Power distribution network capacity-increase-free access system based on dynamic coordination

Abstract

The invention relates to the technical field of power distribution network operation management, and discloses a power distribution network capacity-increase-free access system based on dynamic coordination. The system comprises a multi-source data acquisition and fusion unit which acquires multi-source data and fuses the multi-source data to generate dynamic power grid state characteristics; the multi-time-scale collaborative optimization architecture unit is used for constructing a layered architecture to realize optimization of different time scales; the space-time load prediction unit is used for predicting load distribution based on the space-time diagram model; and the multi-target dynamic optimization solving unit is used for constructing a model to solve an optimal scheduling instruction. In addition, a layered communication protocol unit is arranged to guarantee communication, and a voltage stability cooperative control unit and a distributed energy storage cooperative control unit are arranged to improve the stability of the power grid. Through cooperation of multiple technologies, capacity-increase-free access of the power distribution network is realized, the operation efficiency is improved, the loss is reduced, the voltage stability is ensured, and challenges brought by distributed energy and electric vehicle charging are effectively dealt with.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network operation management, and in particular to a distribution network capacity-increase-free access system based on dynamic collaboration. Background Art

[0002] With the rapid development of distributed energy, such as the widespread access of distributed photovoltaics to distribution networks, and the continued growth in the number of electric vehicles, their charging demand has brought tremendous pressure to the distribution network, and the operation and management of the distribution network faces many challenges.

[0003] From the perspective of distributed photovoltaics, its output is obviously intermittent and volatile. Affected by weather factors, such as cloud cover and changes in light intensity, photovoltaic output is difficult to accurately predict. When the rapid movement of clouds causes a sudden change in light intensity, photovoltaic output may fluctuate greatly in a short period of time, making it difficult to maintain the power balance of the distribution network. Traditional distribution network planning and scheduling methods are difficult to adapt to this unstable power access. If they cannot be effectively dealt with, it is easy to cause problems such as local grid voltage fluctuations and power quality degradation, and may even affect the safe and stable operation of the grid. Moreover, distributed photovoltaics are usually dispersedly connected to various nodes of the distribution network. The uncertainty of their access location and capacity increases the complexity of grid flow calculation and optimal scheduling.

[0004] In terms of electric vehicle charging demand, its temporal and spatial distribution is extremely uneven. During the morning and evening peak hours on weekdays, the demand for charging piles near urban commercial areas and residential areas is concentrated, while the demand drops sharply during other periods. This concentrated charging behavior will cause a sharp increase in the load of the local distribution network, exceeding the rated capacity of the lines and transformers, accelerating equipment aging, shortening equipment life, and increasing the operation and maintenance costs of the power grid. There are also differences in the charging demand for electric vehicles in different regions. The charging demand characteristics between the city center and the suburbs and between different functional areas are different, which brings great difficulties to the planning and load forecasting of the distribution network.

[0005] The historical operation data of the distribution network shows that the traditional method based on fixed capacity planning and static scheduling can no longer meet the growing and complex electricity demand. In some areas where distributed photovoltaic and electric vehicle charging demand is concentrated, problems such as voltage over-limit and line overload frequently occur. To solve these problems, the method of capacity expansion and transformation was often used in the past, that is, increasing the capacity of substations and replacing transmission lines with larger cross-sections. However, this method requires huge investment and a long construction period. It not only requires a large amount of capital investment, but also causes many inconveniences to urban construction and residents' lives, and may not fundamentally solve the randomness and volatility problems brought about by distributed energy and electric vehicle charging.

[0006] Traditional distribution network capacity expansion solutions respond to load growth by replacing large-capacity transformers or building additional lines. However, this method has poor economy and insufficient flexibility, and cannot effectively cope with the fluctuations of distributed photovoltaics and the random charging loads of electric vehicles, resulting in repeated capacity expansion. Centralized scheduling strategies rely on the unified issuance of instructions by township-level scheduling centers, but there are problems of response lag and computational bottlenecks, unable to adjust the power of photovoltaic inverters and the charging speed of electric vehicles in real time, and unable to support minute-level scheduling. In the single-level regulation method, the village level only relies on the charge and discharge of local energy storage, and the user side independently responds to time-of-use electricity prices, which leads to resource waste, low energy storage utilization rate, high photovoltaic curtailment rate, and serious voltage over-limit problems.

[0007] To address these problems, the system proposed in the present invention realizes the capacity-free access of the distribution network by establishing a dynamic coordination mechanism among the three levels of townships, villages, and users. This system can improve the operation efficiency and stability of the power grid, reduce the dependence on physical capacity expansion, lower costs, and enhance the adaptability to distributed energy and random loads. Through this multi-level coordination mechanism, more accurate load forecasting, faster response, and more efficient resource utilization can be achieved, providing strong support for the sustainable development of the distribution network. Summary of the Invention

[0008] The purpose of the present invention is to provide a capacity-free access system for a distribution network based on dynamic coordination to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solution: A capacity-free access system for a distribution network based on dynamic coordination, the system includes:

[0010] A multi-source data acquisition and fusion unit, which is used to obtain the real-time operation data of the distribution network through a multi-source data acquisition module. The multi-source data includes historical load data, meteorological data, distributed photovoltaic output data, electric vehicle charging demand data, and grid topology data; based on a graph neural network, topological feature extraction and fusion are performed on the real-time operation data to generate dynamic grid state features;

