Real-time scheduling method and system for mobile energy storage power distribution network driven by topology awareness prior under ice disaster toughness

Abstract

The invention discloses a topology awareness priori driven mobile energy storage power distribution network real-time scheduling method and a topology awareness priori driven mobile energy storage power distribution network real-time scheduling system under ice disaster toughness. The method comprises the following steps: constructing a line fault space-time distribution model changing along with icing evolution, and forming a multi-period and multi-scene ice disaster dynamic fault set; on the basis of the fault set and the multi-scene power flow, combining the topology and toughness indexes of the power distribution network, and screening by using a graph convolutional network to obtain a scale-controlled mobile energy storage candidate access node set; a random scheduling model is established under the power grid-road integrated constraint, and a mobile energy storage arrangement position, a driving path and a charging and discharging plan are jointly optimized. According to the method, ice disaster dynamic fault modeling is combined with candidate set screening assisted by topology perception learning, the dimensionality of a large-scale random mixed integer optimization model containing mobile energy storage is effectively reduced while line time sequence failure characteristics are fully reflected, the calculation burden is relieved, and the calculation efficiency is improved. And the toughness of the power distribution network in a multi-ice disaster scene and the timeliness of pre-disaster scheduling are improved.

Description

Technical Field

[0001] This invention relates to the field of power system optimization and dispatching technology, specifically to a real-time dispatching method for mobile energy storage distribution networks driven by topology perception prior under ice disaster resilience. Background Technology

[0002] With the increasing frequency and intensity of extreme freezing and ice storms, various types of faults caused by icing, such as tower collapses, line breaks, and flashovers, pose a serious threat to the safe operation of power systems. Among these, the distribution network, due to its dispersed structure and limited redundancy, has become the most vulnerable link in the power grid. The evolution of ice loads over time and space leads to the continuous accumulation of stress levels on lines and their propagation throughout the network. Faults exhibit obvious temporal expansion and spatial clustering characteristics, exacerbating both source-load imbalance and voltage exceedance risks, and making it difficult for traditional dispatch strategies based on static fault assumptions to reflect the evolution of disasters in a timely manner. To improve power supply reliability under extreme conditions, mobile energy storage systems, which can be transported by vehicles and flexibly connected to different nodes, are widely used for emergency power supply and microgrid formation, demonstrating good potential in enhancing critical load support and improving system resilience.

[0003] To address the aforementioned issues, existing technologies primarily analyze and optimize distribution network operation under ice storms or other extreme events by constructing typical fault scenarios, network reconfiguration models, and recovery scheduling models incorporating mobile energy storage. These methods have achieved some success in reducing load shedding and shortening outage times. However, these approaches often simplify ice storms into a few static fault scenarios, making it difficult to characterize the spatiotemporal evolution of line faults. Furthermore, considering the deployment location, travel path, and multi-period charging and discharging plans of mobile energy storage in the model often requires a large number of binary variables and scenario coupling constraints, easily leading to a high-dimensional, stochastic mixed-integer optimization problem with high solution costs and long computation times. Existing scheduling strategies still fall short in balancing the ability to characterize the dynamic evolution of faults with the timeliness of pre-disaster pre-planning and rapid post-disaster adjustment. A more efficient and topology-aware pre-disaster scheduling method for distribution networks is urgently needed. Summary of the Invention

[0004] Objective: This invention proposes a real-time scheduling method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience. It constructs a spatiotemporal line fault model driven by evolving ice loads, and then uses a graph convolutional network to select a candidate set for mobile energy storage system site selection. Based on this, a stochastic pre-disaster scheduling model coupled with the power grid and roads is established to jointly optimize the site selection, routes, and charging / discharging plans of the mobile energy storage system. This at least partially solves the problems of existing scheduling strategies in balancing the ability to characterize dynamic fault evolution with the timeliness of pre-disaster pre-planning and rapid post-disaster adjustment.

[0005] Technical Solution: To achieve the above-mentioned objectives, this invention proposes a real-time dispatching method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience, comprising the following steps:

[0006] Based on the characteristics of icing evolution over time and space, stress and failure processes of distribution lines per unit length are modeled. An initial set of distribution network fault scenarios is generated using a sampling method, and the initial set of distribution network fault scenarios is reduced by a variational Bayesian-Gaussian mixture model to obtain a representative set of ice disaster fault scenarios.

[0007] Under the real-time power flow assessment framework, the marginal mitigation contribution of different nodes to the system over-limit risk is compared with the same power support amplitude. In each representative ice storm fault scenario, the marginal mitigation contribution of each time period during the day is aggregated to obtain the resilience index of the nodes under global ice storm uncertainty. Based on the structural characteristics of the distribution network topology, the resilience index is used as the supervision label of the nodes. The topology-aware learning model is used to evaluate the comprehensive contribution of each node to load loss and network loss improvement, and a set of mobile energy storage candidate access nodes with controlled scale is selected as prior information for random scheduling.

[0008] Under the combined constraints of the integrated power grid-road network and the prior knowledge of candidate access nodes, a pre-disaster stochastic scheduling model is established, which includes the deployment location, travel path and charging and discharging plan of mobile energy storage. The model is solved under multiple ice disaster scenarios with the objective of minimizing the weighted sum of load loss and network loss, and the deployment scheme of mobile energy storage in the distribution network is obtained.

[0009] This invention also provides a real-time dispatching system for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience, comprising:

[0010] The dynamic fault modeling module for ice disasters is used to model the stress and failure process of a unit length of distribution line based on the characteristics of ice accretion evolving over time and space. It generates an initial set of distribution network fault scenarios using a sampling method and reduces the initial set of distribution network fault scenarios using a variational Bayesian-Gaussian mixture model to obtain a representative set of ice disaster fault scenarios.

