A power distribution network source-load-storage unified intelligent regulation method and system

Abstract

This disclosure relates to a unified intelligent control method and system for power distribution network sources, loads, and storage. The method includes: dividing source, load, and storage resources into a three-level hierarchical architecture, comprising an upper system layer, a middle line layer, and a lower distribution area layer; further dividing each layer of resources into three levels of hierarchical standards, including a priority control level, a supplementary control level, and a basic response level; aggregating the three-level hierarchical architecture using a preset lightweight scheduling strategy to obtain a source-load-storage aggregation architecture; aggregating all devices in the lower distribution area layer of the source-load-storage aggregation architecture, and aggregating devices with similar characteristics into the same virtual device unit; designing a lightweight unified control model based on a lightweight sensing architecture and a multi-dimensional controllable resource index system; and integrating the lightweight unified control model with artificial intelligence algorithms to establish a hierarchical control command mechanism for the source-load-storage aggregation architecture, and using the hierarchical control command mechanism to perform unified source-load-storage control of the aggregation architecture.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent power control, specifically to a unified intelligent control method and system for power distribution network sources, loads and storage. Background Technology

[0002] Against the backdrop of the accelerated construction of new power systems and the advancement of "dual carbon" goals, the distribution network, as a key link connecting power production and consumption, has integrated a large number of distributed new energy sources, controllable loads, and energy storage devices. With increasingly frequent interactions among multiple entities involved in power generation, load, and storage, its safe, stable, economical, and efficient operation faces severe challenges. Intelligent regulation technology for power generation, load, and storage has become the core support for solving these problems.

[0003] Currently, extensive research has been conducted both domestically and internationally on the regulation and control of power distribution networks, resulting in a series of technological achievements. However, many shortcomings remain to be overcome in light of practical application needs. International research on hierarchical aggregation of power distribution networks, lightweight sensing, and regulation technologies is relatively mature, particularly in the integration of edge computing and AI. Research institutions in the US and Europe have invested heavily in intelligent regulation and control of power distribution networks, focusing on building distributed energy resource aggregation platforms, promoting model compression and lightweighting, adopting distributed optimization and coordinated control methods, and promoting the application of edge AI technologies. Model compression and lightweighting simplifies deep learning models through techniques such as knowledge distillation, pruning, quantization, and network architecture search (NAS). Lightweight network models such as Google's MobileNet and Facebook's EfficientNet are widely used. Distributed optimization and coordinated control often employ methods such as multi-agent systems (MAS). Edge AI frameworks developed by companies like Google and IBM, such as TensorFlow Lite and Edge AI, are also widely used in power grid regulation and control. These technologies integrate power distribution resources, optimize regulation strategies, and achieve efficient power grid regulation and renewable energy consumption.

[0004] In recent years, domestic research on the joint regulation of power generation, load and energy storage has developed rapidly. Under the background of smart distribution networks and energy internet, the focus is on hierarchical regulation, integrated model construction, lightweight technology and edge intelligence applications. Research teams from Tsinghua University and North China Electric Power University have proposed a power generation, load and energy storage regulation strategy based on model predictive control (MPC) and robust optimization. The Institute of Automation of the Chinese Academy of Sciences has developed a series of lightweight deep learning models. State Grid, China Southern Power Grid and other enterprises are actively promoting the implementation of edge intelligence technology. Huawei's Atlas 200 edge computing module and Baidu's Paddle Lite framework are also applied to real-time monitoring and regulation of distribution networks, gradually improving the level of distribution network regulation.

[0005] In addition, scholars at home and abroad have also conducted in-depth research on key technologies for power generation, load and storage regulation in distribution networks. Reference [1] proposed a fully distributed dynamic economic dispatching method without a central dispatcher. Reference [2] proposed a fully distributed algorithm based on the alternating direction multiplier method. Reference [3] considered the two major stakeholders, the load dispatching center and the sub-network operator, to establish a two-level optimization model. Reference [4] proposed a target-level cascading theory to decouple the dispatching of distribution networks and microgrids. Meanwhile, scholars have integrated new technologies such as 5G communication and blockchain into the control model. Reference [5] points out that 5G communication is the key to power system technology innovation. Reference [6] promotes wind power consumption through load transfer and distributed optimization scheduling model. Reference [7] proposes the "cloud, pipe, edge, and terminal" architecture of distribution Internet of Things. Reference [8] constructs a multi-source-load-storage backup system and uses conditional value at risk (CVaR) to measure the uncertainty of wind power output. Reference [9] introduces blockchain into the virtual power plant model. Reference

[10] summarizes foreign experience of virtual power plants and proposes suggestions for domestic application. Reference

[11] clarifies the definition and operating characteristics of virtual power plants. Reference

[12] proposes a new scheduling model that considers the degradation of energy storage systems.

[0006] Meanwhile, scholars have explored hierarchical aggregation technology driven by both numerical and analog models, unified regulation model based on lightweight perception, and efficient regulation method based on lightweight model. Among them, data mining technology has been widely used in multi-source heterogeneous data fusion, state assessment and optimization regulation. Reference

[13] proposed a heterogeneous data fusion method, and Reference

[14] proposed a distribution network operation state assessment method based on big data technology. In terms of optimization regulation, Reference

[15] applied reinforcement learning to reactive power control of power systems, and Reference

[16] proposed a distribution network reactive power optimization method based on deep belief network. Cluster control has become an important regulation direction under high penetration distributed photovoltaic access. Reference

[17] pointed out the impact of distributed photovoltaic (DPV) access on distribution network, Reference

[18] proposed that high penetration DPV distribution network is suitable for cluster control, and Reference

[19] carried out research on coordinated planning control within the cluster. Reinforcement learning and deep reinforcement learning techniques have also been widely applied to multi-timescale scheduling of power systems. Single-agent deep reinforcement learning has yielded rich research results. Reference

[20] applied the RL algorithm to single-time economic scheduling. Multi-agent deep reinforcement learning is still in the exploratory stage. Reference

[21] conducted relevant research on distributed power scheduling. In response to the uncertainty of new energy output, scholars have adopted methods such as fuzzy optimization [8] and robust optimization

[19] . Reference

[20] adopted a multi-timescale control framework to reduce the impact of prediction errors.

[0007] Despite this, existing technologies still have significant shortcomings: the rationality and dynamic adaptability of hierarchical aggregation are insufficient; current cluster partitioning is mostly based on known loads, failing to fully consider dynamic factors such as future load growth and fluctuations in renewable energy output; and the coordination mechanisms between different levels are imperfect, failing to address the conflicting interests of different stakeholders. The adaptability of lightweight sensing and unified regulation models needs improvement; lightweight models do not fully integrate the characteristics of multi-source heterogeneous data in distribution networks, and unified regulation models have poor versatility and insufficient consideration for the synergistic effect of multi-timescale regulation. Furthermore, the practicality and robustness of artificial intelligence regulation methods are insufficient; deep reinforcement learning suffers from training limitations. The challenges include high difficulty, high sample dependence, and weak generalization ability. The coordination mechanism of multi-agent deep reinforcement learning is immature and has a weak ability to cope with emergencies. The integration and application of new technologies are not deep enough. The integration of 5G, blockchain, big data and other technologies with regulation technologies is in the initial stage and has not formed a systematic solution, making it difficult to release synergistic effects. The ability to cope with uncertainty is insufficient. The handling of uncertainties in new energy output and load demand is too conservative. Factors such as the degradation of energy storage systems and differences in equipment response characteristics are not fully considered. The prediction accuracy of the regulation model deviates from the actual operating effect, making it difficult to meet the regulation needs of the distribution network under the high penetration of new energy access.

