptation to various ARM-based rural edge gateway hardware. This facilitates deployment, enhances portability, and significantly lowers the barrier to entry for rural deployments. This model achieves clear decoupling of cloud and edge responsibilities, with core optimization calculations and real-time scheduling completed independently at the edge, without cloud involvement. In scenarios involving communication interruptions or network instability in rural power distribution networks, the edge can autonomously perform multi-energy coordinated scheduling and equipment control based on locally cached data such as time-of-use pricing and energy storage health, ensuring the continuous and stable operation of the power distribution network. This deployment method effectively adapts to the weak communication and low computing power conditions in rural areas, significantly improving system reliability and environmental adaptability.

[0110] Example 2 is an experimental example based on the rural smart distribution network energy management system that supports multi-energy coordination in Example 1.

[0111] 1. Simulation Platform and Examples

[0112] Simulation platform: MATLAB R2022a + Simulink; the power distribution network model is built based on the OpenDSS engine. The random seed is fixed at 0 to ensure the reproducibility of simulation results.

[0113] Test case: A modified IEEE 13-node distribution system was used, with 1MWp photovoltaic power and 500kW / 1000kWh energy storage integrated. The load curve used typical daily agricultural load data (including irrigation and residential loads) from a county. The simulation timescale was 24 hours, and the control cycle was 15 minutes. The hardware environment consisted of an Intel i5-8250U processor and 8GB of RAM.

[0114] Control group: Based on the standard particle swarm optimization algorithm (population 30, iterations 80 times).

[0115] 2. Optimize algorithm parameter calibration

[0116] Table 2 is the parameter calibration table for the optimization algorithm.

[0117] Table 2

[0118]

[0119] 3. Simulation Results

[0120] Computational efficiency: In a single round of optimization with 80 iterations, the average time is 3.5 seconds, which is basically equivalent to the control PSO algorithm.

[0121] Dispatch economy: Under typical daily operating conditions, the total system operating cost, i.e., the cost of purchasing electricity plus the penalty for abandoning electricity plus the energy storage loss, is improved compared with the control group.

[0122] Impact on energy storage lifespan: By introducing loss costs, the frequency and depth of energy storage utilization can be effectively controlled, which helps to extend the lifespan of the storage.

[0123] Voltage qualification rate: After the system is put into operation, the 24-hour qualification rate of the grid connection point voltage meets the national standard requirements.

[0124] This invention proposes a lightweight adaptive energy collaborative management system for rural power distribution networks. By employing a two-point collaborative optimization and adaptive search radius adjustment mechanism, the algorithm structure is effectively simplified while ensuring optimization performance. Simulation results show that the system can operate stably in typical rural power distribution network scenarios and possesses practical application potential.

[0125] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0126] 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. A rural smart power distribution network energy management system supporting multi-energy synergy, characterized in that, It adopts a three-layer collaborative architecture of cloud, edge, and terminal, including a terminal-side intelligent data acquisition and execution unit, an edge-side edge computing control unit, and a cloud-side non-real-time management unit; The edge computing control unit is equipped with a lightweight adaptive optimization algorithm based on dual-point dynamic collaboration. The algorithm adopts a dual-center collaboration and search radius adaptive adjustment mechanism. The edge computing control unit has edge autonomy capabilities and can complete multi-energy collaborative scheduling of the power distribution network based on local cached data when communication is interrupted. The cloud-side non-real-time management unit sends the day-ahead electricity price and daily energy storage health estimation model to the edge computing control unit according to a preset non-real-time cycle. It receives and stores historical data uploaded by the edge computing control unit and analyzes it. It only triggers data interaction and model updates during the non-real-time cycle and does not participate in millisecond / second-level real-time scheduling. It only performs non-real-time management tasks. The end-side intelligent acquisition and execution unit completes the acquisition of multi-energy and operation data of the power distribution network and executes the control commands issued by the edge computing control unit.

2. The rural smart distribution network energy management system supporting multi-energy synergy according to claim 1, characterized in that, The end-side intelligent data acquisition and execution unit includes a photovoltaic inverter, an energy storage converter, and a smart meter; The non-real-time management tasks performed by the cloud-side non-real-time management unit include the issuance of day-ahead electricity prices, updating of the daily energy storage health estimation model, and storage and analysis of historical operation data.

3. The rural smart distribution network energy management system supporting multi-energy synergy according to claim 1, characterized in that, The execution flow of the lightweight adaptive optimization algorithm based on two-point dynamic collaboration includes data acquisition and state initialization, two-point dynamic collaborative optimization, real-time control and command issuance of edge-end collaboration, equipment health management and long-term economic optimization, and local autonomous deployment of the system.

4. The rural smart distribution network energy management system supporting multi-energy synergy according to claim 3, characterized in that, The data acquisition and state initialization process includes normalizing the acquired multi-energy data, operational data, and cloud-based data, and initializing two core state points for the scheduling strategy: development point A and exploration point B. Independent search radii are bound to the two core state points and initial values ​​are assigned. The dimensions of development point A and exploration point B are consistent with the number of distribution network scheduling decision variables.

5. A rural smart distribution network energy management system supporting multi-energy synergy according to claim 3, characterized in that, The two-point dynamic collaborative optimization first evaluates the scheduling strategies corresponding to development point A and exploration point B by constructing a fitness function. The fitness function formula is as follows: ； In the formula, The fitness function value; For electricity purchase costs; The cost of punitive measures for curtailing renewable energy; Cost of energy storage equipment per charge and discharge cycle.

6. A rural smart distribution network energy management system supporting multi-energy synergy according to claim 5, characterized in that, After completing the fitness evaluation, the dual-point dynamic collaborative optimization performs a point exchange operation to ensure that development point A is the optimal scheduling strategy point obtained by the current optimization and exploration point B is the exploratory scheduling strategy point. Then, candidate solutions are generated based on development point A, exploration point B and their bound search radii. The generation of candidate solutions adopts one of two modes: single-dimensional perturbation and multi-dimensional perturbation.

7. A rural smart distribution network energy management system supporting multi-energy synergy as described in claim 6, characterized in that, After the candidate solutions are generated, they are subjected to inverse normalization, decision variable boundary truncation, and dynamic constraint verification of energy storage charge state. The candidate solutions with the best fitness are selected from the candidate solutions corresponding to development point A and exploration point B, respectively, and development point A and exploration point B are updated.

8. A rural smart distribution network energy management system supporting multi-energy synergy as described in claim 6, characterized in that, In the dual-point dynamic collaborative optimization, the search radius is dynamically adjusted according to a preset iteration interval. This includes updating the core state points of the archives based on local archives with a fixed capacity, and updating the search radius through preset coefficients, so that the search radius of development point A gradually shrinks to a preset lower limit and the search radius of exploration point B gradually expands to a preset upper limit.

9. A rural smart distribution network energy management system supporting multi-energy synergy according to claim 3, characterized in that, The real-time control and command issuance of the edge-end collaboration includes scheduling strategy decomposition, local rapid response and communication deployment. The local rapid response means that when the grid connection point voltage exceeds the preset threshold, the photovoltaic inverter is called to provide reactive power support for voltage. When the reactive power regulation capability is insufficient, the energy storage is called to perform active / reactive power coordinated regulation.

10. A rural smart distribution network energy management system supporting multi-energy synergy according to claim 1, characterized in that, The system is encapsulated using containerization technology and adapted to edge gateway hardware with advanced reduced instruction set machine architecture. Control commands are issued via power line carrier or wireless communication and use a fixed data interaction cycle.

Images