[0011] A multi-time scale collaborative optimization architecture unit, which is used to construct a multi-time scale collaborative optimization architecture. The architecture includes a township-level global optimization layer, a village-level real-time regulation layer, and a user terminal response layer; the township-level global optimization layer adopts a distributed robust optimization model to generate a day-ahead benchmark scheduling plan based on the dynamic grid state features; the village-level real-time regulation layer adopts an adaptive model predictive control algorithm to perform intraday rolling correction based on the scheduling plan; the user terminal response layer adopts a game theory-driven load allocation strategy to achieve second-level dynamic power regulation;

[0012] A spatio-temporal load forecasting unit for establishing a load forecasting model based on a spatio-temporal graph model. The load forecasting model fuses travel chain characteristics, photovoltaic output volatility, and spatio-temporal distribution data of electric vehicle charging, and uses a spatio-temporal graph convolutional network to predict the future multi-period load distribution and output the load forecasting result.

[0013] A multi-objective dynamic optimization solving unit for constructing a multi-objective dynamic optimization model. The model takes the minimum grid loss, the lowest energy storage life loss, and the maximum new energy consumption as optimization objectives, and uses the alternating direction method of multipliers for distributed solution to generate an optimal scheduling instruction under the global power balance constraint.

[0014] Preferably, the township-level global optimization layer adopts a distributed robust optimization model to generate a day-ahead benchmark scheduling plan based on the dynamic grid state characteristics, including:

[0015] Construct an uncertainty set, which includes the photovoltaic output fluctuation range, load forecasting error, and electric vehicle charging demand deviation;

[0016] Establish a robust objective function, which robustly optimizes the distributed power generation output and energy storage charge and discharge strategies by minimizing the grid operation cost in the worst-case scenario and combining linear decision rules;

[0017] Use a distributed iterative algorithm to solve the objective function, and achieve the decomposition and coordination of global constraints through information interaction between adjacent nodes;

[0018] The robust objective function is:

[0019]

[0020] where u is the decision variable vector, including the distributed power generation output and energy storage charge and discharge power, w is the uncertainty parameter vector, including the photovoltaic output P pv and the load demand P load , Ω is the uncertainty set, c g is the unit power generation cost coefficient, c ess is the energy storage cycle loss coefficient, γ is the regularization coefficient, and T is the optimization time window length.

[0021] Preferably, the village-level real-time regulation layer adopts an adaptive model predictive control algorithm to perform intraday rolling correction based on the scheduling plan, including:

[0022] Construct a dynamic state space equation, which includes the time-varying relationships of node voltage, energy storage state of charge, and line power;

[0023] Design a rolling optimization objective function, which combines a weighted tracking error term and a control variable rate of change term, and updates the model parameters in real time by integrating an online parameter identification technique; introduce a constraint softening mechanism, and when a hard constraint conflict is detected, ensure the feasibility of the optimization problem by relaxing the boundary conditions and adding penalty terms;

[0024] The dynamic state space equation is as follows:

[0025] x k+1 = A k x k + B k u k + E k d k

[0026] where x k+1 is the state vector at time step k + 1, x k is the state vector at time step k, u k is the control vector at time step k, including the reactive power output of the PV inverter, d k is the disturbance vector at time step k, including load fluctuations, A k , B k , E k are time-varying system matrices.

[0027] Preferably, the user terminal response layer adopts a game theory-driven load distribution strategy to achieve second-level dynamic power regulation, including:

[0028] Define a user utility function, which combines the electricity price incentive coefficient, the charging delay cost, and the power regulation revenue to construct a non-cooperative game model;

[0029] Design a Nash equilibrium solution algorithm, and use the distributed gradient projection method to iteratively calculate the optimal response strategies of each user;

[0030] The user utility function is as follows:

[0031]

[0032] where U i (p i ) is the user utility function, p i is the actual regulation power of user i, is the demand power, α i is the electricity price sensitivity coefficient, β i is the charging delay penalty coefficient, λ i is the regulation cost coefficient.

[0033] Preferably, the load forecasting model based on the spatio-temporal graph model includes:

[0034] Construct a dynamic graph structure, where the nodes in the graph represent the topological nodes of the distribution network, and the edge weights are jointly determined by the electrical distance and the real-time power flow;

[0035] Design a spatio-temporal convolutional layer, and extract the spatio-temporal correlation features of the load distribution through the alternating operations of one-dimensional convolution in the time dimension and graph convolution in the space dimension;

[0036] Adopt an attention mechanism to dynamically adjust the feature weights of different time steps and spatial nodes, and improve the adaptability of the prediction model to emergencies;

[0037] The spatio-temporal convolution operation is as follows:

[0038] H (l+1) =σ(Φ temp *(Θ spa *H (l) ))

[0039] where H (l+1) is the feature matrix of the (l + 1)-th layer, H (l) is the feature matrix of the l-th layer, Φ temp is the time convolution kernel, Θ spa is the spatial graph convolution parameter, * represents the one-dimensional convolution operation, ★ represents the graph convolution operation, and σ is the activation function.