[0008] The references are as follows: [1]Xu T, Wu W, Sun H, et al. Fully distributed multi-area dynamiceconomic dispatch method with second-order convergence for active distribution networks[J]. IET Generation, Transmission and Distribution.2017, 11(16): 3955-3965. [2]Zheng W, Wu W, Zhang B, et al. Fully distributed multi-areaeconomic dispatch method for active distribution networks[J]. IET Generation, Transmission and Distribution. 2015, 9(12): 1341-1351. [3]Ye C, Miao S, Li Y, et al. Hierarchical scheduling scheme for AC / DC hybrid active distribution network based on multi-stakeholders[J]. Energies. 2018, 11(10). [4]Xie M, Ji X, Hu X, et al. Autonomous optimized economic dispatch of active distribution system with multi-microgrids[J]. Energy. 2018, 153:479-489. [5] Zhang Ning, Yang Jingwei, Wang Yi, Chen Qixin, Kang Chongqing. 5G communication for ubiquitous power Internet of Things: technical principles and typical applications [J]. Proceedings of the CSEE, 2019, 39(14):4015-4025. [6] Chen Houhe, Wang Zixuan, Zhang Rufeng, Jiang Tao, Li Xue, Li Guoqing. Distributed optimization scheduling modeling of wind power grid-connected system with virtual power plant [J]. Proceedings of the CSEE, 2019, 39(09):2615-2625. [7] Lü Jun, Luan Wenpeng, Liu Riliang, Wang Peng, Lin Jiaying. Distribution Internet of Things Architecture Based on Comprehensive Sensing and Software Definition [J]. Power System Technology, 2018, 42(10):3108-3115. [8] Lü Mengxuan, Lou Suhua, Liu Jianqin, Wu Yaowu, Wang Zhidong. Coordination optimization of multiple types of backup power in virtual power plants with high proportion of wind power[J]. Proceedings of the CSEE, 2018, 38(10):2874-2882+3138. [9] She Wei, Hu Yue, Yang Xiaoyu, Gao Shan, Liu Wei. Virtual power plant operation and scheduling model based on energy blockchain network [J]. Proceedings of the CSEE, 2017, 37(13):3729-3736.

[10] Wei Xiangxiang, Yang Dechang, Ye Bin. Operation mode and implications of virtual power plant in energy internet [J]. Electric Power Construction, 2016, 37(04):1-9.

[11] Asmus P. Microgrids, virtual power plants and our distributed energy future[J]. The Electricity Journal, 2010, 23(10): 72–82.

[12] Zhou Bin, Cao Yijia, Li Canbing, et al. Optimal scheduling of virtual power plant with battery degradation cost [J]. IET Generation, Transmission & Distribution, 2016, 10(3): 712–725.

[13] Pan S, Morris T, Adhikari U. Classification of disturbances and cyber-attacks in power systems using heterogeneous time-synchronized data[J]. IEEE Transactions on Industrial Informatics, 2015, 11(3): 650-662.

[14] Liu Jiangdong, Wu Zhijian, Huang Zhenyong, et al. Modeling and analysis of influencing factors of planned power outages in distribution networks based on big data mining [J]. Journal of Test & Measurement Technology, 2021, 35(6).

[15] Vlachogiannis JG, Hatziargyriou N D. Reinforcement learning for reactive power control[J]. IEEE transactions on power systems, 2004, 19(3):1317-1325.

[16] Shao Meiyang, Wu Junyong, Shi Chen, et al. Reactive power optimization of distribution network based on data-driven and deep belief network [J]. Power System Technology, 2019, 43(6):10.DOI:CNKI:SUN:DWJS.0.2019-06-003.

[17] Shao Chenxu, Tang Yi. Multi-objective optimization method for three-phase load imbalance in distribution network with distributed photovoltaic power generation considering voltage over-limit risk [J]. Acta Energiae Solaris Sinica, 2023, 44(12):53-64. DOI:10.19912 / j.0254-0096.tynxb.2022-1216.

[18] Chen Chuliang, Li Xiaolu, Ji Kunhua, et al. Distribution network cluster division and optimized operation considering source-load-storage matching [J]. Electric Power Construction, 2023, 44(9):80-93.DOI:10.12204 / j.issn.1000-7229.2023.09.008.

[19] Chen Zexi, Wang Pu, Xiao Wanfang, et al. Comprehensive optimization configuration of flexible multi-state switches and distributed power sources in distribution networks considering reliability benefits [J]. Electric Power Construction, 2022, 43(6):8. DOI:10.12204 / j.issn.1000-7229.2022.06.010.

[20] Jasmin EA, Ahamed TPI, Raj VP J. Reinforcement learning approaches to economic dispatch problem [J]. International Journal of Electrical Power&Energy Systems, 2011, 33 (4): 836 - 845.

[21] Li D, Yu L, Li N, et al. Virtual - action - based coordinatedreinforcement learning for distributed economic dispatch [J]. IEEETransactions on Power Systems, 2021, 36 (6): 5143 - 5152.

[22] Liu J, Li J. Interactive energy-saving dispatch considering generation and demand side uncertainties: a Chinese study[J]. IEEETransactions on Smart Grid. 2018, 9(4): 2943-2953.

[23] Li Peng, Wang Jiahao, Li Canbing, et al. Collaborative optimization operation method of integrated energy system in industrial park considering source load uncertainty and equipment variable operating condition characteristics [J]. Proceedings of the CSEE, 2023(020):043.

[24] Kalantar-Neyestanaki M, Cherkaoui R.Coordinating DistributedEnergy Resources and Utility-Scale Battery Energy Storage System for PowerFlexibility Provision Under Uncertainty[J]. IEEE Transactions on SustainableEnergy, 2021, PP(99):1-1.DOI:10.1109 / TSTE.2021.3068630. Summary of the Invention To address the problems existing in the prior art, this disclosure proposes a unified intelligent control method and system for power distribution network sources, loads, and storage to solve at least one of the aforementioned technical problems. The technical solution adopted in this disclosure is as follows: In a first aspect, this disclosure provides a unified intelligent control method for power generation, load, and energy storage in a distribution network, the method comprising: S100. According to the physical node distribution of distribution network substations, transmission lines and distribution transformer areas, the source-load-storage resources are divided into the upper system layer, the middle line layer and the lower transformer area layer. The upper system layer, the middle line layer and the lower transformer area layer are arranged from top to bottom to form a three-level hierarchical architecture. S200, based on equipment type, functional characteristics, operating characteristics and quantitative models of various equipment, divides resources at each level into a three-level classification standard including priority control level, supplementary control level and basic response level; S300. The three-level hierarchical architecture is aggregated hierarchically by a preset lightweight scheduling strategy to obtain a source-load-storage aggregation architecture. S400: Aggregate all devices in the lower layer of the source-load-storage aggregation architecture, and aggregate devices with similar characteristics into the same virtual device unit. S500, based on a pre-built lightweight perception architecture and a multi-dimensional adjustable resource index system, designed a lightweight unified adjustment model; S600: By integrating the lightweight unified adjustment model with artificial intelligence algorithms, a hierarchical control command mechanism is established for the source-load-storage aggregation architecture. The source-load-storage aggregation architecture is then subject to unified source-load-storage control through this hierarchical control command mechanism.

[0009] Optionally, in S300, the lightweight scheduling strategy includes sequentially performing upward resource aggregation and dimensionality reduction, intra-layer resource optimization decision-making, and downward instruction decomposition and implementation.

[0010] Optionally, upward resource aggregation and dimensionality reduction includes: aggregating the basic response-level resources and priority control-level resources at the lower level of the distribution area into a virtual energy storage unit for the distribution area; aggregating multiple virtual energy storage units of the same distribution area within the same line at the middle level of the line into a line aggregation unit; and aggregating the line aggregation units, priority control-level resources, and supplementary control-level resources within the radiation range of each substation at the upper level of the system into a virtual control resource pool for the entire network.

[0011] Optionally, the intra-layer resource optimization decision includes: performing intra-layer resource optimization for the upper layer of the system, the middle layer of the line, and the lower layer of the transformer area, respectively.

[0012] Optionally, the downward instruction decomposition and implementation includes: parsing the global control instructions for the power distribution network's source, load, and storage, and decomposing the global control instructions from top to bottom to each layer and the equipment resources in each layer of the source, load, and storage aggregation architecture, thereby ensuring that the instructions can be executed accurately at the layer and even the device level.

[0013] Optionally, in S300, The total output of the virtual energy storage unit in the transformer area is the sum of the net output of household photovoltaic, distributed energy storage, electric vehicles and adjustable loads in the transformer area. The equivalent capacity and equivalent charging and discharging power are obtained by summing up the capacity and power constraints of each device. The total output of the line aggregation unit is the sum of the net output of the virtual energy storage units in each transformer area within the line, and it meets the maximum transmission power constraint of the line. The total output of the network-wide virtual control resource pool is the sum of the net output of each line aggregation unit within the substation's radiation range and the output of centralized photovoltaic power stations, thermal power plants, centralized energy storage power stations, and wind power plants, and it meets the network-wide supply and demand balance constraints.

