Power distribution box coordination control method and system for distributed energy access

Abstract

The application discloses a power distribution box coordination control method and system for distributed energy access, belongs to the technical field of power distribution box coordination control methods, and comprises the following steps: a local access perception model based on a power distribution box node is established, and data of distributed energy equipment is collected in real time; a distributed energy access profile of the power distribution box node is generated based on time and space feature fusion, and is used for describing distributed energy output timing, spatial concentration and influence correlation on the system; an embedded micro real-time power flow simulation engine is adopted, local power flow direction, stress distribution and power flow backflow risk are dynamically predicted and quantified according to the access profile, and a risk assessment result is output; the application has the beneficial effects that: by embedding a regulation and control intention generator and a power flow modeling engine in each power distribution box, the power distribution box has the ability of local perception, real-time judgment and autonomous regulation and control, and no longer depends on a central control system, so that the rapid response capability of the system in the face of sudden power flow fluctuation or operation abnormality is remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of distribution box coordinated control method, and specifically relates to a distribution box coordinated control method and system for distributed energy access. BACKGROUND

[0002] There are still a series of technical and systematic deficiencies in the current distribution box coordinated control method, which is more prominent under the background of high penetration rate of distributed energy, diversification of power users and rapid growth of source and load behavior uncertainty, and restricts the intelligent level and sustainable development ability of the operation of the distribution system.

[0003] Firstly, the traditional distribution box mainly undertakes passive response functions such as fault protection and current limiting tripping, and the control mode is mainly driven by static rules, which lacks dynamic running state sensing ability and cannot make active judgment and control response to the real-time access behavior of distributed energy, load fluctuation trend or power quality change. Especially in the presence of typical disturbance events such as rapid change of photovoltaic output and centralized charging of electric vehicles, control lag, reaction failure and fault misjudgment are prone to occur. Secondly, most of the existing control schemes still rely on the central scheduling architecture, and the distribution box as the end device only executes the strategy instructions issued by the upper level, lacking local decision and autonomous judgment ability. This top-down mode is difficult to accurately cover the local characteristics of each node due to communication delay, untimely data backhaul and insufficient control granularity in actual operation, especially after large-scale distributed source access, the dynamic linkage of power flow between nodes is frequent, and the central control system is difficult to balance the contradiction between local fluctuation and overall stability, and the real-time and adaptability of regulation and control is obviously insufficient. In addition, most of the current control methods lack modeling means of spatio-temporal behavior fusion, and the time periodicity of distributed energy output and the spatial coupling of load response behavior cannot be effectively identified and modeled. The system can only make single-point control judgment based on the current state, and lacks the ability of forward-looking identification of trend imbalance or potential risk. In the network with multiple access points and frequent power flow reversal, it cannot accurately predict the high-risk operating state, nor can it develop proactive control strategies in advance, so that the whole system is more likely to be unstable or out of control in a short time. On the other hand, the control strategy itself is too rigid, and the heterogeneity of different types of distributed resources and the flexibility level difference of loads are not fully considered, so that in the case of response, it is not possible to selectively adjust the priority response object, for example, to control non-interruptible loads and movable loads together, which may cause user experience to decline or even equipment failure.

[0004] Furthermore, current distribution boxes generally lack direct communication and coordination mechanisms. Even if some systems achieve point-to-point information transmission, it is limited to data sharing rather than control intent coordination, making it impossible to establish dynamic strategy consensus during operation and thus preventing the formation of a multi-node collaborative network with regional autonomy. This information isolation means that each distribution box can only control its own node resources independently, making it difficult to achieve joint optimization behaviors such as resource complementarity, response sharing, and load transfer. From a system perspective, this wastes the potential value of distributed resources and increases the redundancy configuration cost of the entire system. Moreover, current strategies lack a closed-loop feedback mechanism after execution, making it impossible to verify the accuracy and effectiveness of the strategy through actual execution results, let alone self-correct and adaptively adjust the control model. This leads to repeated execution of ineffective or inefficient control logic once the strategy deviates from the actual operation, which will exacerbate system operation errors and risk exposure in the long run. In addition, most traditional systems have not introduced a control strategy evolution mechanism. Even if initial optimization logic is deployed, it is difficult to learn from experience in multiple rounds of operation. The controller remains in a low-intelligence mode of repeated triggering and execution, which is seriously out of step with the requirements of dynamic evolution of intelligent control in the current complex power grid environment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for coordinated control of distribution boxes for distributed energy access, thereby solving some of the drawbacks and shortcomings pointed out in the background art.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems includes the following steps:

[0007] A local access sensing model based on distribution box nodes is established to collect data from distributed energy devices in real time. Based on the fusion of the time and space features, a distributed energy access profile of the distribution box nodes is generated to describe the timing, spatial concentration, and correlation of the output of distributed energy and its impact on the system.

[0008] An embedded micro real-time power flow simulation engine is used to dynamically predict and quantify the local power flow direction, stress distribution, and energy flow backflow risk based on the access profile, and output the risk assessment results. A control intention generator is introduced to construct a local operating state model based on the risk assessment results, distributed energy regulation capabilities, and load flexibility levels, and generate a goal-oriented autonomous control plan. A summary of the autonomous control plan is broadcast among the distribution box nodes through a neighborhood communication unit to trigger a collaborative negotiation mechanism to generate coordinated execution actions, including load shifting, energy storage charging and discharging postponement, and distributed energy output limitation or priority strategies.

[0009] After executing the coordinated execution action, the load balancing, voltage stability and harmonic suppression effects are continuously monitored, the monitoring results are analyzed for deviation from the autonomous regulation plan, and the parameters of the regulation intention generator and the electric energy flow simulation engine are adaptively set using a dynamic control mapping function.

[0010] Further, the establishment of the local access awareness model based on the distribution box node includes: high-frequency sampling of real-time electrical parameters such as power, voltage and frequency of the distributed energy equipment; recording the photovoltaic power generation sunshine output period, the energy storage equipment charging and discharging period and the electric vehicle charging behavior period time information; and obtaining the geographical coordinates, electrical topology position and electrical distance spatial information between the distributed energy equipment and the main line and transformer.

[0011] Further, the time and space feature fusion generates an access profile using a multi-modal feature fusion algorithm, forms a time feature model in the time dimension according to the operation cycle and output change law, forms a space feature model in the space dimension according to the electrical distance and line impedance parameters, and fuses the two types of models into the access profile through a weight adaptive mechanism.

[0012] Further, the access profile is used to construct a prediction model based on historical and real-time data, determine high-risk periods and sensitive distribution box nodes, and map the prediction results into a preventive collaborative control strategy.

[0013] Further, the neighborhood communication unit establishes a regional energy flow collaborative regulation network by sharing the access profile and autonomous regulation plan summary, and uses a local game algorithm in the collaborative negotiation mechanism to determine the coordinated execution action without revealing the device information.