[0040] Preferably, the alternating direction multiplier method is used for distributed solution to generate an optimal scheduling instruction under the global power balance constraint, including:

[0041] Decompose the global optimization problem into a township-level main problem and a village-level sub-problem, and realize variable coupling by introducing a consistency constraint;

[0042] Design a residual feedback mechanism to dynamically adjust the penalty factor according to the solution deviation of the village-level sub-problem to accelerate the algorithm convergence; adopt a parallel computing architecture and use GPU to accelerate matrix operations to realize the real-time solution of large-scale distribution network optimization problems.

[0043] Preferably, the system further includes:

[0044] A hierarchical communication protocol unit for designing a hierarchical communication protocol, which combines 5G network slicing and edge computing technologies to realize the millisecond-level issuance of township-level optimization instructions, the real-time synchronization of village-level control data, and the fast feedback of user terminal status information.

[0045] Preferably, the hierarchical communication protocol includes:

[0046] Define the data encapsulation format of the township-level optimization instruction, and the format includes a timestamp, a target power value, and a priority identifier;

[0047] Design a compression and transmission protocol for village-level control data, and use sparse matrix coding and differential coding techniques to reduce the amount of data transmitted;

[0048] Construct an edge preprocessing mechanism for user terminal status information, and reduce the redundancy of uplink data through local filtering and feature extraction;

[0049] The compression and transmission protocol satisfies:

[0050]

[0051] Among them, S is the original data matrix, is the compression reconstruction matrix, R is the relative error rate, and ∈ is the preset error threshold.

[0052] Preferably, the system further includes:

[0053] A voltage stability coordinated control unit, used to construct a voltage stability evaluation module, which calculates the power grid voltage stability margin in real time based on the Lyapunov exponent analysis method and generates a preventive control strategy;

[0054] A distributed energy storage coordinated control unit, used to design a distributed energy storage coordinated control strategy, which coordinates the charging and discharging timing of multi-node energy storage through a consensus algorithm to avoid power oscillation and overload risks; The voltage stability margin calculation formula is:

[0055]

[0056] Among them, η is the system stability margin index, P i is the real-time power of node i, P i,max is the power limit of node i, is the node set.

[0057] Preferably, the voltage stability evaluation module includes:

[0058] Construct a reduced-order power grid dynamic model, which extracts the dominant oscillation mode through modal analysis; Design an online calculator for stability indicators, and iteratively update the coefficient matrix of the Lyapunov function based on real-time measurement data;

[0059] Generate adaptive PID control parameters, and dynamically adjust the response intensity of the reactive power compensation device according to the change rate of the stability margin; The Lyapunov function is:

[0060]

[0061] Among them, Δδ is the generator power angle deviation vector, M is the diagonal matrix of inertia coefficients, and V is the energy function.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] The multi-source data acquisition and fusion unit of the present invention can widely collect various types of data such as historical load data, meteorological data, distributed photovoltaic output data, electric vehicle charging demand data, and power grid topology data. Through graph neural networks, topological feature extraction and fusion are carried out, and the generated dynamic power grid state features are comprehensive and accurate, providing a solid data foundation for subsequent optimization decisions. Compared with traditional data acquisition and analysis methods, this multi-source data fusion technology can dig deeper into the potential connections between data. For example, it can combine meteorological data and historical load data to more accurately predict load fluctuations caused by weather changes, greatly improving the perception ability of the power grid operation state and laying a foundation for achieving precise scheduling and control.

[0064] The township-level global optimization layer, village-level real-time regulation layer, and user terminal response layer constructed by the multi-time scale collaborative optimization architecture unit achieve collaborative optimization at different time scales. The township-level global optimization layer uses a distributed robust optimization model to generate a daily benchmark scheduling plan, considering uncertainty factors such as photovoltaic output fluctuations, load prediction errors, and electric vehicle charging demand deviations, improving the robustness of the scheduling plan. The village-level real-time regulation layer performs intraday rolling correction based on an adaptive model predictive control algorithm and can respond in a timely manner to real-time changes in the power grid, such as sudden changes in load or abnormal fluctuations in distributed power generation output. The user terminal response layer uses a game theory-driven load distribution strategy to achieve second-level dynamic power regulation, motivating users to participate in power grid regulation and improving the flexibility and stability of the power grid. This hierarchical collaborative optimization method can more effectively balance the power grid operation cost, improve power supply reliability, and adapt to the dynamic characteristics of distributed energy and electric vehicle charging compared with a single-time scale scheduling method.

[0065] The load forecasting model based on a spatio-temporal graph model established by the spatio-temporal load forecasting unit integrates travel chain characteristics, photovoltaic output volatility, and spatio-temporal distribution data of electric vehicle charging, and uses a spatio-temporal graph convolutional network to predict the future multi-period load distribution. This model fully considers the correlation of load in time and space. For example, by analyzing the spatio-temporal distribution laws of residents' travel patterns and electric vehicle charging behaviors, it can more accurately predict the load demands in different regions at different times. Compared with traditional load forecasting models, it can capture load change trends more timely and accurately, providing a more reliable basis for the scheduling and planning of the power grid and reducing power grid operation problems caused by inaccurate load forecasting, such as line overload and voltage instability.