[0014] Optionally, in step S400, the step of aggregating devices with similar characteristics into the same virtual device unit specifically includes: S401: At the data fusion level, multi-dimensional operational data of each device is collected, and after data preprocessing and feature extraction, multi-dimensional operational characteristics of each device are obtained. S402: Based on the multi-dimensional operational characteristics of each device, the data similarity between different combinations of devices is calculated. S403: By fusing data similarity and preset physical compatibility constraints, the aggregated similarity between different combinations of devices is obtained; S404: Combine different devices with similarity exceeding a preset similarity threshold into a single virtual device unit.

[0015] Optionally, in S500, the lightweight sensing architecture is preferably a three-level flat architecture of "terminal sensing - direct transmission - edge decision-making" to solve the pain points of high data latency and large redundancy in traditional architecture.

[0016] Optionally, in S500, the construction method of the multi-dimensional adjustable resource indicator system includes: using the lightweight sensing architecture to collect the core control information of the source-load-storage aggregation architecture, and establishing an adjustable resource indicator system covering dimensions such as control range, control cost, and time scale.

[0017] Optionally, the method may further include: S700. The source-load-storage unified control effect of the hierarchical and graded control instruction mechanism is verified and optimized through a multi-dimensional evaluation system.

[0018] A second aspect of this disclosure provides a unified intelligent control system for power distribution network sources, loads, and energy storage, the system comprising: The three-level hierarchical architecture module is used to divide the source-load-storage resources into the upper system layer, the middle line layer, and the lower distribution area layer according to the physical node distribution of the distribution network substations, transmission lines, and distribution areas. The upper system layer, the middle line layer, and the lower distribution area layer are arranged from top to bottom to form a three-level hierarchical architecture. The three-level classification standard module is used to divide resources at each level into three levels, including priority control level, supplementary control level and basic response level, based on equipment type, functional characteristics, operating characteristics and quantitative models of various equipment. The source-load-storage aggregation architecture module is used to perform hierarchical aggregation of the three-level hierarchical architecture through a preset lightweight scheduling strategy to obtain the source-load-storage aggregation architecture. The aggregation module within the distribution area is used to aggregate all devices in the lower layer of the source-load-storage aggregation architecture, and to aggregate devices with similar characteristics into the same virtual device unit. The adjustment model module is used to design a lightweight unified adjustment model based on a pre-built lightweight perception architecture and a multi-dimensional adjustable resource index system. The unified control module is used to integrate the lightweight unified control model with artificial intelligence algorithms to establish a hierarchical control command mechanism for the source-load-storage aggregation architecture. The hierarchical control command mechanism is used to perform unified control of the source-load-storage aggregation architecture.

[0019] Optionally, the system may further include: The verification and optimization module is used to verify the source-load-storage unified control effect of the hierarchical and graded control instruction mechanism through a multi-dimensional evaluation system.

[0020] In a third aspect, this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the unified intelligent control method for power generation, load, and storage in a distribution network as described above.

[0021] In a fourth aspect, this disclosure provides an electronic device including a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the unified intelligent control method for power generation, load and storage in a distribution network as described above.

[0022] Compared with the prior art, this disclosure has the following advantages: (1) Significantly improved resource aggregation efficiency: The hierarchical aggregation architecture based on dual-drive digital and analog systems proposed in this disclosure achieves scientific classification and efficient aggregation of source, load, and storage resources from physical topology to functional characteristics through a three-level hierarchical architecture of "system-line-transformer area" and a three-level hierarchical structure of "priority control level, supplementary control level, and basic response level". The upward aggregation and dimensionality reduction mechanism transforms massive scattered devices into high-level virtual control units, which greatly reduces the computational complexity of system-level scheduling and solves the problems of data explosion and computational bottleneck in traditional scheduling.

[0023] (2) Significantly faster control response speed: The lightweight sensing architecture constructed in this disclosure adopts a three-level flat design of "terminal sensing - direct transmission - edge decision", skipping the intermediate forwarding nodes in the traditional cloud-edge-device architecture, reducing the transmission volume by more than 70%, and controlling the latency to within 1 second. Combined with the application of the simplified particle swarm optimization algorithm in the lightweight unified control model, it realizes rapid control decisions at the minute or even second level, which is significantly better than the hourly response cycle of traditional centralized scheduling.

[0024] (3) Enhanced Physical Feasibility of Aggregation Results: The dual-drive similarity aggregation rule proposed in this disclosure extracts common features of multi-dimensional operational data at the data fusion level to calculate data similarity, and introduces a physical constraint model at the equipment mechanism level to calculate physical compatibility. The two are then weighted and fused as the aggregation criterion. This mechanism ensures that the aggregated virtual control unit not only has feature similarity at the data level, but also meets the physical operational feasibility requirements, effectively avoiding the problem that the aggregation results cannot be actually executed due to pure data-driven clustering.

[0025] (4) Deep integration of artificial intelligence algorithms and lightweight models: This disclosure organically integrates simplified long short-term memory network algorithms, deep reinforcement learning algorithms, and lightweight models to form a collaborative model of "lightweight computing + intelligent decision-making". The simplified long short-term memory network algorithm reduces computational complexity while ensuring prediction accuracy. The deep reinforcement learning algorithm learns the optimal strategy autonomously in the simulated environment using the lightweight model. The distributed energy aggregation flexibility solution algorithm improves computational efficiency by more than 40% and reduces computational conservatism by 50% through closed-loop sampling training, achieving high-precision and high-efficiency intelligent control under the condition of limited edge computing resources.

[0026] (5) Precise decomposition and closed-loop feedback of control commands: The hierarchical control command mechanism established in this disclosure, which decomposes from top to bottom and provides feedback from bottom to top, precisely decomposes the global control objectives to line aggregation units, transformer area virtual energy storage units, and individual devices, ensuring that commands at each level strictly meet equipment constraints, power balance constraints, and line transmission constraints. Simultaneously, through real-time feedback adjustments and a multi-dimensional evaluation system, quantitative assessment and iterative optimization of the control effect are achieved, ensuring the safe and stable operation of the distribution network in scenarios with a high proportion of new energy access.

[0027] (6) Multi-scenario adaptability and economic optimization: The multi-dimensional controllable resource index system established in this disclosure covers three dimensions: control range, control cost, and time scale. Combined with the scenario-based strategy library, it can flexibly call the optimal control strategy according to different operating scenarios such as load peak, valley, and fault recovery, so as to minimize the control cost under the premise of ensuring power grid safety and improve the overall economic efficiency of distribution network operation. Attached Figure Description

[0028] The accompanying drawings, which form part of this application, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0029] Figure 1 A flowchart of a unified intelligent control method for power distribution network sources, loads and storage provided in Embodiment 1 of this disclosure.

[0030] Figure 2 This is an example diagram of the three-level hierarchical architecture described in Embodiment 1 of this disclosure.

[0031] Figure 3 This is a topology diagram of the power distribution network used in the simulation test of Embodiment 1 of this disclosure.

[0032] Figure 4 This is a schematic diagram of the integrated dispatchable power of the transformer area in the simulation test of Embodiment 1 of this disclosure.

[0033] Figure 5 This is a schematic diagram of the photovoltaic dispatchable power in the simulation experiment of Embodiment 1 of this disclosure.

[0034] Figure 6 This is a schematic diagram of the load dispatchable power in the simulation test of Embodiment 1 of this disclosure.

[0035] Figure 7 This is a schematic diagram of the integrated scheduling output in the simulation test of Embodiment 1 of this disclosure.

[0036] Figure 8 This is a schematic diagram of the photovoltaic dispatch output in the simulation test of Embodiment 1 of this disclosure.

[0037] Figure 9This is a schematic diagram of the load dispatch output in the simulation test of Embodiment 1 of this disclosure.

[0038] Figure 10 This is a schematic diagram of the source-load-storage data perception and AI training process supporting scheduling decisions in the simulation experiment of Embodiment 1 of this disclosure.

[0039] Figure 11 This is a schematic diagram of the multi-dimensional evaluation system for the simulation experiment in Embodiment 1 of this disclosure.

