Power system pre-disaster configuration method, system and device fusing multi-modal data

Abstract

The application relates to a power system pre-disaster configuration method, system and equipment fusing multi-modal data. The method comprises the following steps: acquiring multi-modal data; performing cross-modal fusion on the multi-modal data to obtain high-dimensional fusion features; inputting the high-dimensional fusion features into a pre-constructed differentiable disaster penetration gradient model, and obtaining disaster penetration data, line capacity attenuation results and differentiable disaster penetration loss results through the differentiable disaster penetration gradient model; determining multi-objective optimization indexes based on the disaster penetration data and the high-dimensional fusion features; determining power constraint conditions according to the differentiable disaster penetration loss results; optimizing power system decision variables according to the multi-objective optimization indexes and the power constraint conditions to obtain target decision variables of the power system, and the target decision variables are used for indicating pre-disaster configuration results of the power system. Thus, the method can improve the system resilience of the pre-disaster configuration of the power system in the face of disasters.

Description

Technical Field

[0001] The present application relates to the technical field of power system disaster prevention, and in particular to a method, system, and device for pre-disaster configuration of a power system integrating multimodal data. Background Art

[0002] The stability of power systems is crucial for ensuring our daily lives and work. However, the vulnerability of power systems to extreme disasters, such as typhoons and earthquakes, is becoming increasingly prominent. To enhance power system resilience, recent research has focused on combining multimodal data fusion with multi-objective optimization techniques.

[0003] In related technologies, feature splicing or fixed weight weighting is usually used to integrate multimodal data, and the disaster diffusion model is used independently of resource allocation optimization to achieve disaster simulation and resource allocation separately. The computational complexity of multi-objective optimization is high and the results are dispersed.

[0004] Based on this, the applicant discovered during the implementation process that the relevant technology at least had the problem of insufficient resilience of the power system in complex disaster scenarios. Summary of the Invention

[0005] Based on this, the purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defect of insufficient resilience of power systems to disasters in the existing technology. This application provides a pre-disaster configuration method, system and equipment for power systems that integrate multimodal data.

[0006] In a first aspect, the present application provides a method for pre-disaster configuration of a power system by integrating multimodal data, the method comprising:

[0007] Acquire multimodal data, which includes at least geographic data, meteorological data, power grid data, and economic data;

[0008] Perform cross-modal fusion on multimodal data to obtain high-dimensional fusion features;

[0009] The high-dimensional fusion features are input into the pre-built differentiable disaster penetration gradient model. Through the differentiable disaster penetration gradient model, disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results are obtained.

[0010] Determine multi-objective optimization indicators based on disaster penetration data and high-dimensional fusion features; and determine power constraints based on the results of differentiable disaster penetration losses;

[0011] According to the multi-objective optimization indicators and power constraints, the power system decision variables are optimized to obtain the target decision variables of the power system. The target decision variables are used to indicate the pre-disaster configuration results of the power system.

[0012] In one embodiment, cross-modal fusion is performed on multimodal data to obtain high-dimensional fusion features, including:

[0013] Perform spatiotemporal registration processing on each modal data in the multimodal data to obtain multi-dimensional spatiotemporal coupling tensor data;

[0014] The spatiotemporal coupling tensor data is weightedly fused to obtain high-dimensional fusion features.

[0015] In one embodiment, weighted fusion processing is performed on the spatiotemporal coupling tensor data to obtain high-dimensional fusion features, including:

[0016] Encode each modal vector contained in the spatiotemporal coupling tensor data to obtain the query vector, key vector, and value vector corresponding to each modal vector;

[0017] Determine the attention weight corresponding to each modal vector based on the query vector and the corresponding key vector corresponding to each modal vector;

[0018] Based on the attention weight corresponding to each modal vector, the value vector corresponding to each modal vector is weightedly fused to obtain a high-dimensional fusion feature.

[0019] In one embodiment, disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results are obtained through a differentiable disaster penetration gradient model, including:

[0020] Based on multimodal data, determining the initial value of disaster penetration of the current power system node and the initial value of disaster penetration of the adjacent power system node; the adjacent power system node is directly connected to the current power system node in the power grid diagram;

[0021] Through the differentiable disaster penetration gradient model, the disaster penetration initial value of the current power system node and the disaster penetration initial value of the adjacent power system node are convoluted and iterated to obtain the disaster penetration data of the current power system node.

[0022] In one embodiment, disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results are obtained through a differentiable disaster penetration gradient model, including:

[0023] Determine the line capacity attenuation result based on the disaster penetration data of the current power system node, the disaster penetration data of the adjacent power system nodes, and the preset line capacity;

[0024] Through the differentiable disaster penetration gradient model, the differentiable disaster diffusion coefficient is obtained according to the high-dimensional fusion characteristics;

[0025] The differentiable disaster penetration loss results are determined based on the differentiable disaster diffusion coefficient, the disaster penetration data of the current power system node, and the disaster penetration data of the adjacent power system nodes.