[0066] A multi-objective dynamic optimization model for the construction of a multi-objective dynamic optimization solution unit, with the minimum grid loss, the lowest energy storage life loss, and the maximum new energy consumption as the optimization objectives, is solved distributively using the alternating direction multiplier method to generate optimal scheduling instructions under the global power balance constraint. In this way, while ensuring the safe and stable operation of the power grid, various performance indicators of the power grid can be comprehensively optimized. For example, while meeting the load demand, new energy such as distributed photovoltaics can be maximally consumed to reduce the phenomenon of light curtailment; the charge and discharge strategies of energy storage can be reasonably arranged to reduce the energy storage life loss and extend the service life of energy storage equipment; the power flow distribution of the power grid can be optimized to reduce the grid loss and improve the operation efficiency of the power grid, thereby realizing the economic and efficient operation of the distribution network.

[0067] The hierarchical communication protocol designed by the hierarchical communication protocol unit, combined with 5G network slicing and edge computing technologies, realizes the millisecond-level issuance of optimization instructions at the township level, the real-time synchronization of village-level control data, and the rapid feedback of user terminal status information. By defining a reasonable data encapsulation format, designing an efficient data compression and transmission protocol, and constructing an edge preprocessing mechanism, the communication efficiency is improved, the data transmission delay and redundancy are reduced, and the accurate and timely transmission of control instructions and data at all levels is ensured, providing a strong communication guarantee for the realization of real-time control and optimized scheduling of the distribution network.

[0068] The voltage stability assessment module constructed by the voltage stability collaborative control unit calculates the power grid voltage stability margin in real time based on the Lyapunov exponent analysis method and generates preventive control strategies. This helps to timely detect potential unstable factors in the power grid voltage. By adjusting reactive power compensation devices and other means, the stability of the power grid voltage is maintained, and the anti-interference ability of the power grid is improved. The distributed energy storage collaborative control strategy designed by the distributed energy storage collaborative control unit coordinates the charge and discharge timing of multi-node energy storage through a consensus algorithm, avoiding the risks of power oscillation and overload, and giving full play to the role of distributed energy storage in suppressing power fluctuations and improving power quality, further enhancing the stability and reliability of the distribution network. Brief Description of the Drawings

[0069] Figure 1 It is the working principle diagram of the distribution network free-capacity access system based on dynamic collaboration described in the present invention;

[0070] Figure 2 It is the working principle diagram of the township-level global optimization layer;

[0071] Figure 3 It is the working flow chart of the load forecasting model based on the spatio-temporal graph model;

[0072] Figure 4 It is the working flow chart of data transmission and interaction of the hierarchical communication protocol unit. Detailed Implementation Manner

[0073] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0074] Please refer to Figures 1-4 , the present invention provides a technical solution: a distribution network capacity-free access system based on dynamic collaboration. Through the collaborative work of a multi-source data acquisition and fusion unit, a multi-time-scale collaborative optimization architecture unit, a spatio-temporal load forecasting unit, and a multi-objective dynamic optimization solving unit, the system realizes the efficient management and optimization of capacity-free access to the distribution network.

[0075] Multi-source data acquisition and fusion unit: The multi-source data acquisition module is used to obtain the real-time operation data of the distribution network, including historical load data, meteorological data, distributed photovoltaic output data, electric vehicle charging demand data, and grid topology data. Then, based on the graph neural network, topological feature extraction and fusion are performed on these real-time operation data to generate dynamic grid state features. The graph neural network can effectively process data with topological structures. By extracting and fusing the features of different types of data, it provides an accurate and comprehensive information basis for subsequent optimization and decision-making.

[0076] Multi-time-scale collaborative optimization architecture unit: A multi-time-scale collaborative optimization architecture including a township-level global optimization layer, a village-level real-time regulation layer, and a user terminal response layer is constructed. The township-level global optimization layer adopts a distributed robust optimization model to generate a daily benchmark scheduling plan based on the dynamic grid state features; the village-level real-time regulation layer uses an adaptive model predictive control algorithm to perform intraday rolling correction according to the scheduling plan; the user terminal response layer realizes second-level dynamic power regulation by means of a game theory-driven load distribution strategy. This hierarchical architecture can perform refined management of the distribution network according to the requirements and characteristics of different time scales, improving the stability and reliability of the system.

[0077] Spatio-temporal load forecasting unit: A load forecasting model based on a spatio-temporal graph model is established. This model integrates travel chain features, photovoltaic output volatility, and spatio-temporal distribution data of electric vehicle charging. Using a spatio-temporal graph convolutional network, it predicts the future multi-period load distribution and outputs the load forecasting results. The spatio-temporal graph model can fully consider the correlation of load in time and space, improve the accuracy of load forecasting, and provide a reliable basis for the scheduling and optimization of the distribution network.

[0078] Multi-objective dynamic optimization solution unit: Construct a multi-objective dynamic optimization model with the objectives of minimizing power grid losses, minimizing energy storage life losses, and maximizing new energy consumption. The alternating direction multiplier method is used for distributed solution to generate optimal scheduling instructions under the global power balance constraint. The multi-objective dynamic optimization model can comprehensively consider the trade-off relationships among multiple optimization objectives, improve the calculation efficiency through the distributed solution method, and achieve the optimal operation of the distribution network.