[0040] Figure 12 This is an architecture diagram of a unified intelligent control system for power distribution network sources, loads and storage provided in Embodiment 2 of this disclosure. Detailed Implementation

[0041] The present disclosure will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0042] The following detailed descriptions are exemplary and intended to provide further detailed explanation of this disclosure. Unless otherwise specified, all technical terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure.

[0043] In the current power distribution network source-load-storage regulation, the following technical problems generally need to be overcome: (1) The randomness of photovoltaic power output, the spatiotemporal uncertainty of electric vehicle charging, and the adjustable load response characteristics form highly dispersed data sources. It is necessary to achieve cross-level (station-line-transformer-household) resource similarity aggregation while retaining key control characteristics. At the same time, the aggregation process needs to balance the contradiction between data compression rate and control accuracy. Excessive simplification will lead to distortion of the equivalent control capability of the cluster, while excessive fidelity cannot meet the data burden reduction requirements of minute-level real-time calculation. This poses an extremely high challenge to the design of dual-drive aggregation rules for digital and analog systems.

[0044] (2) In order to overcome the multi-level transmission delay of cloud, edge and terminal, it is necessary to obtain real-time data of user side through the direct control architecture of power grid. However, the safety constraints (voltage deviation, feeder current) under multiple boundary scenarios (normal / fault / extreme weather) will greatly increase the complexity of the control strategy. Although the lightweight method can reduce the computational dimension, it is necessary to solve the problem of insufficient model generalization ability - a single model needs to adapt to both fluctuating power output and high-frequency load changes, and ensure the accuracy and robustness of control commands under the condition of limited computing resources.

[0045] (3) The simplified feature space of the lightweight model may weaken the AI ​​algorithm’s ability to perceive random fluctuations in new energy sources (such as second-level output fluctuations in photovoltaics and peak-valley migration of electric vehicle charging). It is necessary to develop a reinforcement learning framework that is compatible with feature compression. At the same time, minute-level supply and demand peak regulation requires the algorithm to simultaneously complete ultra-short-term prediction, hierarchical instruction generation and multi-objective optimization (power quality improvement and feeder load balancing). The economy, safety and speed of the regulation objectives will further increase the decision-making complexity. It is urgent to build a quantifiable multi-dimensional evaluation system to verify the actual effectiveness.

[0046] Example 1: like Figure 1 As shown, this disclosure provides a unified intelligent control method for power distribution network sources, loads and storage, the method including steps S100 to S700.

[0047] S100. Based on the physical node distribution of distribution network substations, transmission lines, and distribution transformer areas, the source-load-storage resources are divided into an upper system layer, a middle line layer, and a lower transformer area layer. These three layers are arranged sequentially from top to bottom to form a three-tiered architecture. The three-tiered architecture is as follows: Figure 2 As shown.

[0048] Furthermore, in S100, the upper layer of the system is centered on the substation, covering all source, load and storage resources within the substation's radiation range. As the decision-making core for the entire network's regulation and control, it undertakes the functions of balancing the supply and demand of the entire network and responding to the superior dispatch instructions. Among them, the types of substations include centralized energy storage power stations, large-scale photovoltaic power stations, wind power plants and thermal power plants, etc.

[0049] Furthermore, in S100, the middle layer of the line takes the transmission line as a unit, covers multiple distribution substations connected by the same transmission line, aggregates substation-level distributed power sources, distributed energy storage and commercial aggregated loads connected by the same transmission line, and is mainly responsible for optimizing the allocation of resources within the region and alleviating the transmission pressure on the line.

[0050] Furthermore, in S100, the lower layer of the distribution area is based on the distribution area as a unit, covering the terminal resources within the same distribution area, including residential photovoltaic, household energy storage and adjustable loads, etc., serving as the basic unit for resource aggregation, and undertaking the function of real-time data collection and feedback of adjustment potential.

[0051] S200, based on equipment type, functional characteristics, operating characteristics and quantitative models of various equipment, divides resources at each level into a three-level classification standard including priority control level, supplementary control level and basic response level.

[0052] The following section uses photovoltaic power plants and energy storage equipment as typical examples to introduce the quantitative models of their output characteristics.

[0053] 1) Quantitative Model of Photovoltaic Power Plant Output Characteristics: Output is affected by light intensity, ambient temperature, and module efficiency. The formulas for centralized and distributed photovoltaic power plants are basically the same, differing only in rated power and the number of modules. The formula can be: (1) in, Photovoltaic power stations at all times Output power (kW); The rated output power (kW) of a photovoltaic power station is generally above 1000kW for centralized photovoltaic systems and 5-20kW for residential distributed photovoltaic systems. :time The actual light intensity (W / m²); Light intensity under standard test conditions (1000W / m²); :time Ambient temperature (°C); : Ambient temperature under standard test conditions (25℃); Temperature coefficient of photovoltaic modules (typically -0.002 to -0.005 °C), negative values ​​indicate that module efficiency decreases as temperature increases.

[0054] For residential distributed photovoltaic systems located on the lower level of the transformer substation, their output is also affected by the roof orientation and shading conditions; therefore, a shading factor can be introduced. (0≤ ≤1) is corrected, and the formula for residential distributed photovoltaic power generation can be obtained after correction: (2) in, This refers to the rated power of residential photovoltaic systems.

[0055] 2) Quantitative model of energy storage device output characteristics: Centralized energy storage (upper layer, lithium battery, flywheel energy storage) is used for grid peak shaving and backup, including charging and discharging power constraints, energy storage capacity constraints, and energy storage capacity update formulas, as shown below: Charge and discharge power constraints: (3) Energy storage capacity constraints: (4) Energy storage capacity update formula: (5) in, Centralized energy storage at any time The charging and discharging power (kW) is positive for charging and negative for discharging. Rated charging power (kW) of centralized energy storage; Rated discharge power (kW) of centralized energy storage; :time The remaining energy storage capacity (kWh); , : The maximum and minimum remaining capacity (kWh) of energy storage, generally 20%-30% of the rated capacity; Time step (h), usually 15 min or 30 min; The overall charging and discharging efficiency of centralized energy storage (typically 0.85-0.95); during charging... During discharge .

[0056] Distributed energy storage has similar output characteristics to centralized energy storage, with only slight differences in charging and discharging power and efficiency.

[0057] Furthermore, in S200, the priority control level includes centralized and distributed energy storage with a response speed of no more than 10 seconds, a regulation accuracy error of no more than 5%, and a fluctuation coefficient of no more than 0.1. The priority control level can participate in the real-time control of the entire network first, and quickly smooth out power output fluctuations.

[0058] Furthermore, in S200, the supplementary control level includes: distributed wind power plants, centralized photovoltaic power plants, commercial adjustable loads, and thermal power plants with a response speed greater than 10 seconds and not greater than 60 seconds, a regulation capacity of not less than 100kW, and a fluctuation coefficient greater than 0.1 and not greater than 0.3. The supplementary control level can participate in the control as a supplement and undertake large-capacity regulation tasks.

[0059] Furthermore, in S200, the basic response level includes: residential photovoltaic systems, adjustable residential loads, and electric vehicles with a response speed greater than 60 seconds, a regulation capacity less than 100kW, and a fluctuation coefficient greater than 0.3. The basic response level may not directly participate in the overall network control; instead, it may participate in the mid-level control of the transmission lines after being aggregated by the distribution area.

[0060] S300. The three-level hierarchical architecture is aggregated by a preset lightweight scheduling strategy to obtain a source-load-storage aggregation architecture.

[0061] Furthermore, in S300, the lightweight scheduling strategy includes sequentially performing upward resource aggregation and dimensionality reduction, intra-layer resource optimization decision-making, and downward instruction decomposition and implementation.

[0062] Furthermore, upward resource aggregation and dimensionality reduction includes: aggregating the basic response-level resources and priority control-level resources at the lower level of the distribution area into virtual energy storage units for the distribution area; aggregating multiple virtual energy storage units of the same distribution area within the same line at the middle level of the line into line aggregation units; and aggregating the line aggregation units, priority control-level resources, and supplementary control-level resources within the radiation range of each substation at the upper level of the system into a virtual control resource pool for the entire network.

[0063] Furthermore, the total output of the virtual energy storage unit in the transformer area is the sum of the net output of residential photovoltaic, distributed energy storage, electric vehicles, and adjustable residential loads within the transformer area. The equivalent capacity and equivalent charging and discharging power are obtained by summing up the capacity and power constraints of each device.