[0011] The graph convolutional pre-decision module under fault evolution is used to compare the marginal mitigation contribution of different nodes to the system's over-limit risk under the same power support amplitude in the power flow assessment framework of real injection. In each representative ice disaster fault scenario, the marginal mitigation contribution of each time period during the day is aggregated to obtain the resilience index of the node under global ice disaster uncertainty. Based on the structural characteristics of the distribution network topology graph, the resilience index is used as the supervision label of the node. The topology-aware learning model is used to evaluate the comprehensive contribution of each node to load loss and network loss improvement, and a set of mobile energy storage candidate access nodes with controlled scale is selected as prior information for random scheduling.

[0012] The intelligent scheduling module is used to establish a pre-disaster stochastic scheduling model that includes the location of mobile energy storage, travel path and charging and discharging plan under the combined constraints of the integrated power grid-road network and the prior knowledge of candidate access nodes. Under multiple ice disaster scenarios, the model is solved with the objective of minimizing the weighted sum of load loss and network loss to obtain the layout scheme of mobile energy storage in the distribution network.

[0013] The present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the real-time dispatching method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience as described above.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the real-time scheduling method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience as described above.

[0015] Beneficial effects:

[0016] (1) Under extreme icing conditions such as ice storms, traditional methods often simplify line faults into a few static scenarios, which are difficult to reflect the spatiotemporal characteristics of ice load accumulation and fault propagation along the network. This invention constructs a spatiotemporal distribution model of distribution line faults driven by the evolution of icing loads, describing the probability of line failure and the evolution of line disconnection status in multiple time periods. This provides dynamic fault scenario inputs that are more consistent with the actual ice disaster process for pre-disaster scheduling, which helps to identify high-risk lines and key power supply channels in advance, improve the accuracy of fault prediction and the pertinence of pre-disaster planning.

[0017] (2) Although mobile energy storage systems have high flexibility, directly incorporating them into stochastic optimization will significantly amplify the model dimensionality and solution difficulty. This invention adopts a mobile energy storage pre-decision method that integrates topology-aware graph convolutional networks. It assesses the importance of nodes based on power flow and resilience indicators in multiple ice disaster scenarios, retains only a limited set of candidate access nodes, and jointly optimizes the deployment location, travel path, and charging and discharging plan of mobile energy storage under the constraints of grid-road integration. While compressing the scale of 0 / 1 decision variables and reducing the burden of stochastic mixed integer optimization, it can still significantly reduce load loss, network loss, and voltage deviation, meeting the engineering application needs within a limited pre-disaster time window. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the method of the present invention;

[0019] Figure 2 This is a diagram of the IEEE 33-node system architecture in an embodiment of the present invention.

[0020] Figure 3 This is a process diagram of ice storm attacking the power distribution network in an embodiment of the present invention;

[0021] Figure 4 This is a dynamic change diagram of a typical faulty circuit after reduction in an embodiment of the present invention;

[0022] Figure 5 This is a probability diagram of typical scenarios occurring after reduction in the embodiments of the present invention;

[0023] Figure 6 This is a comprehensive robust cost score diagram for each node in the implementation case of this invention. Detailed Implementation

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] Reference Figure 1 This invention proposes a real-time scheduling method for mobile energy storage distribution networks driven by topology sensing under ice disaster resilience, comprising the following steps:

[0026] Step 1: Based on the characteristics of ice accretion evolving over time and space, model the stress and failure process of the power line to obtain the spatiotemporal distribution of power line faults in multiple time periods and scenarios, providing a dynamic fault scenario set that reflects the evolution of ice disaster for subsequent scheduling models;

[0027] Step 2: Based on the dynamic fault scenario and multi-scenario power flow results, and combined with the distribution network topology and resilience index, construct the graph structure characteristics, use the topology-aware learning model to evaluate the comprehensive contribution of each node to load loss and network loss improvement, and screen out a set of mobile energy storage candidate access nodes with controlled scale as prior information for random optimization to compress the decision space.

[0028] Step 3: Under the combined constraints of the integrated power grid-road network and the prior knowledge of candidate nodes, establish a pre-disaster stochastic scheduling model that includes the location of mobile energy storage deployment, travel path and charging and discharging plan. Under multiple ice disaster scenarios, solve the pre-layout scheme with the objective of minimizing the weighted sum of load loss and network loss.

[0029] The following detailed implementation process of using the method of this invention for distribution network optimization is illustrated with a specific example. In this implementation example, a modified IEEE 33-bus distribution system is selected as the benchmark simulation example. The system topology is shown in the attached figure. Figure 2As shown in the figure. In this system, some nodes are designated as Class I load nodes and the rest as Class II load nodes. Wind turbines and distributed photovoltaics are configured to construct an operating environment with uncertainties in new energy sources. For the ice disaster part, a line fault model based on the evolution of icing load is adopted. Multiple sets of dynamic fault scenarios of the distribution network are generated using the quasi-Monte Carlo method and then clustered to obtain several typical representative scenarios. These are used as inputs to the dynamic fault modeling and mobile energy storage scheduling model of this invention to verify the effectiveness of the proposed method in reducing load loss and network loss and improving the resilience of the distribution network under ice disasters.

[0030] Based on the constructed model data, the specific implementation steps of the method of the present invention are as follows:

[0031] (1) Modeling the probability of line failure in ice storm scenarios.