[0079] The following further illustrates the present invention with specific examples using Embodiments 1 to 6:

[0080] Embodiment 1:

[0081] This embodiment details how the township-level global optimization layer uses a distributed robust optimization model to generate a day-ahead benchmark scheduling plan to cope with the uncertainties in the distribution network and ensure that the power grid operation cost can be effectively controlled even in the worst scenarios.

[0082] The township-level global optimization layer uses a distributed robust optimization model to generate a day-ahead benchmark scheduling plan. First, an uncertainty set is constructed, which covers the fluctuation range of photovoltaic power output, load forecasting error, and electric vehicle charging demand deviation. These uncertainty factors are common in the operation of the distribution network and have a significant impact on the formulation of the scheduling plan. By constructing the uncertainty set, these uncertain factors can be taken into account to improve the robustness of the scheduling plan.

[0083] A robust objective function is established, and the formula is:

[0084]

[0085] Among them, u is the decision variable vector, including the output of distributed power sources and the charge and discharge power of energy storage. The reasonable adjustment of these variables can optimize the operation of the power grid. w is the uncertainty parameter vector, including the photovoltaic power output P pv , the load demand P load , and the uncertainties of these parameters pose challenges to the scheduling plan. Ω is the uncertainty set, c g is the unit power generation cost coefficient, reflecting the relationship between power generation cost and power generation power. c ess is the energy storage cycle loss coefficient, reflecting the loss during the charge and discharge process of the energy storage. γ is the regularization coefficient, used to balance the weights of different terms. T is the length of the optimization time window, determining the time range of the scheduling plan. This function robustly optimizes the output of distributed power sources and the charge and discharge strategy of energy storage through the linear decision rule by minimizing the power grid operation cost in the worst scenario, making the scheduling plan more reliable in the face of uncertainties.

[0086] The distributed iterative algorithm is used to solve the objective function, and the decomposition and coordination of global constraints are realized through information interaction between adjacent nodes. In an actual distribution network, there are numerous and interconnected nodes. The distributed iterative algorithm can make full use of the information interaction between nodes, decompose complex global constraints into multiple local constraints for processing, and improve the solution efficiency and system scalability.

[0087] Embodiment 2:

[0088] This embodiment is used to describe the village-level real-time regulation layer. By adopting the adaptive model predictive control algorithm for intraday rolling correction, the response ability to the real-time changes of the distribution network is improved, and the stable operation of the power grid is ensured.

[0089] The village-level real-time regulation layer adopts the adaptive model predictive control algorithm to perform intraday rolling correction based on the scheduling plan. First, a dynamic state-space equation is constructed, and the formula is:

[0090] x k+1 =A k x k +B k u k +E k d k

[0091] Among them, x k+1 is the state vector at time step k + 1, which includes information such as node voltage and energy storage state of charge, and reflects the operating state of the distribution network at this moment. x k is the state vector at time step k, u k is the control vector at time step k, which includes control variables such as the reactive power output of the photovoltaic inverter. By adjusting these variables, the operation of the power grid can be regulated. d k is the disturbance vector at time step k, including factors such as load fluctuations. These disturbances will affect the stable operation of the power grid. A k , B k , E k are time-varying system matrices, which describe the dynamic relationships between the state vector, the control vector, and the disturbance vector.

[0092] Design a rolling optimization objective function. By weighting the tracking error term and the control variable rate of change term, and combining the online parameter identification technology to update the model parameters in real time. When a hard constraint conflict is detected, a constraint softening mechanism is introduced. By relaxing the boundary conditions and adding penalty terms, the feasibility of the optimization problem is ensured. In actual operation, the parameters of the distribution network will change with time and environment. The online parameter identification technology can obtain these changes in real time to make the model more accurate. The constraint softening mechanism can, in the face of a hard constraint conflict, ensure that the optimization problem can continue to be solved through reasonable adjustment and maintain the stable operation of the power grid.

[0093] Example 3:

[0094] This example illustrates how the user terminal response layer uses a game theory-driven load distribution strategy to achieve second-level dynamic power regulation, motivating users to reasonably adjust their electricity consumption behaviors and improving the flexibility and stability of the distribution network.

[0095] The user terminal response layer uses a game theory-driven load distribution strategy to achieve second-level dynamic power regulation. First, define the user utility function, with the formula:

[0096]

[0097] where U i (p i ) is the user utility function, which measures the benefits of users adjusting power. p i is the actual adjustment power of user i, which is a variable that users can control. is the demand power, reflecting the basic electricity consumption demand of users. α i is the price sensitivity coefficient, reflecting the sensitivity of users to price changes. β i is the charging delay penalty coefficient, used to penalize the behavior of users with charging delays. λ i is the adjustment cost coefficient, considering the costs incurred by users for adjusting power. Through this function, combined with the electricity price incentive coefficient, charging delay cost, and power adjustment benefits, a non-cooperative game model is constructed.

[0098] Design a Nash equilibrium solution algorithm and use the distributed gradient projection method to iteratively calculate the optimal response strategies of each user. In the non-cooperative game model, each user pursues the maximization of its own utility. Through the Nash equilibrium solution algorithm, a balanced state can be found such that the strategy of each user is the best response to the strategies of other users, thus achieving the optimization of the entire distribution network. The distributed gradient projection method can perform calculations distributively at the user terminals, improving the calculation efficiency and system flexibility.