[0014] In the above scheme, specifically:

[0015] Each distribution box internally constructs a distributed energy access profile based on time and space features, including:

[0016] Node power change trend, electrical distance between node and main line, energy type and regulation response capability, time concentration of node operation behavior.

[0017] Through the neighborhood communication mechanism, each distribution box shares its access profile information to construct a decentralized and dynamically evolving regional energy flow collaborative regulation network. A prediction model of risk significance and regulation coupling coefficient is constructed to identify high-risk periods and sensitive nodes in future operation, and to generate regulation strategies in advance, including load shifting, energy storage pre-discharge, distributed energy suppression or preferential output;

[0018] A node-level risk regulation mapping function is introduced, defined as follows:

[0019] ;

[0020] wherein:

[0021] represents the control intention index function of the th distribution box node at time , used to determine the current and future control dominance of the node in the regional network; represents the risk significance weight of the node at time , which integrates the load mutation rate, voltage fluctuation amplitude and power flow reverse probability; represents the local resource adjustable capacity function, which is composed of the remaining capacity of energy storage, the variable interval of photovoltaic and the flexibility factor of load; represents the cooperative control coupling coefficient with the adjacent node , reflecting the tightness of the power flow coupling between the node and its adjacent nodes (the smaller the coupling path impedance, the greater the coupling coefficient); represents the integral of the dynamic state within the control window period , which embodies the cumulative evaluation of the system on the evolution of the continuous risk and the regulation potential; when reaches the system preset threshold, it indicates that the node has entered or is about to enter the regulation priority state; the control system will generate a regulation plan instruction set in advance according to the access profile and resource capacity of the node, and the regulation plan instruction set includes the following contents:

[0022] migrating flexible load to load trough period;

[0023] starting pre-charging or pre-discharging of the energy storage system;

[0024] limiting the output of intermittent sources such as photovoltaic in peak period;

[0025] organizing the power reconstruction path cooperative response between adjacent distribution boxes.

[0026] Further, the risk assessment result includes power flow direction, feedback risk and local overload probability, and the corresponding regulation target threshold is determined through the node-level risk-regulation mapping function.

[0027] Further, the autonomous regulation plan at least includes the following instruction sets: i) starting or inhibiting specified distributed energy equipment; ii) adjusting the charging and discharging power and duration of energy storage equipment; iii) switching, delaying or limiting the movable load; iv) sending a cooperative control request to adjacent distribution box nodes.

[0028] Further, the load flexibility level adopts the criticality index method, which weights and scores important load, interruptible load and movable load according to real-time adjustable power proportion, business tolerance and recovery cost.

[0029] Further, it is characterized that the dynamic control mapping function adopts a recurrent neural network to learn the monitoring feedback and the regulation deviation online, to update the control parameters at a time resolution of 0.5-5 seconds, and to realize fast self-adaptation to external working condition changes.

[0030] A power distribution box coordination control system for distributed energy access is composed of a plurality of power distribution box nodes interconnected through a neighborhood communication network, characterized in that each power distribution box node comprises:

[0031] a) a local access perception unit for real-time acquisition of distributed energy device state and environmental data; b) a space-time feature fusion unit for constructing a distributed energy access profile based on the data, describing the output time sequence, spatial concentration and influence correlation on the system; c) a risk assessment unit with an embedded embedded real-time power flow simulation engine, for dynamically predicting and quantifying the local power flow direction, stress distribution and energy flow backflow risk according to the access profile, and outputting the risk assessment result; d) a regulation intention generation unit for constructing a local operating state model according to the risk assessment result, distributed energy regulation capacity and load flexibility level, and generating a target-oriented autonomous regulation plan; e) a neighborhood communication and collaborative negotiation unit for broadcasting the autonomous regulation plan summary among the power distribution box nodes, triggering the collaborative negotiation mechanism and aggregating the negotiation results; f) a coordination decision unit for generating coordinated execution actions, including load peak shifting, energy storage charge and discharge postponement, and distributed energy output restriction or priority strategy, by comprehensively considering the negotiation results of the node and neighboring nodes; g) an execution unit for implementing the coordinated execution actions; h) a monitoring and feedback correction unit for continuously monitoring the load balancing, voltage stability and harmonic suppression effect, analyzing the deviation between the monitoring results and the autonomous regulation plan, and adaptively setting the parameters of the regulation intention generation unit and the risk assessment unit using a dynamic control mapping function.

[0032] The power distribution box coordination control method for distributed energy access has the following remarkable beneficial effects under the background of widespread access of distributed energy, multi-point divergence of power flow, and highly uncertain load:

[0033] By embedding a regulation intention generator and an energy flow modeling engine in each power distribution box, it has the ability of local perception, real-time judgment and autonomous regulation, and no longer relies on the central control system, significantly improving the system's rapid response capability when facing sudden energy flow fluctuations or abnormal operation; through the neighborhood communication and regulation intention broadcasting mechanism, the power distribution boxes can share regulation intentions and resource states, build a regional control consensus network, realize horizontal deployment of distributed energy resources, load peak shifting and mutual assistance response, and avoid the island effect and resource waste caused by single-point regulation.

[0034] The access profile and risk prediction model can accurately identify the behavior characteristics and adjustment capacity of distributed devices such as photovoltaic, energy storage, electric vehicles, and effectively alleviate the problems of voltage out-of-limit, power feedback, branch overload and the like caused by large-scale access, and guarantee the stability of power grid operation; through the introduction of a dynamic risk assessment function and a strategy deviation feedback mechanism, the system can identify high-risk nodes and time periods in advance, and continuously optimize the control strategy through adaptive fine-tuning of control parameters. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The figure is a function relationship diagram of the local perception and fusion model of the distribution box node side distributed energy of the application.

[0036] Figure 2 The figure is a flow chart of intelligent simulation, control and collaborative processing of the distribution box.

[0037] Figure 3 The figure is a dynamic learning optimization closed-loop mechanism diagram of the distribution box node monitoring driven.

[0038] Figure 4 The figure is a function relationship diagram of the local perception and regional coordination of the D3 distribution box node of embodiment 1.

[0039] Figure 5 The figure is a function relationship diagram of autonomous control plan distribution and feedback closed loop of embodiment 2.

[0040] Figure 6 The figure is a schematic diagram of the multi-module coordination and regional coordination control structure dominated by the D3 node of embodiment 3. DETAILED DESCRIPTION

[0041] The specific embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0042] In combination with the accompanying Figure 1 The local access perception model is constructed at the distribution box node side, and the construction of the local access perception model is based on real-time monitoring devices inside or around the distribution box, including voltage, current, frequency sensors, digital data acquisition terminals and edge processing units. Multidimensional operation data of distributed energy devices connected to the node are collected through high-frequency sampling, and the distributed energy devices can include photovoltaic power generation components, wind power devices, energy storage systems and electric vehicle charging facilities, etc. The types of collected data include but are not limited to device output power, voltage fluctuation frequency, load change rate, energy storage charge and discharge state, electric vehicle charging behavior trajectory, and related environmental factors such as light intensity, temperature, sunshine period, etc.