[0032] During the evolution of an ice storm, wind and icing processes often occur simultaneously, affecting various power distribution lines. The wind speed at a given location during a given time period is defined as:

[0033]

[0034]

[0035]

[0036] in, and They are respectively Time-of-use power distribution lines Wind speed and maximum wind speed at the location, in units of ; For Gaussian kernel, From the storm center to the line Distance from the geometric center Radius affected by wind speed It is a numerically stable term; This refers to the wind speed in the direction perpendicular to the line. and They are respectively Wind direction and azimuth of the line during the time period.

[0037] Based on the wind speed and environmental conditions at the location of the line, the cumulative icing thickness of the line can be expressed as:

[0038]

[0039] in, for Time-of-day routes The cumulative icing thickness, For unit duration, and The densities of ice and water are respectively. The liquid water content in the air. The rainfall rate.

[0040] In extreme ice storm conditions, ice-covered power distribution lines exposed to the air bear not only horizontal wind loads. It also needs to resist the vertical load generated by the weight of the ice itself. As the ice thickness continues to increase, the overall load... When the line's load-bearing capacity is exceeded, a line breakage fault may occur. The wind load and ice load per unit length of line can be expressed as:

[0041]

[0042]

[0043]

[0044] in, and They are respectively Wind load and ice load per unit length of line during the time period, in units of , As a constant factor, take , This is the line span coefficient. The diameter of the line. Let be the acceleration due to gravity, and take . ; for Comprehensive load per unit length of line during a given time period.

[0045] When the load exceeds this threshold, its load-bearing capacity decreases exponentially with the increase of the generated strain, leading to transmission line failure. Assuming the cross-section of the transmission line is elliptical, the failure probability per unit length of the distribution line can be expressed as:

[0046]

[0047]

[0048] in, for The probability of line failure per unit length over a given time period and These are two threshold values ​​for line faults; for Time-of-day routes The probability of failure, For the line Length, It is a constant, taken as 0.6931.

[0049] (2) Generation of dynamic fault scenarios in distribution networks under ice storms

[0050] A two-stage sampling strategy using the Quasi-Monte Carlo (QMC) method and the Determinantal Point Process (DPP) is adopted: First, QMC is used to generate candidate samples with uniform coverage and low dissimilarity in the high-dimensional storm parameter space to reduce sampling variance and avoid coverage gaps; then, DPP is introduced into the candidate set to perform diversity-driven subset selection, which prioritizes the retention of samples with greater dissimilarity, thereby reducing sample clustering and improving coverage of boundary and extreme combinations.

[0051]

[0052] in, For the scene Downline exist The fault status for a given time period, with 0 indicating a fault and 1 indicating normal operation. For the line In the scene Down The time period extracted is located in Uniformly distributed random numbers.

[0053] By iteratively executing the above sampling process, a predetermined number of distribution network fault scenarios are generated. Then, a Variational Bayesian Gaussian Mixture Model (VB-GMM) is introduced to reduce this scenario set, ultimately constructing... This study presents a representative set of dynamic fault scenarios during ice storms and their probabilities of occurrence. This set of typical scenarios provides a more accurate and reliable data foundation for subsequent evaluation of dynamic adjustment strategies for mobile energy storage grid connection points.

[0054] (3) Based on dynamic fault scenarios and multi-scenario power flow results, construct a candidate set generated by graph convolutional network.

[0055] Extreme weather-driven faults cause the available topology and power flow constraints of distribution networks to evolve over time. Simultaneously, loads, distributed generation, reactive power compensation devices, and energy storage devices exhibit significant time-varying characteristics and mutual coupling on an intraday scale. To determine the priority deployment locations of mobile energy storage before a disaster, this section, within a real-world power flow assessment framework, measures the improvement in system over-limit risk if a small power input is provided at a node, and performs robust aggregation over time to obtain a value score for each node. This indicator reflects both the network structure and the coupling effect between equipment and load-source output, providing physically interpretable hot-start candidates for subsequent optimization.

[0056]

[0057] in, for Time-based system normalization exceeding limits risk. For the line exist Apparent tidal amplitude over a period of time For the line Maximum capacity, For the line exist The running status of a time period is 1 if running, and 0 otherwise; For the set of all lines, It is a numerically stable term.

[0058] To compare the mitigation effects of different nodes on system over-limit risk under the same power support amplitude, we consider the following at the nodes. Injecting small amount of active power At the equilibrium node, the same amount is extracted to maintain power flow conservation, the risk of exceeding the limit is calculated again, and the node is defined. exist Marginal mitigation contribution over time period:

[0059]

[0060]

[0061]

[0062]

[0063] in, and They are nodes exist Load and local equivalent power generation during the time period, The set of all nodes. This represents the net load of the entire network. and These are the upper and lower limits for injected or absorbed active power, corresponding to the upper and lower limits for the output of mobile energy storage. This is a scaling factor used to scale the "system net load" to the magnitude of a single small disturbance during injection. This means to use a scalar Crop to Within the range; and For the node Apply The marginal mitigation amount obtained from the current flow is then calculated again. For nodes exist Marginal mitigation contribution over time period Indicates at node Deploying mobile energy storage can reduce the risk of exceeding limits.

[0064] When considering a single ice storm evolution scenario, nodes During the period The marginal mitigation contribution is Since actual ice storms have multiple possible evolution paths, this invention obtains a representative scenario set based on multi-scenario ice storm simulation and scenario reduction. and its probability of occurrence In each representative scenario Below, for each time period of the day Quantile aggregation is performed, and then a weighted average is calculated according to the scenario probability to obtain the node. Comprehensive robustness value under global ice disaster uncertainty :

[0065]

[0066] The comprehensive robust value In this paper, this is referred to as the resilience index. Given the multi-scenario and multi-period context of ice storms, directly optimizing the pre-disaster deployment of mobile energy storage across all busbars in the entire network would result in high decision-making dimensionality, heavy computational burden, and difficulty in timely supporting pre-disaster decision-making. Therefore, this paper introduces a data-driven node selection step before optimization, pre-selecting a group of high-potential access nodes based on the contribution of each busbar to mitigating over-limit risks. Considering the significant graph topology characteristics of the distribution network, this invention employs a topology-aware GCN to model and rank node importance, using its output candidate set as a priori constraint for the subsequent optimization model.