[0064] Furthermore, the total output of the line aggregation unit is the sum of the net output of the virtual energy storage units in each transformer area within the line, and it meets the maximum transmission power constraint of the line. Furthermore, the total output of the network-wide virtual control resource pool is the sum of the net output of each line aggregation unit within the substation's radiation range and the output of centralized photovoltaic power plants, thermal power plants, centralized energy storage power plants, and wind power plants, and it meets the network-wide supply and demand balance constraints.

[0065] It's important to note that upward resource aggregation and dimensionality reduction aims to aggregate low-level, scattered resources into high-level virtual control units, thereby consolidating fragmented resources and reducing computational complexity. This can be understood in the following three steps: 1) Substation Lower-Level Aggregation: This forms a virtual energy storage unit (VESS) within the substation, aggregating basic response and priority control resources to meet equipment constraints and power balance. The core formula is as follows: Total output of the virtual energy storage unit in the distribution area: (6) Equivalent capacity of the virtual energy storage unit in the transformer area: (7) The equivalent charge / discharge power of the virtual energy storage unit in the transformer area: , (8) in, For the first Each virtual energy storage unit in the transformer area is at a certain time. Net output (kW); , , , These represent the number of residential photovoltaic systems, distributed energy storage systems, electric vehicles, and adjustable residential loads within the transformer substation area. The equivalent maximum capacity (kWh) of the virtual energy storage unit in the transformer area; , The equivalent rated charging and discharging power (kW).

[0066] Meanwhile, the formula for quantifying the flexible domain parameters of the transformer area (i.e., the adjustment potential range of the virtual energy storage unit in the transformer area) can be: ; in, This represents the maximum output power of the distribution area (full photovoltaic power generation + energy storage discharge - minimum load). This represents the minimum output power of the transformer area (zero photovoltaic output + energy storage charging - maximum load).

[0067] By aggregating the data at the lower level of the distribution area, each distribution area can be fed back upwards only in the form of a "virtual energy storage unit of the distribution area", thus avoiding the data transmission and calculation of scattered devices.

[0068] 2) Line Aggregation: Forming Line Aggregation Units (LAGs) to aggregate multiple VESSs within the same line. Considering line transmission constraints, the core formula can be as follows: Total output: (9) Line transmission constraints: (10) in, For the first Line aggregation unit at time Net output (kW); This refers to the number of transformer substations within the line. This is the maximum transmission power (kW) of the line, to avoid the aggregated output exceeding the line's carrying capacity.

[0069] 3) Upper-level aggregation of the system: forming a network-wide virtual control resource pool GVCP, aggregating LAG, priority control level, and supplementary control level resources within the substation's coverage area. The core formula can be as follows: Total output of the entire network's virtual control resource pool: (11) Overall supply and demand balance constraints: (12) in, For the entire network virtual control resource pool at all times Total output (kW); The number of lines within the substation's coverage area; The power output (kW) of a centralized photovoltaic power station; The power output (kW) of a thermal power plant; The output (kW) of the centralized energy storage power station; Power output (kW) of the wind farm; The power supply capacity (kW) of the upstream power grid; This represents the total load of the entire network (kW).

[0070] Furthermore, the intra-layer resource optimization decision includes: performing intra-layer resource optimization for the upper layer of the system, the middle layer of the line, and the lower layer of the transformer area, respectively.

[0071] Furthermore, the downward instruction decomposition and implementation includes: parsing the global control instructions for the power distribution network's source, load, and storage, and decomposing the global control instructions from top to bottom to each layer of the source, load, and storage aggregation architecture and the equipment resources in each layer, thereby ensuring that the instructions can be executed accurately at the layer and even the device level.

[0072] It is important to note that the purpose of decomposing and implementing commands downwards is to break down global control commands level by level to each level and individual device, ensuring the feasibility of the commands. This can be understood in the following three steps: 1) System upper-level command decomposition: The overall network control target is decomposed to the line aggregation unit (LAG), centralized energy storage, and various power plants to meet the constraints of line transmission and regulation potential. The core formula can be shown below: (13) The constraints can be as follows: (14) (15) (16) (17) (18) The above constraints ensure that upper-level instructions are allocated reasonably, thus avoiding line overload.

[0073] 2) Line-level command decomposition: Decompose line commands to the Virtual Energy Storage Units (VESS) of each transformer area. The core formula is as follows: Control commands for the virtual energy storage unit in the distribution area: (19) Constraints: (20) in, The control commands to be undertaken by the virtual energy storage unit in the kth distribution area are to ensure that the actual output of the line aggregation unit reaches the target value.

[0074] 3) Substation Lower-Level Command Decomposition: Substation commands are decomposed to individual devices, strictly adhering to device constraints. Device constraints may include the following: Constraints of distributed energy storage: ;(twenty one) Distributed energy storage constraints are used to meet output and energy storage capacity constraints.

[0075] Constraints on residential solar power (output reduction): ;(twenty two) in, The total reduction in photovoltaic power generation in the distribution area must meet the following requirements after the reduction: ;(twenty three).

[0076] Adjustable load constraints for residents: ;(twenty four).

[0077] Electric vehicle constraints: (25) ;(26).

[0078] S400 aggregates all devices in the lower layer of the source-load-storage aggregation architecture, grouping devices with similar characteristics into the same virtual device unit. The core logic of S400 is "data fusion to extract common features + device mechanism to ensure physical feasibility". It obtains the aggregation similarity by weighted fusion of data similarity and physical compatibility constraints. Different devices with aggregation similarity exceeding a preset similarity threshold are regarded as devices with similar aggregation characteristics and are aggregated, taking into account both data transmission efficiency and control feasibility.

[0079] Furthermore, in S400, the multi-dimensional operational features include extracted time-domain features, frequency-domain features, and environmental correlation features.

[0080] Furthermore, in S400, the data similarity is calculated using the cosine similarity method, and is obtained by weighted calculation based on the multi-dimensional operating features of different device combinations.

[0081] Furthermore, in S400, the physical compatibility constraint is calculated based on the intersection range of the physical constraints of charging and discharging power, capacity, and temperature of each device.

[0082] Furthermore, in S400, the aggregated similarity is a weighted fusion value of data similarity and physical compatibility constraints.

[0083] Furthermore, the similarity threshold Depending on actual needs, the preferred preset value is 0.8.

[0084] Furthermore, in S400, the step of aggregating devices with similar characteristics into the same virtual device unit may specifically include S401 to S404.

[0085] S401: At the data fusion level, multi-dimensional operational data from various devices is collected. After data preprocessing and feature extraction, multi-dimensional operational characteristics of each device are obtained. S401 can be understood through the following steps: 1) Multi-dimensional operation data collection: The collected multi-dimensional operation data includes real-time operation data (collected every 15 minutes, including power, voltage, etc.), historical operation data (the past 12 months, used to analyze patterns), and environmental correlation data (real-time light intensity, wind speed, etc.).

[0086] 2) Data Preprocessing and Feature Extraction: First, the multi-dimensional operational data mentioned above is preprocessed, and then multi-dimensional operational features are extracted. Multi-dimensional operational features include time-domain features, frequency-domain features, and environmental correlation features, etc.

[0087] Among them, time-domain features can be further extracted through equations (27) and (28), Fourier transform can be performed through equation (29) to extract the main frequency and spectrum amplitude to obtain frequency-domain features, and environmental correlation features can be extracted through equation (30).

[0088] The formula for extracting time-domain features is shown below: , (27) , (28) in, For average output, For maximum output, Minimum output This is the fluctuation coefficient.

[0089] The formula for obtaining frequency domain features is shown below: (29) in, main frequency for The frequency corresponding to the maximum value.

[0090] The formula for extracting environmental association features is shown below: (30) in, For covariance, , Let Variance be the variance.

[0091] S402: Based on the multi-dimensional operating characteristics of each device, the data similarity between different combinations of devices is calculated.

[0092] Furthermore, the formula for calculating the data similarity between different combinations of devices using the cosine similarity method can be as follows: (31) in: , respectively equipment and equipment The A pre-processed operational characteristic; equipment and equipment To form a device combination, and ≠ ; For the first The weights of each running feature; The closer the similarity is to 1, the better the device... and equipment The more similar their operating characteristics, the more suitable they are for subsequent aggregation.

[0093] S403: By fusing data similarity and preset physical compatibility constraints, the aggregated similarity between different combinations of devices is obtained.