[0043] Meanwhile, the location information and spatial attributes of the distributed energy equipment are combined, including the electrical topology location in the power distribution network, the electrical distance to the main line or transformer, the geographic coordinates and other spatial attributes, to construct a time and space feature fusion model. The fusion model unifies the above-mentioned time series data and spatial distribution data through a multi-modal data processing method, analyzes the output periodicity, volatility and load response lag characteristics of various types of distributed energy in the time dimension, forms a device output time feature model, calculates the energy coupling degree and geographic agglomeration between devices in the space dimension based on the impedance parameters and electrical connection relationship of the power distribution network line, forms an electrical space feature model, and further fuses the time and space features through a weight self-adaptive algorithm to generate a distributed energy access profile at the node level of the distribution box. The access profile can comprehensively reflect the output time series of the distributed energy connected to the node, the spatial concentration and the potential influence correlation of the operating indicators such as system power flow and voltage stability.

[0044] In combination with the drawings Figure 2As shown, the dynamic local grid state prediction and risk quantification of the distribution box node is realized by the embedded micro real-time power flow simulation engine, specifically: the simulation engine is based on the completed distributed energy access profile, combined with real-time collected data and historical operation curve of the node, executes the power flow prediction model and thermal stability evaluation algorithm, dynamically simulates the change of power flow direction of each node, the stress distribution state of line and equipment and the possible reverse flow risk of energy flow in future multiple time windows, and comprehensively outputs the quantitative risk assessment index, which is used to describe the operation uncertainty and potential abnormal risk that the local grid may face; after obtaining the above risk assessment results, the system further calls the control intention generator module, constructs the current operation state model of the distribution box node according to the risk level, the key risk nodes involved in the prediction results, the adjustable capacity of the current distributed energy equipment (such as the power regulation range of the photovoltaic converter, the state of charge SOC of the energy storage system, etc.) and the flexibility level of the local load (indicating the controllability and delayability of various loads participating in dispatching), and generates autonomous control plan with the goal of reducing prediction risk, improving power quality and ensuring voltage stability, which includes control target, control mode and expected control effect, etc.; then, through the neighborhood communication unit set in each distribution box node, the summary information of the autonomous control plan is transmitted in the form of broadcast among the adjacent nodes, which is used to trigger the collaborative negotiation mechanism, in this process, each node performs responsive analysis based on its own ability, current state and received control summary, and participates in the local optimization negotiation process, and finally jointly generates the coordinated execution control action, which covers but is not limited to load peak shifting strategy (time sequence distribution of delayable load among different nodes to reduce peak value), energy storage system charge and discharge postponement (automatic adjustment of energy storage response time and strategy according to grid state) and dynamic limitation or priority operation strategy of distributed energy output (such as limiting part of photovoltaic power output or preferentially guaranteeing new energy power supply for necessary load in weak grid period).

[0045] After executing the coordinated execution action, the load balancing, voltage stability and harmonic suppression effect are continuously monitored, combined with the attached Figure 3As shown, three core indicators are monitored by the monitoring module: first, the load balancing condition, which monitors the matching degree of power input and output in a unit of time and the dynamic response of load, is used to determine whether the load adjustment achieves the preset peak shifting target; second, the voltage stability, which collects the phase voltage and line voltage fluctuation range at the distribution box node, analyzes whether it exceeds the allowed deviation value or has voltage flicker problems; third, the harmonic suppression effect, which measures the total harmonic distortion (THD), high-frequency harmonic content and asymmetric waveform distribution in voltage and current signals in real time, evaluates the impact of distributed energy, especially nonlinear load access, on power quality. The above monitoring results are compared and analyzed with the control target parameters set in the original autonomous control plan to form a control deviation matrix, including power difference vector, voltage stability offset metric and harmonic target deviation index; and further sent to the dynamic control mapping function module for processing. The dynamic control mapping function is based on a recurrent neural network (RNN) architecture, which combines historical operation data and current deviation characteristics of the node to execute an online learning algorithm; without human intervention, the mapping relationship is updated in real time, which is a real-time function corresponding rule that quickly converts the monitored operation deviation into the control parameters that should be adjusted next.

[0046] Next, the key parameters in the control intention generator and the embedded micro real-time power flow simulation engine are adaptively adjusted, such as adjusting the load flexibility level weight, autonomous control priority coefficient, simulation prediction step, power flow sensitivity weight, etc., so that the next period control plan is more in line with the current operation situation in terms of strategy formulation and risk prediction accuracy, realizing high matching between control parameters and dynamic state of the power grid, ensuring the continuous safe and stable operation of distributed energy under high proportion access, and thus building a dynamic learning optimization closed-loop mechanism driven by monitoring, and comprehensively improving the autonomous control intelligence level and system collaborative adaptation ability of the distribution box node.

[0047] Embodiment 1:

[0048] In combination with the accompanying Figure 4 As shown, a suburban residential community is introduced to distributed energy reconstruction, the project covers 8 residential buildings totaling 120 households, and the deployed equipment includes: a total capacity of 350kW of rooftop photovoltaic system, 3 sets of distributed energy storage units (each battery capacity 150kWh / power output 60kW), and 12 public electric vehicle charging piles (total maximum concurrent power 180kW). The community manages the electricity load and distributed energy in each region through 6 medium-sized intelligent distribution box nodes. This paper focuses on the distribution box node numbered D3, which serves 2 residential buildings with 24 households, locally accesses photovoltaic capacity of 60kW, storage system capacity of 50kWh, and is equipped with 2 7kW charging piles, with a daily load peak of about 45kW.

[0049] In establishing the local access awareness model of D3 distribution box node, the system first configures real-time monitoring equipment, continuously collects all accessed photovoltaic, energy storage, electric vehicle charger output power, voltage, frequency and other electrical parameters at a time interval of 200 ms, and the sample is as follows: at 12:00 on June 10, 2024, the real-time output power of the roof photovoltaic in node D3 is 48.6kW, the voltage on the inverter side is 236.5V, the output frequency is 49.95Hz, the energy storage system is in charging state, the power is-11.3kW (negative sign indicates charging), the voltage is 240.3V, the frequency is 50.02Hz, and one of the two charging piles is charging a vehicle at a power of 5.2kW; such data is collected about 43,200 groups per day, and preliminary data preprocessing is performed.

[0050] At the same time, the model records time series feature information closely related to the operation behavior of the equipment, for example, by combining meteorological API with photovoltaic inverter data, a sunshine output period curve of the photovoltaic system is established, and an example is as follows: in June 2024, the daily average effective power generation period of the photovoltaic system is between 6:05 and 18:35, and the peak output period is concentrated between 10:50 and 13:20, and the system identifies it as a high sensitivity period; for energy storage equipment, by analyzing its operation log in the past two weeks, it can be identified that its charging and discharging period is generally concentrated in the early morning 2:00-5:30 charging and the evening 17:30-20:00 discharging, and the duration is stable at about 3 hours; the electric vehicle charging behavior period is obtained by analyzing the vehicle owner App log and the charging pile operation data, and the average charging initiation time is 18:15 on weekdays, and the average duration is 2 hours, while on weekends, the average charging starts at 14:30, and the peak concurrent time is 19:00-21:00.