[0067] GCN drives the evolution of bus features through multi-layered interactions, with its core being the characterization of information diffusion and topological dependencies. In each layer, buses pass messages and aggregate along network edges to neighboring buses, forming dynamic feature propagation channels. Furthermore, bidirectional message passing extracts geometric relationship patterns within a spatial correlation framework, constructing a hierarchical topological representation layer by layer.

[0068] This invention employs a node representation graph structure that integrates temporal and static features. Each node corresponds to a node in a distribution network, and its feature vector is composed of the two types of features mentioned above: temporal features are obtained from various power curves, including active and reactive loads, photovoltaic and wind power output, energy storage charging and discharging power, and reactive power compensation output, over 24 time periods of a typical operating day, used to characterize the dynamic behavior of nodes during daily operation and regulation; static features aggregate node connection methods, reference voltage amplitude, whether photovoltaic or wind power is connected, distributed generation capacity, whether it is a critical load node, node degree, static var compensator (SVC) and parallel capacitor (CB) parameters, reflecting the static attributes of nodes in terms of topological location and regulation capability. Based on this, the comprehensive robustness value calculated above is not used as node feature input, but as a supervisory label for each node, used to quantify the comprehensive mitigation of overload and load shedding risks after deploying mobile energy storage at that node, thereby guiding the GCN to learn from the above local temporal and static features and predict the global resilience contribution of nodes.

[0069] After standardization and unification of dimensions, the features are fused and input into the GCN, enabling the model to jointly learn time response and spatial adaptability under topological constraints, thereby obtaining a candidate set for mobile energy storage pre-deployment. The loss function of GCN is as follows:

[0070]

[0071] in, For nodes The supervisory label, when mobile energy storage is deployed to this node, can reduce network boundary violations and is defined as a positive class sample. It is 1 if it is true, otherwise it is 0; For the Sigmoid function, For the model to nodes Output the real number fractions that do not pass the sigmoid function. For positive class weights, and The number of positive and negative class samples. Negative class weights It is the natural logarithm.

[0072] (4) Under the combined effect of the power grid-road network integration constraint and the prior of candidate nodes, a pre-disaster stochastic scheduling model is established.

[0073] Given the set of all possible failure scenarios induced by the evolution of ice storms, this invention aims to minimize the joint network loss and load loss of the system. The objective function can be expressed as:

[0074]

[0075] in, This is a collection of ice storm failure scenarios. Scene The probability of occurrence, branch road The resistance, For the scene of Branch road during the period The squared variable of the current; For nodes The importance weights are set differently for first-level and second-level users. For the scene of Time period node The power of the load shedding This is a collection of all time periods.

[0076] The specific operating models for each device include:

[0077] 1) Operational constraints of mobile energy storage

[0078]

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090] in, The maximum number of configurable mobile energy storage units. For the first Configuration decision variables for mobile energy storage in Taiwan; For the first Taiwan Mobile Energy Storage Scenarios Does the time period coincide with the node? connect; For the first Taiwan Mobile Energy Storage Time period from node Drive to the node Travel time, To ensure that, under the influence of disasters and congestion, at all times node With nodes The equivalent driving distance between them For the first Taiwan Mobile Energy Storage The actual vehicle speed during the time period node With nodes The geometric distance of the road under normal operating conditions. For the first Taiwan's mobile energy storage operates at ideal vehicle speeds under zero traffic congestion conditions; and The first Taiwan Mobile Energy Storage Scenarios of Time period and Time period connection node binary variables, A fixed timeframe for completing the access operation at the node; and The first Taiwan Mobile Energy Storage Scenarios of A binary variable representing the charging / discharging state over a given time period; , The first Taiwan Mobile Energy Storage Scenarios of Active power during charging / discharging over a given period; and The first Taiwan Mobile Energy Storage Scenarios of Energy storage during a specific period of time, and For charge / discharge efficiency; and For the first The upper and lower limits of the energy of mobile energy storage in Taiwan; To linearize the auxiliary binary variables, a linearization method is used to convert the bilinear coupling into equivalent linear constraints.

[0091] 2) Operating constraints of distributed power sources

[0092]

[0093]

[0094]

[0095] in, and For distributed power sources in various scenarios of The contribution and ineffectiveness of effort during a given period , , , These are the upper and lower limits of the active and reactive power of the distributed power source. and These are the upper and lower limits of the power factor angle.

[0096] 3) Operational constraints of stationary energy storage systems

[0097]

[0098]

[0099]

[0100]

[0101] in, , They are nodes Fixed energy storage scenario of Active power during charging / discharging over a given period; and They are nodes Fixed energy storage scenario of and Energy storage during a specific period of time, and For charge / discharge efficiency; and These are the upper and lower limits of ESS energy.

[0102] 4) Operational constraints of reactive power compensation equipment

[0103]

[0104]

[0105]

[0106] in, For the scene exist Number of capacitor banks put into operation during the period The compensation power for each group of capacitors. This is the upper limit for the number of CB groups put into operation. This is the maximum number of times that can be adjusted within the scheduling cycle. For SVC in the scene Time period Internally compensated reactive power, This represents the maximum capacity of the SVC.