[0099] Example 4:

[0100] This example details the construction and working principle of a load forecasting model based on a spatio-temporal graph model, providing strong support for the scheduling and planning of the distribution network through accurate load forecasting.

[0101] For the load forecasting model based on the spatio-temporal graph model, first construct a dynamic graph structure. The nodes in the graph represent the topological nodes of the distribution network, and the edge weights are jointly determined by the electrical distance and the real-time power flow. This dynamic graph structure can accurately reflect the electrical connection relationship and power transmission situation between the nodes in the distribution network, providing a basis for subsequent feature extraction.

[0102] Design a spatio-temporal convolutional layer to extract spatio-temporal correlation features of load distribution through alternating operations of one-dimensional convolution in the time dimension and graph convolution in the space dimension. The spatio-temporal convolution operation formula is as follows:

[0103] H (l+1) = σ(Φ temp *(Θ spa ★H (l) ))

[0104] Where, H (l+1) is the feature matrix of the (l + 1)-th layer, which contains the feature information extracted after the convolution operation of this layer. H (l) is the feature matrix of the l-th layer, Φ temp is the time convolution kernel, which is used to extract features in the time dimension. Θ spa is the spatial graph convolution parameter, which is used to extract features in the space dimension. * represents the one-dimensional convolution operation, ★ represents the graph convolution operation, and σ is the activation function, which is used to increase the non-linear expression ability of the model. Through this alternating operation, the correlation features of load distribution in time and space can be fully explored.

[0105] Adopt an attention mechanism to dynamically adjust the feature weights of different time steps and spatial nodes, and improve the adaptability of the prediction model to emergencies. In the actual operation of the distribution network, emergencies may occur, such as sudden load changes caused by extreme weather. The attention mechanism can dynamically adjust the feature weights according to the importance of different time steps and spatial nodes, making the model pay more attention to abnormal situations and improving the accuracy of prediction.

[0106] Example 5:

[0107] This example elaborates on the specific process of using the alternating direction method of multipliers for distributed solution of the multi-objective dynamic optimization solution unit, and how to achieve real-time solution of large-scale distribution network optimization problems through relevant mechanisms, so as to improve the operation efficiency of the distribution network.

[0108] Use the alternating direction method of multipliers for distributed solution to generate the optimal scheduling instructions under the global power balance constraint. First, decompose the global optimization problem into township-level main problems and village-level sub-problems, and realize variable coupling by introducing consistency constraints. In a large-scale distribution network, the global optimization problem is very complex and difficult to solve directly. By decomposing it into sub-problems, the complexity of the problem can be reduced, and at the same time, the coordination and unity between sub-problems are ensured through consistency constraints, ensuring that the final solution satisfies the global power balance constraint.

[0109] Design a residual feedback mechanism to dynamically adjust the penalty factor according to the solution deviation of the village-level sub-problems, accelerating the convergence of the algorithm. When there is a deviation between the solution of the village-level sub-problem and the expectation, the residual feedback mechanism can dynamically adjust the penalty factor according to the deviation magnitude, enabling the algorithm to converge to the optimal solution faster. Adopt a parallel computing architecture and utilize GPU to accelerate matrix operations to achieve real-time solution of large-scale distribution network optimization problems. When dealing with large-scale distribution network data, the amount of matrix operations is huge, and traditional computing methods are difficult to meet the real-time requirements. Utilizing the parallel computing ability of GPU can significantly improve the matrix operation speed, thus achieving real-time solution of large-scale distribution network optimization problems and providing support for the real-time scheduling of the distribution network.

[0110] Example 6:

[0111] This example focuses on the key units in the system to ensure communication efficiency, maintain voltage stability, and optimize energy storage management. Through the coordinated operation of the hierarchical communication protocol unit, voltage stability coordinated control unit, and distributed energy storage coordinated control unit, the comprehensive performance of the distribution network is comprehensively improved to ensure its safe, stable, and efficient operation.

[0112] The system sets up a hierarchical communication protocol unit, aiming to design a set of efficient hierarchical communication protocols to meet the data transmission requirements between different levels of the distribution network. In the design of the data encapsulation format of the township-level optimization instructions, a timestamp, target power value, and priority identifier are particularly incorporated. The timestamp accurately records the generation time of the instructions, providing guarantee for the timeliness of the instructions during transmission and execution, ensuring that the scheduling instructions can play their roles in a timely manner according to the real-time grid status; the target power value clarifies the specific regulation target, enabling the devices receiving the instructions to clearly know the adjustment direction; the priority identifier plays a key role in resource allocation in the case of limited communication resources, preferentially transmitting important and urgent instructions to ensure the smooth progress of the core regulation tasks of the power grid.