[0051] In terms of spatial information, the awareness model calls the GIS interface and connects with the distribution network topology information, automatically extracts the geographic coordinates, belonging distribution line, and electrical distance to the main line and substation transformer of each distributed energy equipment in the node. In the example, the electrical distance from the photovoltaic system in node D3 to the substation transformer is 0.42 ohms, and the energy storage system is connected to the D3 main bus, and the electrical reactance of the line to the main line is 0.12+j0.24 ohms. Such spatial information and electrical parameters together constitute the basis variables for subsequent power flow simulation and access profile generation.

[0052] In summary, the local access perception model of the D3 node forms a complete data structure, including high-frequency electrical operation data (such as P = 48.6kW, U = 236.5V), output cycle behavior characteristics (such as the high-sensitivity time period of photovoltaic output 10:50-13:20), device operation frequency model (such as the discharge cycle of energy storage 17:30-20:00), and spatial topology characteristics (such as impedance path = 0.54Ω). The perception model not only has the characteristics of real-time, time sequence, and space fusion, but also can provide a dynamic image of distributed energy operation for subsequent access profile generation. On this basis, the application can further perform distributed power flow prediction, risk assessment, and autonomous regulation, laying a micro-modeling foundation for urban-level high-density distributed energy management. Through simulation experiment testing, the node access profile based on the above perception model used for power flow simulation has a mean square error (MSE) of 1.86% on the test set, which is much lower than 4.21% when the behavior cycle modeling is not introduced, verifying the feasibility and accuracy of the model under actual deployment conditions.

[0053] Taking the data of June 12, 2024 as an example, the photovoltaic system of the D3 node presents a typical "parabolic" change in the output curve of the day, the sunshine duration is 12 hours and 31 minutes, the effective power generation interval is 5:54 to 18:25, the peak value appears at 11:58, and the system takes 15 minutes as the time window to extract the output mean and standard deviation, obtaining the typical fluctuation characteristics of the photovoltaic system of the node on the time axis: the output stability index (in the middle of the day) is 0.91, the maximum change rate in the early morning is 4.5kW / 15min; the energy storage system analyzes its continuous 7-day charge and discharge records and finds that the average discharge start time is 17:35, the daily average discharge power is 14.7kW, the fluctuation is not large, and at least 12 hours of sleep state is maintained after each discharge; the electric vehicle charging operates from 18:15 to 20:35 on this day, with outputs of 6.2kW and 6.4kW respectively, and the two lasts for about 110 minutes. In summary, a feature vector group based on typical cycles is constructed in the time dimension: the photovoltaic output curve envelope function , the energy storage cycle function , and the electric vehicle load behavior function , which are used to describe the output fluctuation, operation rhythm, and peak-valley cooperation capability of the node in the time domain.

[0054] At the same time, the electrical location relationship and energy transmission capability model are established in the spatial dimension according to the distribution network topology structure. First, based on the geographic coordinates and wiring diagram of each device in the D3 node, the following key electrical parameters are obtained through network modeling: the impedance from the photovoltaic access point to the main line is 0.62Ω, the electric reactance from the energy storage system to the transformer is Ω, electric vehicle charging piles are connected in parallel to the end of the secondary branch, and the average electrical distance between the end node and the main node is 0.78Ω. The system constructs an electrical distance matrix in the spatial dimension. And in conjunction with the line resistivity and node access density Forming a local energy flow sensitivity coefficient matrix This is used to represent the energy coupling strength and location sensitivity between different distributed devices. Simulation results show that when the photovoltaic output suddenly increases to 60kW, the voltage drop on this electrical path is about 1.48%. If the energy storage is located upstream and discharges 15 minutes in advance, the risk of power flow feedback can be effectively mitigated, and the predicted voltage drop can be controlled within 0.94%.

[0055] Subsequently, this invention employs a multimodal feature fusion algorithm to integrate the aforementioned time feature model. , , Spatial feature model The overall mapping is transformed into a unified access contour feature vector. This process is completed by the adaptive fusion unit in the edge computing module of the distribution box, and adopts an improved weighted feature attention mechanism. The weight factors are automatically adjusted according to the real-time operating status. For example, when the photovoltaic output fluctuates sharply at noon, the weights of the time feature model are adjusted. Increased to 0.68, spatial feature weights It drops to 0.32; however, during the peak load period at dusk, the voltage deviation is mainly affected by the spatial topology. It rose to 0.61. The value decreased to 0.39. The final output access contour vector. ,in and These represent the normalized temporal and spatial feature tensors, respectively.

[0056] To verify the effectiveness of the fusion method, the system compared the performance of the non-fusion model and the fusion model in node load prediction and power flow direction judgment for three consecutive days. The results showed that the mean square error of prediction (MSE) of the fusion model decreased by 37.2%, the accuracy of power flow feedback risk identification increased from 84.6% to 94.1%, and the average lead time for voltage fluctuation warning was extended by 6.4 minutes.

[0057] This embodiment introduces a neighborhood communication unit to establish a regional decentralized energy flow collaborative control network by sharing access profiles and autonomous control summaries constructed by each node based on spatiotemporal characteristics. It assumes the test area includes six smart distribution box nodes (D1-D6), each covering residential users, rooftop photovoltaics, some energy storage, and several electric vehicle charging terminals. Each node possesses edge computing power and communication capabilities and has the following characteristics:

[0058] Take node D3 as an example, first build its distributed energy access profile, including:

[0059] ① Node power change trend: the maximum output variation rate ΔP in the past 24 hours is 17.6 kW / h, and the power fluctuation rate during noon is as high as 5.8 kW / 15min;

[0060] ② Electrical distance parameters: the equivalent resistance from D3 to the main line , reactance ;

[0061] ③ Energy type and regulation response capability: D3 contains 60kW photovoltaic, 50kWh energy storage (current SOC is 72%), with ±20kW adjustable response bandwidth, and energy storage charge / discharge response time delay <3 seconds;

[0062] ④ Operation behavior time concentration: load response is concentrated in 17:00-21:00, photovoltaic output is concentrated in 10:30-13:30, and the concentration index .

[0063] Each distribution box node broadcasts the above profile in the form of a compressed feature vector (such as a 32-dimensional encoding vector) in the neighborhood range through the LoRaMesh network (typical communication radius <300 meters, time delay <0.5s), and the adjacent nodes decode and fuse the received information to dynamically establish a node coordination graph network , where the vertices are distribution boxes, the edges represent node pairs with effective power flow coupling, and the connection strength is determined by impedance distance and power flow interdependence. The system will build a risk-resource-collaboration three-dimensional prediction model to identify high-risk operation periods and sensitive nodes within the next 1 hour. For example, the model identifies 12:00-13:00 on June 14 as a high-risk period, and D3 and D5 nodes have a power flow reverse probability , voltage fluctuation prediction , and load peak growth rate exceeding .