[0107] 5) Power flow security constraints in distribution networks

[0108]

[0109]

[0110]

[0111]

[0112]

[0113]

[0114]

[0115] in, and These are the power outflow and inflow nodes, respectively. The set of branches; and Scenes exist Time period is determined by nodes Flow direction The trend of meritorious and ineffective actions; and The lines are respectively Resistance and reactance; and Scenes exist Time period nodes The active and reactive loads; and Scenes exist Time period nodes The active and reactive loads that were removed; and Scenes exist Time period nodes and The square of the voltage amplitude; It is a very large positive number; and Scenes exist Upper and lower limits of the squared term of the voltage amplitude at time points; For the scene exist The upper limit of the square term of the branch current during the time period.

[0116] Because of the existence of the work balance equation and Nonlinear terms are difficult to solve directly, so this invention uses the Big-M method to linearize them:

[0117]

[0118]

[0119]

[0120] in, and It is a non-negative continuous variable.

[0121] After solving the optimization model consisting of the aforementioned objective function and constraints, the decision results of the pre-disaster deployment scheme can be obtained, including the stationary node sequence of each mobile energy storage unit in each time period, as well as the corresponding charging and discharging power and energy state plan. In addition, the model also simultaneously outputs the coordinated control quantities of stationary energy storage, distributed power output, and reactive power regulation devices, which are used to minimize network losses and load shedding and maintain operational constraints under multi-scenario fault conditions.

[0122] Figure 3 This is a process diagram of an ice storm affecting a power distribution network in an embodiment of the present invention. The ice storm begins at 0:00 and lasts for 24 hours. The initial location of the ice storm center is... Ice disaster along the horizontal axis The angle is moving at a speed of 4.5 km / h.

[0123] Figure 4 The presentation showcases 10 typical scenarios that were generated from 300 power grid fault scenarios using QMC sampling and then clustered and reduced using VB-GMM. Figure 5 This represents the probability of occurrence in each typical scenario.

[0124] Figure 6The results demonstrate the comprehensive robustness cost of each node obtained by GCN based on dynamic fault scenarios and multi-scenario power flow results. Nodes 6, 17, 23, and 25 score approximately 0.9, significantly higher than the other nodes; nodes 1, 24, and 30 score approximately 0.1–0.3; and the rest are close to 0. Considering the overall score distribution and the scale of subsequent solutions, [6, 17, 23, 25] is ultimately selected as the candidate set for mobile energy storage for pre-disaster deployment and post-disaster dynamic scheduling.

[0125] To demonstrate the effectiveness of the proposed GCN candidate-guided pre-disaster mobile energy storage pre-deployment method that considers the spatiotemporal evolution characteristics of faults under ice disasters, the following six schemes are designed and compared according to whether or not the dynamic evolution process of grid faults is considered.

[0126] Scheme 1: The method of this invention uses GCN to generate a candidate node set for mobile energy storage deployment, explicitly characterizes the spatiotemporal evolution process of power grid faults under the influence of ice disaster in the pre-disaster pre-deployment model, and considers vehicle travel path and time constraints.

[0127] Option 2: Without setting a candidate set under the same model, optimize the deployment and scheduling of mobile energy storage directly on all network nodes;

[0128] Option 3: Use a Convolutional Neural Network (CNN) instead of the GCN in Option 1 to generate a candidate node set, in order to compare the impact of different data-driven models on the pre-layout effect;

[0129] Option 4: While keeping the model optimization objective of Option 1 unchanged, replace the solver with the particle swarm optimization algorithm and evaluate the differences in performance and efficiency between heuristic algorithms and exact optimization methods.

[0130] 5: The pre-disaster deployment model of mobile energy storage no longer describes the fault sequence evolution during the ice disaster process, but only uses the static results of the final line faults in each scenario for optimization, in order to compare the benefits brought by dynamic fault modeling.

[0131] Option 6: Considering that mobile energy storage does not participate in dispatching during the dynamic evolution of power grid faults under ice storms, evaluate the positive role of mobile energy storage in dispatching.

[0132] Table 1 Comparison of optimization results between Scheme 1 and Schemes 2 and 3

[0133]

[0134] Table 1 compares the optimization results of Schemes 1, 2, and 3 under different numbers of ice storm scenarios. Compared with Scheme 2, Scheme 1 reduces the solution time of the optimization algorithm by 89.52%, 89.35%, and 91.19% for 2, 5, and 10 scenarios, respectively, while the optimization results for network loss and load shedding are not significantly different. Since calculating the access location of mobile energy storage introduces many 0 / 1 decision variables into the optimization problem, the candidate set can effectively narrow the search range of the optimization algorithm under the same number of scenarios, resulting in faster results. Furthermore, the candidate set is derived considering the improvement of line over-limit conditions after applying power to nodes in multiple scenarios and time periods; therefore, the optimization results are similar to those of the global search. In addition, with the increase in the number of scenarios, the algorithm without a candidate set experiences a significant increase in solution time, leading to a sharp increase in memory consumption during the solution process, which is not conducive to adapting to the requirements of rapid optimization with low hardware costs.

[0135] Based on this, Scheme 1 using GCN and Scheme 3 using CNN are compared. Both schemes introduce candidate sets, so the solution time is roughly the same. However, Scheme 3 performs slightly worse than Scheme 1 in terms of operational indicators such as network loss and load shedding. This is because GCN can perform feature propagation and aggregation on the graph structure, and better perceive the topological location, electrical connectivity, and over-limit improvement effect of the IEEE 33-node system under multiple scenarios and time periods during training. This results in a candidate set [6,17,23,25] that takes into account both key branches and high-load nodes and has a relatively dispersed spatial distribution. In contrast, CNN relies on regular convolution kernels, which makes it difficult to fully characterize the graph topology of the radial distribution network. Its candidate set is concentrated near a cluster of downstream nodes [17,18,21,22], failing to cover important adjustment potential locations such as nodes 6, 23, and 25. This limits the available adjustment space, and the overall optimization effect is not as good as Scheme 1.