[0094] Furthermore, physical compatibility constraints are introduced at the device mechanism level to ensure the feasibility of subsequent aggregation. The formula for the physical compatibility constraints can be as follows: (32) in, , respectively equipment No. The maximum and minimum values ​​of each physical constraint (such as charging / discharging power, capacity, temperature, etc.); similarly, P j,k,max , P j,k,min respectively equipment j No. The maximum and minimum values ​​of each physical constraint; The number of physical constraints; The closer the compatibility score is to 1, the better the device performance. and equipment The better the physical constraints match, the better the aggregation will satisfy their respective operational constraints.

[0095] Furthermore, by fusing data similarity and preset physical compatibility constraints, the aggregated similarity between different device combinations can be obtained, which can be done with reference to the following formula: (33) in, These are the weighting coefficients; Sim total ( i | j ) represents equipment and equipment Aggregate similarity.

[0096] S404: Combine different devices with similarity exceeding a preset similarity threshold into a single virtual device unit.

[0097] Furthermore, if the equipment and equipment The aggregated similarity between them exceeded the similarity threshold. Explain the equipment and equipment The aggregation threshold has been reached, therefore the equipment... and equipment Devices are aggregated into the same virtual device unit. By iteratively calculating the aggregation similarity between pairs of devices, devices with similar characteristics are ultimately aggregated to obtain several virtual device units.

[0098] Based on the S500 and a pre-built lightweight sensing architecture and a multi-dimensional adjustable resource index system, a lightweight unified adjustment model is designed. This model addresses the problems of long delays and complex calculations in traditional control methods, achieving a closed-loop connection between "sensing-indicators-control" and adapting to minute-level control requirements.

[0099] Furthermore, in the S500, the lightweight sensing architecture is preferably a three-level flat architecture of "terminal sensing - direct transmission - edge decision-making" to solve the pain points of high data latency and large redundancy in traditional architectures.

[0100] A three-tier flat architecture can include the following components: 1) Terminal sensing layer: Deploy miniaturized, low-power sensing terminals to collect only core information for regulation (real-time photovoltaic output and regulation potential, remaining energy storage capacity and constraints, adjustable load power and delay, etc.).

[0101] 2) Direct transmission layer: adopts point-to-point direct connection mode, skips intermediate forwarding nodes, and transmits data directly to the corresponding control node after data preprocessing, reducing the transmission volume by more than 70%. Combined with data compression algorithm, the delay is controlled within 1 second.

[0102] 3) Edge decision layer: The terminal integrates edge computing nodes and has self-diagnostic functions to ensure the stability of the perception system.

[0103] Furthermore, in S500, the construction method of the multi-dimensional adjustable resource indicator system includes: using the lightweight sensing architecture to collect core control information of the source-load-storage aggregation architecture, and establishing an adjustable resource indicator system covering dimensions such as control range, control cost, and time scale. By establishing an adjustable resource indicator system with three dimensions of "control range, control cost, and time scale," and quantitatively defining it, combined with a scenario-based strategy library, the targeted and economical nature of the control is ensured.

[0104] An adjustable resource indicator system can specifically include the following dimensions and indicators: 1) Regulation range indicator: Defines the regulation boundary of the aggregation unit.

[0105] (34) in, , These are the minimum and maximum adjustable power of the aggregation units (VESS, LAG, GVCP), respectively, which are obtained by summing the output characteristic constraints of the devices within each unit.

[0106] 2) Regulation cost indicators: quantifying the cost of regulation.

[0107] (35) in, For the total cost of regulation, Let the unit adjustment cost (yuan / kWh) be the cost of equipment of type i. , These are the control command power and the initial operating power of the equipment, respectively.

[0108] 3) Time scale indicators: Distinguish between ultra-short-term (1-15 minutes, relying on priority control resources, response time ≤10s) and short-term (15 minutes-2 hours, relying on supplementary control resources, control duration ≥15 minutes) control needs.

[0109] The scenario-based strategy library can cover typical scenarios such as peak load, off-peak load, and fault recovery, ensuring that the control strategies are adapted to different operating conditions.

[0110] S600: By integrating the lightweight unified adjustment model with artificial intelligence algorithms, a hierarchical control command mechanism is established for the source-load-storage aggregation architecture. The source-load-storage aggregation architecture is then subject to unified source-load-storage control through this hierarchical control command mechanism.

[0111] Furthermore, the lightweight unified regulation model can shorten the calculation time while ensuring accuracy through "mathematical simplification + algorithm optimization," achieving unified regulation of source, load, and storage. The core process and design points are as follows: 1) Simplified model design: Mathematically, minor parameters are ignored while core variables are retained; physically, based on the equivalent characteristics of dual-drive aggregation of digital and analog models, the control of scattered equipment is transformed into the control of aggregated units, simplifying the calculation.

[0112] 2) Algorithm optimization design: A simplified particle swarm optimization algorithm (PSO) is adopted, with the objective function being "lowest regulation cost + minimum supply-demand balance error", as shown in the following formula: (36) in, , These are the weights for regulation costs and supply-demand balance errors, respectively.

[0113] 3) Model operation process: acquire real-time data and indicator parameters → call the scenario-based strategy library to generate a preliminary plan → PSO algorithm optimizes and outputs precise instructions → instructions are decomposed to individual devices → real-time feedback and adjustment to form a closed-loop control.

[0114] The core constraint of the lightweight unified regulation model is preferably the balance of supply and demand across the entire network, so as to simultaneously meet the physical constraints of equipment and aggregation units, achieve minute-level regulation, and solve the problems of insufficient accuracy and delayed response of traditional scheduling.

[0115] Furthermore, by integrating the aforementioned lightweight unified adjustment model with artificial intelligence algorithms, a collaborative mode of "lightweight computing + intelligent decision-making" is formed to enhance the adaptive control capability. The integration method can refer to the following steps: 1) Prediction algorithm fusion: A simplified LSTM algorithm can be used to combine historical and real-time data to build a 1-24 hour power demand prediction model with prediction error controlled within 5%, providing support for the early formulation of control strategies.

[0116] (37) in, Let be the predicted electricity demand at time t. Data contributing to history , , These are real-time light intensity, wind speed, and ambient temperature data, respectively.

[0117] 2) Decision algorithm fusion: Deep reinforcement learning (DRL) algorithm can be used to construct an intelligent decision model with a lightweight model as the simulation environment. The power grid state is used as the state space, the control command is used as the action space, and the multi-objective optimization is used as the reward function (Equation 38). The model can learn the optimal control strategy autonomously and adapt to complex scenarios.

[0118] (38) in To account for supply and demand imbalance error, To minimize control costs, As an indicator for evaluating power quality, , , These are the weighting coefficients.

[0119] 3) Feedback-based active learning algorithm integration: By combining the power flow distribution mechanism of the distribution network, a distributed energy aggregation flexibility solution algorithm can be designed. Through closed-loop sampling training, key data can be screened to improve computational efficiency by more than 40%, reduce computational conservatism by 50%, and balance computational efficiency and control accuracy.

[0120] Furthermore, a hierarchical control command mechanism is established for the source-load-storage aggregation architecture. This mechanism, characterized by "top-down decomposition and bottom-up feedback," ensures accurate and rapid implementation of commands and can be divided into four stages: 1) Instruction generation: The upper layer of the system combines the LSTM prediction results with the overall supply and demand balance target of the network to formulate the overall target. Combined with cost indicators and strategy library, the priority of LAG regulation is determined, and the AI ​​algorithm dynamically adjusts the target.

[0121] 2) Instruction decomposition: Instructions are decomposed in layers and levels. For example, the upper-level instructions of the system are decomposed to LAG in reference formula (39), and the middle-level instructions of the line are decomposed to VESS and the lower-level instructions of the transformer area are decomposed to individual devices in reference formula (40) to ensure that the instruction decomposition meets the constraints.

[0122] The decomposition of upper-level system instructions to LAG can be done with reference to the following formula: (39) in, This represents the maximum adjustment potential of the j-th line aggregation unit, ensuring that the instructions for each line after decomposition do not exceed its adjustment boundary.

[0123] The decomposition of line-level instructions to VESS and the decomposition of substation-level instructions to individual devices can be done using the following formulas: (40) in, For the rated capacity (energy storage, electric vehicles) or maximum adjustable capacity (adjustable load, photovoltaic) of a single device, ensure reasonable command allocation and prioritize reasonable user-side needs (such as electric vehicle travel and residential air conditioning comfort).