[0064] Based on the above prediction, the system introduces a node-level risk regulation mapping function to accurately evaluate the regulation dominance, which is defined as follows:

[0065] ;

[0066] Wherein the parameters are defined as follows:

[0067] is the risk significance weight, and the calculation formula is:

[0068] ;

[0069] The coefficient value range is as follows:

[0070] Wherein: is the active power output (or load) of the i-th node at time t; is the voltage amplitude change of the i-th node at time t, indicating the degree of voltage fluctuation; is the probability of power reverse sending (i.e. electric energy flows back to the power grid from the user end) of the i-th node at time t. is the resource adjustment capability function, and an example expression is as follows: ; Wherein SOC is the remaining capacity percentage of the energy storage, is the photovoltaic output variation bandwidth, is the load flexibility coefficient,

[0071] , ,

[0072] Wherein SOC is the remaining capacity percentage of the energy storage, is the photovoltaic output variation bandwidth, is the load flexibility coefficient, , , ;

[0073] is the coupling coefficient, which can be expressed as , wherein is the power flow impedance path between the i-th node and the j-th node, The smaller the coupling is, the stronger the coupling is. The system performs integral evaluation in the control window , and the D3 node calculates in the window. If is higher than the preset regulation dominant threshold of the system, , it is determined that the D3 is a "active regulation node", and the D5 and D6 are "response coordination nodes". Based on this, the D3 generates a regulation instruction set as follows:

[0074] 1) Migrate the interruptible load such as the operation period of the domestic water heater from 12:30 to 14:10 (load translation); 2) Start the pre-discharge task of the energy storage system, and discharge 30 minutes at a power of 16kW from 11:55, releasing energy of 7.2kWh; 3) Limit the maximum output of the photovoltaic to 53kW, to reduce the feedback risk; 4) Coordinate the power flow with the D5 node, guide its upstream line to support the load peak period, and share the power flow pressure of 4.5kW through the low impedance path.

[0075]

[0076] ​​​June 15, 2024, was the test date. This was a typical hot and sunny summer day with strong photovoltaic output and a surge in air conditioning load in the residential area, pushing the entire D area power distribution system to the brink of stress. From 9:00 AM, the power flow direction at node D3 showed a significant change. Originally acting as a load terminal (main line → D3 load), after 10:45 AM, due to a sudden increase in photovoltaic output to 63.1kW (exceeding the local transformer branch grid-connected capacity of 57kW), the node current reversed, forming a power flow feedback. The system's real-time monitoring showed that the reverse flow ratio (RFR) rose to 0.83, meaning that over 83% of D3's output was being fed back to the upstream grid. Simultaneously, the node current density... Branch resistance Voltage upward trend The local overload probability (LOP) of D3 is determined by the model to be 0.61 (exceeding the safety threshold of 0.5), which means that there is a significant risk of overload and voltage exceeding the limit in the system at this time.

[0077] The risk assessment module outputs three indicators for node D3 at the current moment: power flow direction index. (This indicates a reversal in the trend) Feedback risk factors Local overload probability The system uses a node-level risk-control mapping function of the following form to convert it into a specific control target threshold:

[0078] ;

[0079] in Indicates that node D3 is at time... The comprehensive score of the urgency of regulation is used to determine the level of its regulatory target, and the coefficient selection meets the following requirements. Configured according to the current system policy , , Substituting the data, we get:

[0080] ;

[0081] The system's preset control levels are as follows: If This indicates "no adjustment needed," 0.45–0.65 indicates "weak adjustment," and 0.65–0.8 indicates "moderate adjustment." This indicates a "forced control" state. Here... Exceeding the highest threshold, node D3 is directly classified as a "priority control node" by the system, and the control intent generator is immediately invoked to load high-priority instruction templates, specifically including:

[0082] 1) Execution of photovoltaic output hard limit: Set the inverter output maximum to 50kW and actively discard 13.1kW of output (using MPPT pruning strategy), with the voltage drop expected to recover to 2.1%; 2) Triggering of energy storage system pre-discharge: Start the energy storage system to discharge at 20kW to assist in absorbing power flow feedback from surrounding nodes and buffer bus pressure for 22 minutes; 3) Execution of load shifting command: Adjust the operating rhythm of the smart water heater and central air conditioning compressor to shift 4kW load to the 14:00 time period; 4) Broadcast linkage request: Send a coupling path reconfiguration request to nodes D2 and D4 to allow some power flow to be diverted through the D2 branch with lower impedance, with an estimated sharing capacity of 5.7kW.

[0083] The risk assessment module continuously monitors the data during this phase, updating the three risk factors every 5 seconds and reassessing them every 15 minutes. After control measures were implemented, at 11:30, node D3... , The voltage drop was successfully reduced to the "no control required" region. Control feedback data showed that the voltage drop was controlled at 1.7%, and no feedback overflow or voltage over-limit events occurred in the system, verifying the rationality of the mapping function and the effectiveness of the dynamic threshold mechanism. In summary, the risk assessment and control mapping mechanism proposed in this invention not only has advantages such as strong interpretability, good real-time performance, and configurable structural parameters, but also can accurately assess the control dominance and intervention level of each distribution box under the premise of multi-node collaboration and non-sharing of information privacy. This achieves a truly refined control strategy of "on demand, on weight, and on capability," significantly improving the robustness and safety margin of distributed autonomous control of the distribution network.

[0084] Example 2:

[0085] Combined with appendix Figure 5 As shown, this embodiment further demonstrates the actual issuance and execution process of the autonomous control plan, and explains the functions, parameter configuration logic, and actual effects of the four types of instruction sets covered in the plan one by one, to support the completeness and practicality of the distribution box coordinated control method for distributed energy access in this invention. Assuming this scenario still occurs during the midday period on June 15, 2024, node D3 predicts that it will face peak feedback pressure and local overload risk from 11:00 to 13:00, and the control mapping function outputs the control index. The threshold for mandatory system control has been exceeded. The autonomous control plan generator immediately invokes the high-response level template to generate the following four types of control instructions:

[0086] i) Start or inhibit specified distributed energy resource devices: Upon detecting that the photovoltaic inverter output is about to break through the on-grid upper limit (measured as 63.1 kW, limit value as 57 kW), the system sends a peak curtailment control instruction (ActivePowerCurtailment) to the D3 node photovoltaic controller, setting the inverter output upper limit to 50 kW, inhibiting 13.1 kW of peak output, thereby avoiding over-voltage protection action or grid disconnection. The instruction is issued in Modbus RTU format according to the protocol DL / T645-2007, and is successfully executed within 5 seconds. After the limit value is measured, the D3 node voltage is reduced from 248.2V to 243.1V, which does not exceed the alarm threshold.