[0136] Table 2 Comparison of optimization results between Scheme 1 and Schemes 4, 5, and 6

[0137]

[0138] Table 2 compares the optimization results of Schemes 1, 4, 5, and 6 under different numbers of ice storm scenarios. With the same number of scenarios, Scheme 1 outperforms Scheme 4 overall in terms of network loss, load shedding, and average voltage deviation. When the number of scenarios is small (S=2, 5), these three indicators can be reduced to about half that of Scheme 4, and a certain degree of improvement is maintained even when the number of scenarios increases to S=10. Scheme 4 uses a particle swarm optimization algorithm for heuristic search, and its solution time is significantly shorter than that of Scheme 1, which is based on stochastic optimization. However, due to the difficulty in systematically handling multi-scenario coupling constraints and discrete decision-making, its optimization level in terms of comprehensive network loss, load shedding, and voltage quality is slightly inferior, demonstrating the advantages of the method described in this paper in resilience improvement problems.

[0139] Under the same number of scenarios, Scheme 1 has lower network loss and load shedding than Scheme 5. At S=5, network loss increases by 70.99%, and load shedding also increases by 72.85% to 79.11%. In Scheme 5, some lines consistently fail at all 24 time points. This fault setting often leads to an overly conservative optimization strategy. Furthermore, because the number of line failures is very high at the beginning of optimization, the optimization algorithm struggles to achieve the lower network loss and load shedding of Scheme 1 from the outset.

[0140] The comparison with Scheme 6 further highlights the role of mobile energy storage dispatch. Under the same number of scenarios, the network loss and load shedding of Scheme 6 are still significantly higher than those of Scheme 1. When S=2, the increase in network loss is approximately 74.71%, and the increase in load shedding is approximately 52.05% to 74.99%. Since Scheme 6 assumes that mobile energy storage does not participate in dispatch, the number of 0 / 1 decision variables in the model is significantly reduced, thus shortening the solution time accordingly. However, at the same time, the system lacks the spatiotemporal reconfiguration capability of mobile energy storage during the fault evolution process, and can only rely on the redundancy of fixed power sources and the network itself to cope with the uncertainty of load and fault, resulting in more loads being cut off and a significant deterioration in network loss and average voltage deviation.

[0141] This invention constructs a spatiotemporal distribution model of distribution line faults driven by the evolution of icing loads, characterizing the failure probability and outage state evolution of lines at multiple time periods and scenario scales, providing dynamic fault inputs that closely resemble actual ice disaster processes for pre-disaster scheduling. It designs a mobile energy storage pre-decision model integrating a topology-aware graph convolutional network, assessing node importance based on power flow and resilience characteristics across multiple ice disaster scenarios, and generating a controlled-scale set of candidate access nodes for mobile energy storage. Furthermore, it establishes a stochastic pre-disaster scheduling model coupled with grid and road network constraints, jointly optimizing the location, travel path, and charging / discharging plans of mobile energy storage in multiple scenarios, with the objective of minimizing the weighted sum of load loss, network loss, and voltage deviation. This invention combines dynamic fault modeling for ice disasters, topology-aware learning priors, and stochastic scheduling of mobile energy storage, effectively reducing load loss levels and network losses under extreme icing conditions, improving voltage deviation, and enhancing the comprehensive resilience and engineering application value of the distribution network in ice disaster scenarios, while considering both solution efficiency and pre-disaster time constraints.

[0142] Based on the same technical concept as the method embodiments, the present invention also provides a real-time dispatching system for mobile energy storage distribution networks driven by topology awareness under ice disaster resilience, comprising:

[0143] The dynamic fault modeling module for ice disasters is used to model the stress and failure process of a unit length of distribution line based on the characteristics of ice accretion evolving over time and space. It generates an initial set of distribution network fault scenarios using a sampling method and reduces the initial set of distribution network fault scenarios using a variational Bayesian-Gaussian mixture model to obtain a representative set of ice disaster fault scenarios.

[0144] The graph convolutional pre-decision module under fault evolution is used to compare the marginal mitigation contribution of different nodes to the system's over-limit risk under the same power support amplitude in the power flow assessment framework of real injection. In each representative ice disaster fault scenario, the marginal mitigation contribution of each time period during the day is aggregated to obtain the resilience index of the node under global ice disaster uncertainty. Based on the structural characteristics of the distribution network topology graph, the resilience index is used as the supervision label of the node. The topology-aware learning model is used to evaluate the comprehensive contribution of each node to load loss and network loss improvement, and a set of mobile energy storage candidate access nodes with controlled scale is selected as prior information for random scheduling.

[0145] The intelligent scheduling module is used to establish a pre-disaster stochastic scheduling model that includes the location of mobile energy storage, travel path and charging and discharging plan under the combined constraints of the integrated power grid-road network and the prior knowledge of candidate access nodes. Under multiple ice disaster scenarios, the model is solved with the objective of minimizing the weighted sum of load loss and network loss to obtain the layout scheme of mobile energy storage in the distribution network.

[0146] It should be understood that the mobile energy storage distribution network real-time dispatch system driven by topology perception under ice disaster resilience in the embodiments of the present invention can realize all the technical solutions in the above method embodiments. The functions of each functional module can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can be referred to the relevant descriptions in the above embodiments, which will not be repeated here.