[0124] 3) Command execution: Relying on the direct transmission mode, the command transmission time is ≤3 minutes, the priority control level device response is ≤10s, the supplementary control level is ≤60s, the basic response level is ≤5 minutes, and the sensing terminal collects the execution status in real time.

[0125] 4) Feedback and Adjustment: Real-time feedback data is received, and AI algorithms quickly analyze deviations, re-optimize instructions, and ensure the balance of supply and demand and stable operation of the entire network.

[0126] S700. The source-load-storage unified control effect of the hierarchical and graded control instruction mechanism is verified and optimized through a multi-dimensional evaluation system.

[0127] Furthermore, the multi-dimensional evaluation system can include power quality indicators, regulation flexibility indicators, regulation accuracy indicators, regulation speed indicators, and regulation economy indicators. By using multi-dimensional evaluation indicators, the evaluation indicators can be quantitatively defined, ensuring that the evaluation results are objective and accurate.

[0128] 1) Power quality indicators: including voltage deviation Frequency deviation To quantify the stability of power grid operation; 2) Regulation flexibility index: defined as the ratio of the total regulation capacity of adjustable resources to the total load of the entire network. The higher the ratio, the greater the flexibility of regulation; 3) Control accuracy index: defined as the deviation in the execution of control commands. The smaller the deviation, the higher the accuracy; 4) Control speed index: defined as the total time from instruction generation to execution completion. The shorter the time, the faster the speed; 5) Regulate economic indicators using unit regulation costs. (RMB / kWh), the lower the cost, the better the economic efficiency.

[0129] Simulation test This simulation experiment uses the Lengbingwu distribution area in Jianshan New District, Haining City, Jiaxing City as the core demonstration area. The distribution network topology is shown in the figure below. Figure 3 As shown, strictly relying on the theoretical system of "dual-drive of digital and analog, hierarchical aggregation, lightweight perception, and artificial intelligence control", the entire process of lightweight aggregation and command decomposition scheduling of power distribution network source-load-storage resources is verified through a combination of offline simulation, semi-physical testing and field testing. The focus is on verifying the robustness of the algorithm under extreme conditions and its adaptability to actual scenarios, ultimately achieving high efficiency, precision and economy in the overall scheduling of the power distribution network.

[0130] The experiment strictly follows the hierarchical aggregation architecture of system-line-distribution area and the similarity aggregation rules of digital and analog dual drive. Taking the physical topology of the distribution network as the boundary, it focuses on the distribution area-level source-load-storage resources (distributed photovoltaic, adjustable load) to carry out closed-loop verification.

[0131] Based on a lightweight sensing architecture, it adopts a direct terminal connection transmission and core parameter acquisition mode to ensure low latency and high availability of data. Based on a multi-dimensional adjustable resource index system, it quantifies the control range, control cost, and time scale to provide boundary constraints for aggregation and scheduling. With a lightweight unified adjustment model and artificial intelligence efficient control methods as the core, it realizes rapid decision-making and dynamic optimization of instructions. It meets the constraints of equipment output, power balance, line transmission, and physical operation of energy storage / load throughout the process, ensuring that the theory is feasible.

[0132] Based on the upward aggregation and dimensionality reduction mechanism, basic response-level resources such as household photovoltaic and residential adjustable loads in the cold ice dock area are aggregated with flexible resources in the area through digital and analog dual-drive aggregation to form a virtual energy storage unit (VESS) in the area.

[0133] Using a 15-minute timescale, the system accurately predicts the overall dispatchable power range of the distribution area for the next 24 hours, simultaneously outputting the dispatchable power ranges for photovoltaic (PV) and load. The overall dispatchable power of the distribution area, the dispatchable power of PV, and the dispatchable power of the load are respectively as follows: Figure 4 , 5 As shown in Figure 6.

[0134] Following a mechanism of downward decomposition and hierarchical control, the overall dispatching targets are progressively decomposed to line aggregation units and transformer area virtual units, ultimately reaching the level of individual units / household resources. The integrated dispatching output, photovoltaic dispatching output, and load dispatching output are respectively as follows: Figure 7 , 8 As shown in Figures 9 and 9.

[0135] In this simulation experiment, the source-load-storage data perception and AI training process supporting scheduling decisions is as follows: Figure 10 As shown, the multi-dimensional evaluation system adopted Figure 11 As shown.

[0136] Through simulation and field verification, this scheme, based on the dual-drive hierarchical aggregation theory of digital and analog models, successfully achieves lightweight aggregation and decomposition of power distribution network source-load-storage resources and closed-loop execution of scheduling commands. The experiment fully verifies the entire technical route of "lightweight aggregation – hierarchical decision-making – command decomposition – closed-loop feedback", forming a replicable and scalable engineering implementation standard, providing reliable support for the coordinated regulation of source-load-storage under high-proportion renewable energy access in the power distribution network.

[0137] Example 2: like Figure 12 As shown, in a second aspect, this disclosure provides a unified intelligent control system for power distribution network sources, loads, and energy storage, the system comprising: The three-level hierarchical architecture module 100 is used to divide the source-load-storage resources into the upper system layer, the middle line layer, and the lower distribution area layer according to the physical node distribution of the distribution network substations, transmission lines, and distribution areas. The upper system layer, the middle line layer, and the lower distribution area layer are arranged from top to bottom to form a three-level hierarchical architecture. The three-level classification standard module 200 is used to divide resources at each level into three-level classification standards, including priority control level, supplementary control level and basic response level, based on equipment type, functional characteristics, operating characteristics and quantitative models of various equipment. The source-load-storage aggregation architecture module 300 is used to perform hierarchical aggregation of the three-level hierarchical architecture through a preset lightweight scheduling strategy to obtain the source-load-storage aggregation architecture. The aggregation module 400 within the distribution area is used to aggregate all devices in the lower layer of the source-load-storage aggregation architecture, and to aggregate devices with similar characteristics into the same virtual device unit. The adjustment model module 500 is used to design a lightweight unified adjustment model based on a pre-built lightweight perception architecture and a multi-dimensional adjustable resource index system. The unified control module 600 is used to integrate the lightweight unified control model with artificial intelligence algorithms to establish a hierarchical control command mechanism for the source-load-storage aggregation architecture, and to perform unified control of the source-load-storage aggregation architecture through the hierarchical control command mechanism. The verification and optimization module 700 is used to verify the source-load-storage unified control effect of the hierarchical and graded control instruction mechanism through a multi-dimensional evaluation system.

[0138] It is worth noting that the system described in Example 2 is only one system implementation of the unified intelligent control method for power generation, load and storage of distribution network described in Example 1, and does not limit the unified intelligent control method for power generation, load and storage of distribution network described in Example 1 to depend on the system described in Example 2.

[0139] Example 3: Embodiment 3 of this disclosure provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the unified intelligent control method for power generation, load and storage of the distribution network as described in Embodiment 1.

[0140] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0141] Example 4: Embodiment 4 of this disclosure provides an electronic device, which includes a processor and a memory. The processor is used to execute a computer program stored in the memory to implement the unified intelligent control method for power generation, load and storage of the distribution network described in Embodiment 1.

[0142] Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0143] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0145] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0146] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0147] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0148] In summary, the unified intelligent control method and system for power distribution network sources, loads, and storage provided in embodiments 1-4 of this disclosure, based on a hierarchical aggregation architecture driven by both digital and analog models, achieves scientific classification and efficient aggregation of source, load, and storage resources from physical topology to functional characteristics through a three-level hierarchical architecture of "system-line-transformer area" and a three-level hierarchical structure of "priority control level, supplementary control level, and basic response level." The upward aggregation and dimensionality reduction mechanism transforms massive amounts of scattered equipment into high-level virtual control units, significantly reducing the computational complexity of system-level scheduling and solving the problems of data explosion and computational bottlenecks in traditional scheduling. The lightweight sensing architecture constructed in this disclosure adopts a three-level flat design of "terminal sensing-direct transmission-edge decision-making," skipping intermediate forwarding nodes in traditional cloud-edge-end architectures, reducing transmission volume by more than 70%, and controlling latency to within 1 second. Combined with the application of a simplified particle swarm optimization algorithm in the lightweight unified control model, it achieves rapid control decisions at the minute or even second level, significantly outperforming the hourly response cycle of traditional centralized scheduling.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit them. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this disclosure. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this disclosure should be covered within the protection scope of the claims of this disclosure.