[0087] ii) Adjust the charging and discharging power and duration of the energy storage device: Considering that the current SOC of the energy storage system under the D3 distribution box is 72%, with a dischargeable capacity of 14.4kWh, the system calculates the optimal discharge power and duration according to the predicted load valley time window, using the following objective function to solve:

[0088] , , ;

[0089] where , represents the required buffer energy. The final parameters are determined as , , the system performs a 30-minute discharge with a cumulative output of 7.99kWh, reducing the main bus feedback current by about 18%. The system records that the voltage rise during the discharge process is maintained at +1.2%, without causing harmonic or frequency fluctuations.

[0090] iii) Switch, delay or limit the shiftable load: The system identifies from the load side data that 6 intelligent water heaters and 8 air conditioner compressors in the D3 node are shiftable loads with delay capability, controlled by the user-authorized edge load agent module. According to the load influence coefficient evaluation model and the current load curve, the system selects to shift the start time of 4 water heaters from 12:05 to 14:00, delayed by 110 minutes, with a cumulative load release of 5.2kW, in addition to setting the output limit of 3 air conditioner compressors to 80%, reducing the running power from 3.5kW to 2.8kW for 1 hour, a total of 2.1kW peak shaving.

[0091] iv) Send a cooperative control request to the adjacent distribution box node: Due to the coupling impedance between D3 and D4, the coupling coefficient This approach is well-suited for coordinated power diversion. When D3 generates a high-risk flag (11:05), the system broadcasts a "coordinated voltage reduction request" via neighborhood communication, including information such as current feedback power, available transfer paths, and requested support window. Upon response, node D4 issues a charge reduction instruction to its own energy storage system, decreasing the charging power from 8kW to 0 during the period from 12:00 to 12:30, thus assisting D3 in maintaining the bus voltage level. Actual test results show that after the two nodes work together, the power flow diversion rate decreases to 6.3%, and voltage stability improves by over 20%.

[0092] This embodiment constructs a complete flexibility level scoring process based on the current load list of node D3 and the operating data of the past 72 hours. There are a total of 36 types of load devices under this node, including residential lighting, water heaters, air conditioner compressors, rice cookers, refrigerators, electric vehicle chargers, etc. The system classifies these loads into three categories according to their control attributes: ① Important loads (such as refrigerators, medical equipment, corridor emergency lighting), ② Interruptible loads (such as air conditioners, water heaters), ③ Displaceable loads (such as washing machines, electric vehicle charging tasks, smart hot water pumps), which are denoted as categories A, B, and C, respectively.

[0093] The flexibility level is calculated using the following weighted scoring formula:

[0094] ;

[0095] in, Indicates load The flexibility score (the larger the value, the easier it is to be manipulated). It represents the real-time adjustable power ratio of the load at the current moment (i.e., the ratio of the current power of the load to the rated power). This indicates the business tolerance level (the smaller the value, the higher the tolerance for regulation). This represents the recovery cost coefficient (a comprehensive score including factors such as power outage restart, user response, and equipment lifespan impact). , , The weighting coefficients for the three dimensions are set by the system's experience as follows. .

[0096] Taking a smart water heater currently in operation within node D3 as an example, its current operating power is 3.8kW, and its rated power is 4.0kW. Business tolerance is based on feedback and configuration from home users. The recovery cost (requiring 5 minutes to reheat) is set to... Substituting into the calculation, we get:

[0097] ;

[0098] Similarly, analyzing an old refrigerator in D3 node, whose rated power is 0.12 kW, has been running for 8 hours in the constant temperature maintenance stage, and the real-time power is 0.05 kW, , because its business is uninterrupted (preserve food), set , the recovery cost is high (restart needs to wait for the refrigeration protection delay), set , substitute:

[0099] ;

[0100] The system classifies the total node load after scoring:

[0101] Flexible level I (high flexibility): , such as water heater, washing machine, air conditioner compressor, intelligent rice cooker;

[0102] Flexible level II (medium flexibility): , such as electric vehicle charger, circulating water pump;

[0103] Flexible level III (low flexibility): , such as refrigerator, gas water heater control module, corridor lighting.

[0104] At 11:00 on June 15, when the D3 node control plan is triggered, the system preferentially selects the load of flexible level I to perform peak shifting control, including delaying the operation of the water heater, limiting the output of the air conditioner, and canceling the pre-arranged start of the rice cooker; Level II loads such as electric vehicle charging tasks are sent to the delayed start queue and set to automatically start to avoid the load peak of 12:00-13:30; Level III loads are all shielded from control and do not participate in any power outage, peak shifting or shifting plan to ensure that users' basic living needs are not disturbed.

[0105] The execution result verifies that the flexible level evaluation mechanism accurately reflects the actual control potential and user experience risk. The data shows that the total controllable load affected by this round of control is 13.4 kW, accounting for 29% of the total load at the time, of which more than 90% are level I or II loads, the user complaint rate is 0, and the device restart failure rate is 0, fully demonstrating that the key index method has achieved a dynamic balance between power regulation potential and control risk under multi-target constraints, effectively supporting the deployability and adaptability of the autonomous control strategy module proposed in the invention in real home energy environment.

[0106] Taking the period of 12:00-13:00 on June 15, 2024 in D3 node as an example, after executing the control plan in this period, the monitoring and feedback unit starts to continuously collect the following key feedback data: load response deviation , node voltage offset , energy waste rate after photovoltaic limit control and the coincidence degree of the energy storage discharge curve and the planned discharge curve For example, at 12:05, the actual discharge power of the energy storage system is 14.2kW, the expected value is 16.0kW, , while the node voltage is 247.5V, the offset value , and the limited top-cut photovoltaic energy is 8.6kWh, the waste rate , the energy storage response coincidence degree .

[0107] Among the above, : load response deviation, indicating the difference between the actual response power of the load and the expected response power at time $t$; : node voltage offset, indicating the deviation between the actual measured voltage of the node and the target reference voltage, usually in volts (V) or can be converted to percentage; : photovoltaic energy waste rate after limit control, indicating the ratio of the reduced energy to the total energy that should have been generated within the time ;

[0108] : reduced energy; : originally available energy;

[0109] : energy storage system discharge coincidence degree, indicating the consistency between the actual discharge power curve of the energy storage system and the regulation and control plan curve, commonly measured by cosine similarity, mean square error or correlation coefficient. The RNN model structure adopts a standard three-layer recurrent network, the input layer receives the above feedback indicators and the system regulation and control history state within the last ten minutes , the middle layer contains 64 recurrent units, using tanh activation function, the output layer corresponds to three key control strategy parameters: photovoltaic limit dynamic threshold , energy storage discharge pre-adjustment delay compensation , load flexibility weight correction factor . The model loss function is the sum of the mean square error of the expected state and the feedback state:

[0110] ;

[0111] Among them: is the square of the load response deviation, measuring the degree of load non-response, and giving a punitive amplification to larger deviations; is the square of the node voltage offset, which strengthens the sensitivity to voltage fluctuations; is the square of the photovoltaic energy waste rate, which increases the punishment intensity of photovoltaic limit generation behavior; Energy storage discharge deviation square, The closer to 1, the smaller the value of the term, indicating that the tracking plan is good. The weight parameter is empirically valued , for highlighting the priority of voltage deviation on system stability. The model fine-tunes the input data of the last period every 5 seconds, using the Adam optimizer, with a learning rate of 0.001, and automatically switches to a 0.5 second high-frequency update mode to deal with sudden abnormalities if the system state fluctuates significantly (such as a pressure drop change > 5% or a power error > 2kW).