[0147] The present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the real-time dispatching method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience as described above.

[0148] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the real-time scheduling method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience as described above.

[0149] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus (systems), electronic devices, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.

[0150] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that each step in the flowchart and combinations thereof 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 device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0151] 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 The function specified in one or more processes.

[0152] 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 Steps of a specified function in one or more processes.

Claims

1. A real-time scheduling method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience, characterized in that, The method includes the following steps: Based on the characteristics of icing evolution over time and space, stress and failure processes of distribution lines per unit length are modeled. An initial set of distribution network fault scenarios is generated using a sampling method, and the initial set of distribution network fault scenarios is reduced by a variational Bayesian-Gaussian mixture model to obtain a representative set of ice disaster fault scenarios. Under the real-time power flow assessment framework, the marginal mitigation contribution of different nodes to the system over-limit risk is compared with the same power support amplitude. In each representative ice storm fault scenario, the marginal mitigation contribution of each time period during the day is aggregated to obtain the resilience index of the nodes under global ice storm uncertainty. Based on the structural characteristics of the distribution network topology, the resilience index is used as the supervision label of the nodes. The topology-aware learning model is used to evaluate the comprehensive contribution of each node to load loss and network loss improvement, and a set of mobile energy storage candidate access nodes with controlled scale is selected as prior information for random scheduling. Under the combined constraints of the integrated power grid-road network and the prior knowledge of candidate access nodes, a pre-disaster stochastic scheduling model is established, which includes the deployment location, travel path and charging and discharging plan of mobile energy storage. The model is solved under multiple ice disaster scenarios with the objective of minimizing the weighted sum of load loss and network loss, and the deployment scheme of mobile energy storage in the distribution network is obtained.

2. The method according to claim 1, characterized in that, The characteristics of icing evolution over time and space are represented as follows: ； in, and They are respectively Time-of-day routes Wind speed and maximum wind speed at the location; For Gaussian kernel, From the storm center to the line Distance from the geometric center Radius affected by wind speed It is a numerically stable term; This refers to the wind speed in the direction perpendicular to the line. and They are respectively Wind direction and azimuth of the line during the time period; for Time-of-day routes The cumulative icing thickness, For unit duration, and The densities of ice and water are respectively. The liquid water content in the air. Rainfall rate; The force model for a unit length of power distribution line is as follows: ； in, and They are respectively Wind load and ice load per unit length of line during time period, A constant factor, This is the line span coefficient. The diameter of the line. It is the acceleration due to gravity; for Comprehensive load per unit length of line during a given time period; The failure model for a unit length of power distribution line is as follows: ； ； in, for The probability of line failure per unit length over a given time period. and These are two threshold values ​​for line faults; for Time-of-day routes The probability of failure, For the line Length, It is a constant.

3. The method according to claim 1, characterized in that, An initial set of distribution network fault scenarios is generated using sampling methods, including: A two-stage sampling strategy using the quasi-Monte Carlo (QMC) method and the determinant point process (DPP) is employed: First, a candidate sample set is generated in the high-dimensional storm parameter space using QMC; then, DPP is introduced into the candidate sample set for diversity-driven subset selection, prioritizing the retention of samples with greater differences from each other, as shown below: ； in, For the circuit in the scene exist The fault status for a given time period, with 0 indicating a fault and 1 indicating normal operation. For the line In the scene Down The time period extracted is located in Uniformly distributed random numbers, for Time-of-day routes The probability of failure.

4. The method according to claim 1, characterized in that, The marginal contribution of a node to mitigating the risk of exceeding system limits is expressed by the formula: ； in, for Time-based system normalization exceeding limits risk. For the line exist Apparent tidal amplitude over a period of time For the line Maximum capacity, For the line exist The running status of a time period is 1 if running, and 0 otherwise; For the set of all lines, It is a numerically stable term; For the node The injected small amount of active power, and They are nodes exist Load and local equivalent power generation during the time period, For the combination of all nodes, This represents the net load of the entire network. and These are the upper and lower limits of injected or absorbed active power, corresponding to the upper and lower limits of mobile energy storage output. This is a scaling factor used to scale the "system net load" to the magnitude of a single small disturbance during injection. This means to use a scalar Crop to Within the range; and For the node Apply The marginal mitigation amount obtained from the current flow is then calculated again. For nodes exist Marginal mitigation contribution over time period Indicates at node Deploying mobile energy storage can reduce the risk of exceeding limits; The resilience index of a node under global ice disaster uncertainty is expressed as: ； in, In the scene Next node During the period The marginal mitigation contribution, This is a representative set of ice storm failure scenarios. For the scene The probability of occurrence; For each scene Below, for each time period of the day The nodes are obtained by performing quantile aggregation and weighting the average according to the scenario probability. Comprehensive robustness value under global ice disaster uncertainty.

5. The method according to claim 1, characterized in that, Based on the structural characteristics of the distribution network topology diagram, including: The nodes in the distribution network are mapped to nodes in the graph structure. The feature vector of a node is composed of temporal features and static features. The temporal features are the features that describe the dynamic behavior of the node during the daytime operation and control process, while the static features are the features that reflect the static attributes of the node in terms of topological location and regulation capability. After unifying the units and standardizing, the features are fused into a topology-aware learning model composed of a graph convolutional neural network. This allows the model to jointly learn time response and spatial adaptability under topological constraints, thereby obtaining a candidate set for mobile energy storage pre-deployment. The loss function of the graph convolutional neural network is as follows: ； in, For nodes The supervisory label, when mobile energy storage is deployed to nodes to reduce network boundary violations, is defined as a positive class sample. It is 1 if it is true, otherwise it is 0; For the Sigmoid function, For the model to nodes Output the real number fractions that do not pass the sigmoid function. For positive class weights, and The number of positive and negative class samples. Negative class weights It is the natural logarithm.