Claims

1. A unified intelligent control method for power distribution network sources, loads, and storage, characterized in that, The method includes: S100. According to the physical node distribution of distribution network substations, transmission lines and distribution transformer areas, the source-load-storage resources are divided into the upper system layer, the middle line layer and the lower transformer area layer. The upper system layer, the middle line layer and the lower transformer area layer are arranged from top to bottom to form a three-level hierarchical architecture. S200, based on equipment type, functional characteristics, operating characteristics and quantitative models of various equipment, divides resources at each level into a three-level classification standard including priority control level, supplementary control level and basic response level; S300. The three-level hierarchical architecture is aggregated hierarchically by a preset lightweight scheduling strategy to obtain a source-load-storage aggregation architecture. S400: Aggregate all devices in the lower layer of the source-load-storage aggregation architecture, and aggregate devices with similar characteristics into the same virtual device unit. S500, based on a pre-built lightweight perception architecture and a multi-dimensional adjustable resource index system, designed a lightweight unified adjustment model; S600: By integrating the lightweight unified adjustment model with artificial intelligence algorithms, a hierarchical control command mechanism is established for the source-load-storage aggregation architecture. The source-load-storage aggregation architecture is then subject to unified source-load-storage control through this hierarchical control command mechanism.

2. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 1, characterized in that, In S100, The upper layer of the system is centered on substations, covering all source, load and storage resources within the substation's radiation range. As the decision-making core for the entire network's regulation and control, it undertakes the functions of balancing the supply and demand of the entire network and responding to the dispatch instructions from higher authorities. Among them, the types of substations include centralized energy storage power stations, large-scale photovoltaic power stations, wind power plants and thermal power plants. The middle layer of the line takes the transmission line as the unit, covers multiple distribution substations connected by the same transmission line, aggregates substation-level distributed power sources, distributed energy storage and commercial aggregated loads connected by the same transmission line, and is mainly responsible for optimizing the allocation of resources within the region and alleviating the transmission pressure of the line. The lower layer of the distribution area is based on the distribution area as a unit, covering the terminal resources within the same distribution area, including residential photovoltaic, household energy storage and adjustable loads. As the basic unit for resource aggregation, it undertakes the function of real-time data collection and feedback of adjustment potential.

3. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 1, characterized in that, In S200, The priority control level includes centralized energy storage and distributed energy storage with a response speed of no more than 10 seconds, a regulation accuracy error of no more than 5%, and a fluctuation coefficient of no more than 0.

1. The supplementary control level includes: distributed wind power plants, centralized photovoltaic power plants, commercial adjustable loads, and thermal power plants with a response speed greater than 10 seconds and not greater than 60 seconds, a regulation capacity of not less than 100kW, and a fluctuation coefficient greater than 0.1 and not greater than 0.

3. The basic response level includes: residential photovoltaic systems, adjustable residential loads, and electric vehicles with a response speed greater than 60 seconds, a regulation capacity of less than 100kW, and a fluctuation coefficient greater than 0.

3.

4. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 1, characterized in that, In S300, the lightweight scheduling strategy includes sequentially performing upward resource aggregation and dimensionality reduction, intra-layer resource optimization decision-making, and downward instruction decomposition and implementation: wherein, Upward resource aggregation and dimensionality reduction include: aggregating the basic response-level resources and priority control-level resources at the lower layer of the distribution area into virtual energy storage units for the distribution area; aggregating multiple virtual energy storage units of the same distribution area within the same line at the middle layer of the line into line aggregation units; and aggregating the line aggregation units, priority control-level resources, and supplementary control-level resources within the radiation range of each substation at the upper layer of the system into a virtual control resource pool for the entire network. Intra-layer resource optimization decisions include: performing intra-layer resource optimization for the upper layer of the system, the middle layer of the line, and the lower layer of the transformer area, respectively; The downward instruction decomposition and implementation includes: parsing the global control instructions for the power distribution network's source, load, and storage, and decomposing the global control instructions from top to bottom to each layer and device resource in the source, load, and storage aggregation architecture, thereby ensuring that the instructions can be executed accurately at the layer and even the device level.

5. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 4, characterized in that, In S300, The total output of the virtual energy storage unit in the transformer area is the sum of the net output of household photovoltaic, distributed energy storage, electric vehicles and adjustable loads in the transformer area. The equivalent capacity and equivalent charging and discharging power are obtained by summing up the capacity and power constraints of each device. The total output of the line aggregation unit is the sum of the net output of the virtual energy storage units in each transformer area within the line, and it meets the maximum transmission power constraint of the line. The total output of the network-wide virtual control resource pool is the sum of the net output of each line aggregation unit within the substation's radiation range and the output of centralized photovoltaic power stations, thermal power plants, centralized energy storage power stations, and wind power plants, and it meets the network-wide supply and demand balance constraints.

6. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 1, characterized in that, In step S400, the step of aggregating devices with similar characteristics into the same virtual device unit specifically includes: S401: At the data fusion level, multi-dimensional operational data of each device is collected, and after data preprocessing and feature extraction, multi-dimensional operational characteristics of each device are obtained. S402: Based on the multi-dimensional operational characteristics of each device, the data similarity between different combinations of devices is calculated. S403: By fusing data similarity and preset physical compatibility constraints, the aggregated similarity between different combinations of devices is obtained; S404: Aggregate different devices with similarity exceeding a preset similarity threshold into the same virtual device unit.

7. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 6, characterized in that, In S400, The multi-dimensional operational features include extracted time-domain features, frequency-domain features, and environmental correlation features; The data similarity was calculated using the cosine similarity method, and was obtained by weighted calculation based on the multi-dimensional operating characteristics of different device combinations. The physical compatibility constraints are calculated based on the intersection range of the physical constraints of charging and discharging power, capacity, and temperature of each device. The aggregated similarity is a weighted fusion value of data similarity and physical compatibility constraints.

8. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 1, characterized in that, In the S500, The lightweight sensing architecture is preferably a three-level flat architecture of "terminal sensing - direct transmission - edge decision" to solve the pain points of high data latency and large redundancy in traditional architectures; The construction method of the multidimensional adjustable resource indicator system includes: using the lightweight sensing architecture to collect the core control information of the source-load-storage aggregation architecture, and establishing an adjustable resource indicator system covering the dimensions of control range, control cost and time scale.

9. The unified intelligent control method for power distribution network sources, loads, and storage as described in claim 1, characterized in that, The method further includes: S700. The source-load-storage unified control effect of the hierarchical and graded control instruction mechanism is verified and optimized through a multi-dimensional evaluation system.

10. A unified intelligent control system for power distribution network sources, loads, and storage, characterized in that, The system includes: The three-level hierarchical architecture module (100) is used to divide the source, load and storage resources into the upper system layer, the middle line layer and the lower distribution area layer according to the physical node distribution of the distribution network substations, transmission lines and distribution areas. The upper system layer, the middle line layer and the lower distribution area layer are arranged from top to bottom to form a three-level hierarchical architecture. The three-level classification standard module (200) is used to classify resources at each level into three levels, including priority control level, supplementary control level and basic response level, based on equipment type, functional characteristics, operating characteristics and quantitative models of various equipment. The source-load-storage aggregation architecture module (300) is used to perform hierarchical aggregation of the three-level hierarchical architecture through a preset lightweight scheduling strategy to obtain the source-load-storage aggregation architecture. The aggregation module (400) within the distribution area is used to aggregate all devices in the lower layer of the source-load-storage aggregation architecture, and to aggregate devices with similar characteristics into the same virtual device unit. The adjustment model module (500) is used to design a lightweight unified adjustment model based on a pre-built lightweight perception architecture and a multi-dimensional adjustable resource index system. The unified control module (600) is used to integrate the lightweight unified control model and artificial intelligence algorithm to establish a hierarchical control command mechanism for the source-load-storage aggregation architecture, and to perform unified control of the source-load-storage aggregation architecture through the hierarchical control command mechanism. The verification and optimization module (700) is used to verify the source-load-storage unified control effect of the hierarchical and graded control instruction mechanism through a multi-dimensional evaluation system.

Images