[0112] During 12:15 to 12:45, through the online learning and updating of the RNN model of the dynamic control mapping function, the system parameters are gradually optimized: the photovoltaic output limit is dynamically relaxed from 50kW to 53kW, the energy storage startup response delay is compressed from the original 30 seconds to 12 seconds, the load response coefficient is adjusted to release an additional 2.6kW adjustment space for flexible level I load without affecting user experience, and the overall control accuracy is significantly improved. The actual effect data shows that at 12:50, the system pressure drop is controlled at +2.1%, the photovoltaic waste rate is reduced to 5.8%, the energy storage discharge coincidence degree is increased to 0.92, and the load adjustment error is reduced by 43%.

[0113] Example 3:

[0114] In combination with the drawings Figure 6 As shown, a certain residential complex pilot deployed the power distribution box coordination control system for distributed energy access described in the present application, including a total of 6 intelligent power distribution box nodes D1 to D6, covering 126 households, a total roof photovoltaic capacity of 380kW, 4 sets of energy storage systems with a total capacity of 210kWh, and 18 electric vehicle charging piles. The system constructs a neighborhood communication network through LoRaMesh, each node is configured according to the modular architecture, and realizes the eight key function modules proposed in the present application. The following takes D3 node as the center, combined with the actual operation data of three consecutive working days (June 12-14), to illustrate the specific implementation and feasibility of the overall system structure and the coordination of each module:

[0115] a) The local access sensing unit is deployed on the D3 node main control board, integrating voltage, current, frequency sampling modules and weather acquisition terminals, with a sampling interval of 200ms, covering photovoltaic output, energy storage SOC, charging pile power, outdoor light intensity, temperature and other data. For example, at 11:00 on June 13, the collected data shows that the D3 roof photovoltaic output is 48.6kW, the energy storage is in a discharging state (-12.3kW), the electric vehicle load is 7.4kW, and the solar radiation intensity is 915W / m².

[0116] ​​b) The spatio-temporal feature fusion unit runs in the D3 edge computing module, which converts the above data into an access profile feature vector through a multi-modal fusion algorithm , where the time dimension contains the photovoltaic output periodic function , the energy storage response window , and the spatial dimension contains the electrical distance, node impedance, and output density matrix. The fusion weight is dynamically adjusted by an adaptive mechanism, such as setting , , the generated profile reveals that the D3 node will have a significant energy flow feedback trend from 11:30 to 13:00.

[0117] c) The risk assessment unit, based on the above access profile, calls the embedded power flow simulation engine (running period 3 seconds) to simulate the power flow direction, voltage deviation, and load distribution of the D3-D5 local area network segment. For example, on June 13th at noon, the D3 node power flow direction will be reversed, with a maximum feedback power of 15.4kW, and a node overload probability , the system determines that it is in a high-risk state and outputs the risk level "R3".

[0118] d) The regulation intention generation unit quickly generates an autonomous regulation plan based on the above risk results, D3 energy storage dischargeable power (17.2kW), and flexible load level (average score 0.78), including starting energy storage pre-discharge (16kW, 30 minutes), limiting photovoltaic (cutting to 52kW), delaying 4 water heaters to run until 14:00, and sending a collaborative regulation request to the D5 node. The plan is generated using state logic tree + priority weight rule, with a generation period of less than 1 second.

[0119] e) The neighborhood communication and collaborative negotiation unit sends a regulation summary to the D2, D4, and D5 nodes through LoRa broadcast, including risk level, demand support value, discharge coordination window, etc., with a communication interval of 1 second, using an encrypted summary package (length not exceeding 128 bytes). The collaborative mechanism triggers the D4 node to delay its energy storage plan, and the D5 node adjusts the electric vehicle charging curve, achieving regional load peak shaving.

[0120] f) The coordination decision unit analyzes all neighborhood responses locally in D3 and integrates them with the local execution plan to generate a final coordinated execution action table, such as: photovoltaic output limit 52kW, energy storage discharge 16kW, load release 5.3kW, external collaborative admitted power 9.2kW, and maximum voltage prediction value controlled from 248.7V to 243.6V.

[0121] g) The execution unit is composed of the internal PLC controller of the distribution box, the load-side intelligent socket, and the power switching module. The instruction issuing includes ModbusRTU, IEC61850 over TCP / IP, etc. After the instruction triggered at 12:00 on June 13, all target devices completed the state switching within 10 seconds, with a success rate of 100%, and the power consumers had no obvious perception.

[0122] h) The monitoring and feedback correction unit continuously tracks the D3 node voltage, load, power flow, energy storage response, and other indicators, records the control deviation, and delivers it to the RNN dynamic control mapping module. Taking the feedback on June 13 as an example: the expected peak shaving load was 21.5 kW, and the actual reduction was 20.3 kW, with a deviation ; The system automatically adjusts the load flexibility coefficient weight and the energy storage response compensation time delay parameters, and corrects the next round of peak shaving target prediction to 19.8 kW. The feedback closed-loop period is 5 seconds.

[0123] After three days of continuous operation, the system realized real-time regulation of each high-load period, with the maximum pressure drop reduced to 2.4%, the number of local power flow reversals reduced by 41%, and the energy waste rate reduced by 13.5%. There were no tripping or misoperation events on the user side, and the system stability, regulation accuracy, and user satisfaction were significantly improved, verifying the engineering feasibility, module interoperability, and adaptive control capability of the system architecture described in the invention in the urban distribution scenario.