6. The method according to claim 1, characterized in that, The pre-disaster stochastic scheduling model aims to minimize both network loss and load loss. The objective function is expressed as: ； in, This is a collection of ice storm failure scenarios. For the scene The probability of occurrence, branch road The resistance, For the scene of Branch road during the period The squared variable of the current; For nodes The importance weighting is different for first-level and second-level users. For the scene of Time period node The power of the load shedding For the set of all lines, The set of all nodes. This is a collection of all time periods.

7. The method according to claim 6, characterized in that, The pre-disaster stochastic dispatch model includes the location of mobile energy storage deployment, travel routes, and charging / discharging plans. Its integrated grid-road network constraints include: Mobile energy storage operation constraints: ； ； ； ； ； ； ； ； ； ； ； ； in, The maximum number of configurable mobile energy storage units. For the first Configuration decision variables for mobile energy storage in Taiwan; For the first Taiwan Mobile Energy Storage Scenarios Does the time period coincide with the node? connect; For the first Taiwan Mobile Energy Storage Time period from node Drive to the node Travel time, To ensure that, under the influence of disasters and congestion, at all times node With nodes The equivalent driving distance between them For the first Taiwan Mobile Energy Storage The actual vehicle speed during the time period node With nodes The geometric distance of the road under normal operating conditions. For the first Taiwan's mobile energy storage operates at ideal vehicle speeds under zero traffic congestion conditions; and The first Taiwan Mobile Energy Storage Scenarios of Time period and Time period connection node binary variables, A fixed timeframe for completing the access operation at the node; and The first Taiwan Mobile Energy Storage Scenarios of A binary variable representing the charging / discharging state over a given time period; , The first Taiwan Mobile Energy Storage Scenarios of Active power during charging / discharging over a given period; and The first Taiwan Mobile Energy Storage Scenarios of Energy storage during a specific period of time, and For charge / discharge efficiency; and For the first The upper and lower limits of the energy of mobile energy storage in Taiwan; To linearize the auxiliary binary variables, a linearization method is used to equivalently transform bilinear coupling into linear constraints; Distributed power source operating constraints: ； ； ； in, and For distributed power sources in various scenarios of The contribution and ineffectiveness of effort during a given period , , , These are the upper and lower limits of the active and reactive power of the distributed power source. and These are the upper and lower limits of the power factor angle; Operating constraints of stationary energy storage systems: ； ； ； ； in, , They are nodes Fixed energy storage scenario of Active power during charging / discharging over a given period; and They are nodes Fixed energy storage scenario of and Energy storage during a specific period of time, and For charge / discharge efficiency; and The upper and lower limits of ESS energy; Operating constraints of reactive power compensation equipment: ； ； ； in, For the scene exist Number of capacitor banks put into operation during the period The compensation power for each group of capacitors. This is the upper limit for the number of CB groups put into operation. This is the maximum number of times that can be adjusted within the scheduling cycle. For SVC in the scene Time period Internally compensated reactive power, This represents the maximum capacity of the SVC. Distribution network power flow security constraints: ； ； ； ； ； ； ； in, and These are the power outflow and inflow nodes, respectively. The set of branches; and Scenes exist Time period is determined by nodes Flow direction The trend of meritorious and ineffective actions; and The lines are respectively Resistance and reactance; and Scenes exist Time period nodes The active and reactive loads; and Scenes exist Time period nodes The active and reactive loads that were removed; and Scenes exist Time period nodes and The square of the voltage amplitude; It is a very large positive number; and Scenes exist Upper and lower limits of the squared term of the voltage amplitude at time points; For the scene exist The upper limit of the square term of the branch current during the time period.

8. A real-time dispatching system for mobile energy storage distribution networks driven by topology sensing priors under ice disaster resilience, characterized in that, include: The dynamic fault modeling module for ice disasters is used to model the stress and failure process of a unit length of distribution line based on the characteristics of ice accretion evolving over time and space. It generates an initial set of distribution network fault scenarios using a sampling method and reduces the initial set of distribution network fault scenarios using a variational Bayesian-Gaussian mixture model to obtain a representative set of ice disaster fault scenarios. The graph convolutional pre-decision module under fault evolution is used to compare the marginal mitigation contribution of different nodes to the system's over-limit risk under the same power support amplitude in the power flow assessment framework of real injection. In each representative ice disaster fault scenario, the marginal mitigation contribution of each time period during the day is aggregated to obtain the resilience index of the node under global ice disaster uncertainty. Based on the structural characteristics of the distribution network topology graph, the resilience index is used as the supervision label of the node. The topology-aware learning model is used to evaluate the comprehensive contribution of each node to load loss and network loss improvement, and a set of mobile energy storage candidate access nodes with controlled scale is selected as prior information for random scheduling. The intelligent scheduling module is used to establish a pre-disaster stochastic scheduling model that includes the location of mobile energy storage, travel path and charging and discharging plan under the combined constraints of the integrated power grid-road network and the prior knowledge of candidate access nodes. Under multiple ice disaster scenarios, the model is solved with the objective of minimizing the weighted sum of load loss and network loss to obtain the layout scheme of mobile energy storage in the distribution network.

9. An electronic device, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, wherein when the programs are executed by the processors, they implement the steps of the real-time scheduling method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps of the real-time scheduling method for mobile energy storage distribution networks driven by topology-aware priors under ice disaster resilience as described in any one of claims 1-7.

Images