[0113] For the village-level regulation data, the system adopts an innovative compressed transmission protocol. This protocol uses sparse matrix coding and differential coding techniques to significantly reduce the data transmission volume. Sparse matrix coding makes full use of the sparse characteristics of the data, only encodes and transmits the key data, and discards a large number of redundant zero elements; differential coding transmits by calculating the differences between the data, avoiding repeated transmission of the same or similar data. The compressed transmission protocol satisfies the formula:

[0114]

[0115] where S is the original data matrix, is the compressed reconstruction matrix, R is the relative error rate, and ∈ is the preset error threshold. Through this strict error control standard, while effectively reducing the data transmission volume, the accuracy of the data is guaranteed, ensuring that the village-level regulation data is not distorted during transmission and providing a reliable basis for subsequent regulation decisions.

[0116] To further optimize the communication between the user terminal and the upper - level node, the system constructs an edge pre - processing mechanism for the user terminal status information. This mechanism performs filtering operations locally on the user terminal to remove noise interference in the data and improve data quality. At the same time, it carries out feature extraction work to extract key features closely related to the operation of the power grid, effectively reducing the redundancy of the uplink data, alleviating the transmission burden of the communication network, and improving the overall communication efficiency.

[0117] The voltage stability collaborative control unit in the system plays a crucial role. The voltage stability assessment module it constructs first constructs a reduced - order power grid dynamic model and accurately extracts the dominant oscillation mode by means of modal analysis. Modal analysis can deeply analyze the dynamic characteristics of the power grid, determine the oscillation mode that has the most critical impact on voltage stability, and provide a core basis for subsequent stability assessment and control.

[0118] Design an online calculator for stability indicators, and iteratively update the coefficient matrix of the Lyapunov function based on real - time measurement data. The formula for the Lyapunov function is:

[0119]

[0120] Among them, Δδ is the generator power angle deviation vector, M is the diagonal matrix of inertia coefficients, and V is the energy function. By updating the coefficient matrix of the Lyapunov function in real - time, the stable state of the power grid can be evaluated more accurately, and potential voltage instability risks can be detected in a timely manner.

[0121] According to the change rate of the stability margin, this module generates adaptive PID control parameters to dynamically adjust the response intensity of the reactive power compensation device. When a decrease in the voltage stability margin is detected, the compensation ability of the reactive power compensation device is automatically enhanced to maintain voltage stability; conversely, the compensation intensity is appropriately reduced to avoid over - compensation. The calculation formula for the voltage stability margin is:

[0122]

[0123] Among them, η is the system stability margin index, P i is the real - time power of node i, P i,max is the power limit of node i, and N is the set of nodes. By calculating the stability margin in real - time, a quantitative basis is provided for the control of the reactive power compensation device, ensuring that the power grid voltage is always within a stable range.

[0124] The distributed energy storage collaborative control unit is also indispensable. The designed distributed energy storage collaborative control strategy coordinates the charging and discharging timings of multiple-node energy storage through a consensus algorithm. During the operation of the distribution network, when power fluctuations or overload risks occur at some nodes, the consensus algorithm can reasonably arrange the charging and discharging sequences and powers of the energy storage according to the real-time states of each node, avoid power oscillations caused by uncoordinated energy storage actions, effectively improve the stability and reliability of the power grid, give full play to the regulating role of distributed energy storage in the distribution network, and optimize the allocation of power resources.

[0125] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0126] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A distribution network capacity-increase-free access system based on dynamic coordination, characterized in that: include: A multi-source data acquisition and fusion unit is used to obtain real-time operation data of the distribution network through a multi-source data acquisition module, wherein the multi-source data includes historical load data, meteorological data, distributed photovoltaic output data, electric vehicle charging demand data and power grid topology data; extract and fuse topological features of the real-time operation data based on a graph neural network to generate dynamic power grid state features; A multi-time scale collaborative optimization architecture unit is used to construct a multi-time scale collaborative optimization architecture, the architecture includes a township-level global optimization layer, a village-level real-time control layer, and a user terminal response layer; the township-level global optimization layer adopts a distributed robust optimization model to generate a day-ahead benchmark dispatch plan based on the dynamic power grid state characteristics; The village-level real-time control layer adopts an adaptive model predictive control algorithm to perform intraday rolling corrections based on the dispatch plan; the user terminal response layer adopts a game theory-driven load distribution strategy to achieve dynamic power regulation in seconds; The spatiotemporal load forecasting unit is used to establish a load forecasting model based on the spatiotemporal graph model. The load forecasting model integrates the travel chain characteristics, the volatility of photovoltaic output and the spatiotemporal distribution data of electric vehicle charging, uses the spatiotemporal graph convolutional network to predict the future multi-period load distribution, and outputs the load forecasting results; The multi-objective dynamic optimization solving unit is used to construct a multi-objective dynamic optimization model. The model takes the minimum grid loss, the minimum energy storage life loss and the maximum new energy consumption as optimization goals, adopts the alternating direction multiplier method for distributed solution, and generates the optimal dispatching instructions under the global power balance constraint.

2. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The township-level global optimization layer adopts a distributed robust optimization model to generate a day-ahead benchmark dispatch plan based on the dynamic power grid state characteristics, including: Constructing an uncertainty set, the set including a photovoltaic output fluctuation range, a load forecasting error, and an electric vehicle charging demand deviation; Establish a robust objective function, which minimizes the grid operation cost under the worst scenario and combines linear decision rules to robustly optimize the distributed generation output and energy storage charging and discharging strategy; A distributed iterative algorithm is used to solve the objective function, and the decomposition and coordination of global constraints are achieved through information interaction between adjacent nodes; The robust objective function is: Among them, u is the decision variable vector, including the distributed power output and energy storage charging and discharging power, and w is the uncertainty parameter vector, including the photovoltaic output P pv , load demand P load , Ω is an uncertain set, c g is the unit power generation cost coefficient, c ess is the energy storage cycle loss coefficient, γ is the regularization coefficient, and T is the optimization time window length.

3. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The village-level real-time control layer adopts an adaptive model predictive control algorithm to perform intraday rolling corrections based on the scheduling plan, including: Constructing a dynamic state space equation, wherein the equation includes a time-varying relationship between node voltage, energy storage state of charge, and line power; Design a rolling optimization objective function, which updates the model parameters in real time by combining the weighted tracking error term and the control variable change rate term with the online parameter identification technology; introduce a constraint softening mechanism, and when a hard constraint conflict is detected, relax the boundary conditions and add a penalty term to ensure the feasibility of the optimization problem; The dynamic state space equation is: x k+1 =A k x k +B k u k +E k d k Among them, x k+1 is the state vector at time step k+1, x k is the state vector at time step k, u k is the control vector at time step k, including the reactive output of the PV inverter, d k is the disturbance vector at time step k, including load fluctuations, A k ,B k ,E k is the time-varying system matrix.

4. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The user terminal response layer adopts a game theory-driven load distribution strategy to achieve second-level dynamic power regulation, including: A user utility function is defined, wherein the function combines an electricity price incentive coefficient, a charging delay cost, and a power regulation benefit to construct a non-cooperative game model; Design a Nash equilibrium solution algorithm and use the distributed gradient projection method to iteratively calculate the optimal response strategy for each user; The user utility function is: Among them, U i (p i ) is the user utility function, p i is the actual regulated power of user i, is the required power, α i is the electricity price sensitivity coefficient, β i is the charging delay penalty coefficient, λ i To adjust the cost factor.

5. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The load forecasting model based on the spatiotemporal graph model includes: Constructing a dynamic graph structure, wherein nodes in the graph represent topological nodes of the distribution network, and edge weights are determined by electrical distance and real-time power flow; Design a spatiotemporal convolution layer to extract the spatiotemporal correlation features of load distribution by alternating one-dimensional convolution in the time dimension with graph convolution in the space dimension. The attention mechanism is used to dynamically adjust the feature weights of different time steps and spatial nodes to improve the adaptability of the prediction model to emergencies; The spatiotemporal convolution operation is: H (l+1) =σ(Φ temp *(I spa ★H (l) )) Among them, H (l+1) is the feature matrix of the l+1th layer, H (l) is the feature matrix of the lth layer, Φ temp is the temporal convolution kernel, Θ spa is the spatial graph convolution parameter, * represents the one-dimensional convolution operation, ★ represents the graph convolution operation, and σ is the activation function.

6. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The distributed solution using the alternating direction multiplier method to generate the optimal scheduling instruction under the global power balance constraint includes: The global optimization problem is decomposed into township-level main problem and village-level sub-problems, and variable coupling is achieved by introducing consistency constraints; A residual feedback mechanism is designed to dynamically adjust the penalty factor according to the solution deviation of the village-level sub-problems to accelerate the convergence of the algorithm.

7. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The system further comprises: The layered communication protocol unit is used to design a layered communication protocol. The protocol combines 5G network slicing and edge computing technology to achieve millisecond-level issuance of township-level optimization instructions, real-time synchronization of village-level control data, and rapid feedback of user terminal status information.

8. The distribution network capacity-increase-free access system according to claim 7, characterized in that: The layered communication protocol includes: Define a data encapsulation format for township-level optimization instructions, the format including a timestamp, a target power value, and a priority identifier; Design a compression transmission protocol for village-level control data, and use sparse matrix coding and differential coding techniques to reduce data transmission volume; Build an edge preprocessing mechanism for user terminal status information to reduce uplink data redundancy through local filtering and feature extraction; The compression transmission protocol meets the following requirements: Among them, S is the original data matrix, is the compressed reconstruction matrix, R is the relative error rate, and ∈ is the preset error threshold.

9. The distribution network capacity-increase-free access system according to claim 1, characterized in that: The system further comprises: A voltage stability collaborative control unit, which is used to construct a voltage stability assessment module, which calculates the voltage stability margin of the power grid in real time based on the Lyapunov index analysis method and generates a preventive control strategy; The distributed energy storage collaborative control unit is used to design a distributed energy storage collaborative control strategy. The strategy coordinates the charging and discharging timing of multi-node energy storage through a consistency algorithm to avoid power oscillation and overload risks. The voltage stability margin calculation formula is: Among them, η is the system stability margin index, P i is the real-time power of node i, P i,max is the power limit of node i, A collection of nodes.

10. The distribution network capacity-increase-free access system according to claim 9, characterized in that: The voltage stability assessment module comprises: A reduced-order power grid dynamic model is constructed, which extracts the dominant oscillation mode through modal analysis; an online calculator of stability indicators is designed to iteratively update the Lyapunov function coefficient matrix based on real-time measurement data; Generate adaptive PID control parameters and dynamically adjust the response strength of the reactive power compensation device according to the stability margin change rate.

Images