Claims

1. A method for coordinated control of distribution boxes for distributed energy access, characterized in that, Includes the following steps: Establish a local access sensing model based on distribution box nodes to collect data from distributed energy devices in real time; Based on the characteristic information reflected by the distributed energy equipment in different time dimensions and spatial locations, the distributed energy access profile of the distribution box node is generated by fusion, which is used to describe the temporality, spatial concentration and correlation of the distributed energy output on the system operation status. An embedded micro real-time power flow simulation engine is used to dynamically predict and quantify the local power flow direction, stress distribution and energy flow backflow risk based on the access profile, and output the risk assessment results. A regulation intention generator is introduced to construct a local operating status model based on the risk assessment results, distributed energy regulation capacity, and load flexibility level, and to generate a goal-oriented autonomous regulation plan. The autonomous control plan is broadcast between the distribution box nodes via the neighborhood communication unit to trigger a collaborative negotiation mechanism to generate coordinated execution actions, including load shifting, energy storage charging and discharging postponement, and distributed energy output restriction or priority strategies. After executing the coordinated action, the load balance, voltage stability and harmonic suppression effect are continuously monitored. The monitoring results are compared with the autonomous control plan to perform deviation analysis. The parameters of the control intention generator and the power flow simulation engine are adaptively tuned using the dynamic control mapping function. Simultaneously, by combining the location information and spatial attributes of distributed energy devices, including their electrical topology location in the distribution network, electrical distance to the main line or transformer, and geographical coordinate spatial attributes, a time and space feature fusion model is constructed. The fusion model uses a multimodal data processing method to unify the modeling of time-series data and spatially distributed data. In the time dimension, it analyzes the output periodicity, volatility, and load response lag characteristics of various distributed energy sources to form a time characteristic model of equipment output. In the spatial dimension, it calculates the energy coupling degree and geographical clustering between equipment based on the impedance parameters and electrical connection relationships of the distribution network lines to form an electrical spatial characteristic model. Furthermore, it fuses the time and spatial characteristics through a weighted adaptive algorithm to generate a distributed energy access profile at the distribution box node level.

2. The distribution box coordination control method for distributed energy access according to claim 1, characterized in that, The establishment of the local access awareness model based on the distribution box node includes: The embedded data acquisition unit of the distribution box node performs high-frequency sampling of real-time electrical parameters such as power, voltage, and frequency of distributed energy equipment. It records the solar power output cycle of photovoltaic power generation, the charging and discharging cycle of energy storage equipment, and the charging behavior cycle of electric vehicles. It also obtains the geographical coordinates, electrical topology attachment points, and equivalent electrical distances between the distributed energy equipment and the main line and transformer. The real-time electrical parameters, time cycle information, and spatial location information are aggregated into a multidimensional dataset. Based on the multidimensional dataset, the local access perception model of the distribution box node is trained or calibrated. During operation, the local access perception model maps the continuously arriving real-time sampling stream into a distributed energy access feature vector that simultaneously encodes the temporality, spatial concentration, and impact indicators of distributed energy output on system operation. The feature vector is then output to the subsequent risk assessment module.

3. The distribution box coordination control method for distributed energy access according to claim 2, characterized in that, The access profile is generated by fusing time and space features using a multimodal feature fusion algorithm. In the time dimension, a time feature model is formed based on the operating cycle and output variation law. In the spatial dimension, a spatial feature model is formed based on the electrical distance and line impedance parameters. The two models are then fused into the access profile through a weight adaptive mechanism.

4. The distribution box coordination control method for distributed energy access according to claim 3, characterized in that, The embedded micro real-time power flow simulation engine takes the distributed energy access profile of the distribution box node as the first input and the historical operation data and real-time monitoring data of the node as the second input to construct a multi-layer prediction model that includes short-term power flow prediction, medium-term risk assessment and sensitive node identification. When the risk index exceeds the preset threshold, a preventive coordinated control command sequence is generated based on the matched control template and combined with the load gap, local voltage deviation and energy storage state parameters, and broadcast to the relevant distribution box node for execution through the neighborhood communication unit.

5. The distribution box coordination control method for distributed energy access according to claim 4, characterized in that, The neighborhood communication unit establishes a regional energy flow collaborative control network by sharing the access profile and the summary of the autonomous control plan, and uses a local game algorithm in the collaborative negotiation mechanism to determine the coordinated execution action without disclosing equipment information.

6. The distribution box coordination control method for distributed energy access according to claim 5, characterized in that, The risk assessment results include the power flow direction, feedback risk, and local overload probability, and the corresponding control target threshold is determined through a node-level risk-control mapping function.

7. The distribution box coordination control method for distributed energy access according to claim 1, characterized in that, The autonomous control plan includes at least the following instruction set: i) starting or suppressing designated distributed energy devices; ii) adjusting the charging and discharging power and duration of energy storage devices; iii) switching, delaying or limiting movable loads; iv) sending coordinated control requests to neighboring distribution box nodes.

8. The distribution box coordination control method for distributed energy access according to claim 7, characterized in that, The load flexibility level adopts the key index method, which scores important loads, interruptible loads, and relocatable loads by weighting them according to the proportion of real-time adjustable power, business tolerance, and recovery cost.

9. The distribution box coordination control method for distributed energy access according to claim 8, characterized in that, The dynamic control mapping function uses a recurrent neural network to learn online from monitoring feedback and control deviations, and updates control parameters with a time resolution of 0.5 seconds to 5 seconds to achieve rapid adaptation to changes in external operating conditions.

10. A distribution box coordination and control system for distributed energy access, comprising several distribution box nodes interconnected via a neighborhood communication network, characterized in that, Each distribution box node includes: a) Local access sensing unit, used to collect real-time data on the status of distributed energy devices and the environment; b) Spatiotemporal feature fusion unit, used to construct a distributed energy access profile based on the data, describing the power output timing, spatial concentration, and correlation with the system's impact; c) Risk assessment unit, with a built-in embedded micro real-time power flow simulation engine, used to dynamically predict and quantify local power flow direction, stress distribution, and energy flow backflow risk based on the access profile, and output risk assessment results; d) Regulation intent generation unit, used to construct a local operating state model based on the risk assessment results, the distributed energy regulation capability, and the load flexibility level, and generate a goal-oriented autonomous regulation plan; e) Neighborhood The communication and coordination negotiation unit is used to broadcast the summary of the autonomous control plan among distribution box nodes, trigger the coordination negotiation mechanism, and aggregate the negotiation results; f) the coordination decision unit is used to integrate the negotiation results of this node and neighboring nodes to generate coordinated execution actions, including load shifting, energy storage charging and discharging postponement, and distributed energy output restriction or priority strategies; g) the execution unit is used to implement the coordinated execution actions; h) the monitoring and feedback correction unit is used to continuously monitor load balance, voltage stability, and harmonic suppression effects, perform deviation analysis between the monitoring results and the autonomous control plan, and adaptively tune the parameters of the control intention generation unit and the risk assessment unit using a dynamic control mapping function; Simultaneously, by combining the location information and spatial attributes of distributed energy devices, including their electrical topology location in the distribution network, electrical distance to the main line or transformer, and geographical coordinate spatial attributes, a time-space feature fusion model is constructed. This fusion model uses multimodal data processing methods to uniformly model time-series data and spatially distributed data. In the time dimension, it analyzes the output periodicity, volatility, and load response lag characteristics of various distributed energy sources, forming a time characteristic model of device output. In the spatial dimension, it calculates the energy coupling degree and geographical clustering between devices based on the impedance parameters and electrical connection relationships of the distribution network lines, forming an electrical spatial feature model. Furthermore, a weighted adaptive algorithm is used to fuse time and spatial features, generating a distributed energy access profile at the distribution box node level.

Images