[0026] In one embodiment, the power system decision variables are optimized based on the multi-objective optimization index and the power constraint conditions to obtain the target decision variables of the power system, including:

[0027] According to the multi-objective optimization index and power constraints, the power system decision variables are optimized to obtain multiple initial decision variables;

[0028] Perform Pareto front compression on multiple initial decision variables to obtain multiple candidate decision variables;

[0029] Based on multiple candidate decision variables, the target decision variable is obtained.

[0030] In one embodiment, a target decision variable is obtained based on a plurality of candidate decision variables, including:

[0031] Determine spatial constraints based on the results of microscopic hazard infiltration losses;

[0032] Based on electrical constraints, economic constraints, and space constraints, a target decision variable is screened out from multiple candidate decision variables.

[0033] In a second aspect, the present application provides a pre-disaster configuration system for a power system integrating multimodal data, the system comprising:

[0034] A multimodal acquisition module is used to acquire multimodal data, where the multimodal data includes at least geographic data, meteorological data, power grid data, and economic data;

[0035] Multimodal fusion module, used to perform cross-modal fusion of multimodal data to obtain high-dimensional fusion features;

[0036] The disaster data determination module is used to input high-dimensional fusion features into a pre-built differentiable disaster penetration gradient model. Through the differentiable disaster penetration gradient model, disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results are obtained;

[0037] The indicator constraint determination module is used to determine multi-objective optimization indicators based on disaster penetration data and high-dimensional fusion features; and to determine power constraints based on the results of differentiable disaster penetration losses;

[0038] The pre-disaster configuration determination module is used to optimize the power system decision variables based on multi-objective optimization indicators and power constraints to obtain the target decision variables of the power system. The target decision variables are used to indicate the pre-disaster configuration results of the power system.

[0039] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0040] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when the computer program is executed by a processor.

[0041] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0042] The present application provides a method, system, and equipment for pre-disaster configuration of a power system that integrates multimodal data. The method, system, and equipment construct high-dimensional fusion features through multi-source data that is cross-modally integrated with multimodal data including geographic data, meteorological data, power grid data, and economic data, and combines a differentiable disaster penetration gradient model to achieve collaborative multi-objective optimization of disaster dynamic simulation and power system decision-making resource allocation. This allows the use of cross-modal fusion and multi-objective optimization to achieve efficient deployment of distributed energy in complex and uncertain environments and improve the system resilience of the pre-disaster configuration of the power system in the face of disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0044] Figure 1 A flowchart of a method for pre-disaster configuration of a power system integrating multimodal data provided in an embodiment of the present application;

[0045] Figure 2 A schematic diagram of a process for determining high-dimensional fusion features provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of a process for determining high-dimensional fusion features provided in an embodiment of the present application;

[0047] Figure 4 A schematic diagram of a process for determining high-dimensional fusion features provided in an embodiment of the present application;

[0048] Figure 5 A schematic diagram of a process for determining high-dimensional fusion features provided in an embodiment of the present application;

[0049] Figure 6A schematic diagram of a process for inputting multi-source data into a final configuration strategy provided in an embodiment of the present application;

[0050] Figure 7 A schematic diagram of the structure of a power system pre-disaster configuration system integrating multimodal data provided in an embodiment of the present application;

[0051] Figure 8 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0053] Existing methods for pre-disaster power system configuration typically integrate multimodal data using feature splicing or fixed weighting. Disaster diffusion models operate independently from the resource allocation optimization process, and multi-objective optimization algorithms suffer from high computational complexity and dispersed solution sets. For example, traditional methods are unable to dynamically adjust the weights for integrating meteorological and geographic data in typhoon disaster scenarios. The disconnect between disaster simulation results and resource allocation plans leads to planning delays, and multi-objective optimization results struggle to meet actual engineering needs.

[0054] To address the above issues, this application proposes a method, system, and equipment for pre-disaster configuration of power systems that integrate multimodal data, addressing the three core flaws of existing technologies: rigid multimodal data fusion, disconnection between disaster simulation and optimization, and low multi-objective search efficiency. To address the multimodal fusion issue, a spatiotemporal attention mechanism is proposed to dynamically establish inter-modal correlations. To address the disconnect between disaster simulation and optimization, a differentiable disaster penetration gradient model is designed to achieve end-to-end joint optimization. To address the multi-objective optimization efficiency issue, a Pareto frontier compression algorithm is developed to focus on high-value solution spaces. These three technological breakthroughs create a synergistic effect, jointly enhancing the power system's disaster response capabilities.

[0055] In an exemplary embodiment, Figure 1 A flowchart of a method for pre-disaster configuration of a power system integrating multimodal data provided in an embodiment of the present application is shown in FIG. Figure 1As shown, a power system pre-disaster configuration method fusing multi-modal data is provided. The method is exemplarily applied to a terminal such as a computer device. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and can be realized through interaction of the terminal and the server. The terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers. The server can be a stand-alone physical server, or a server cluster or distributed system formed by multiple physical servers, or a cloud server providing cloud computing services. In the embodiment, the method includes the following S101 to S104. Wherein:

[0056] S101, acquiring multi-modal data, the multi-modal data at least including geographic data, meteorological data, power grid data, economic data.

[0057] Wherein, the multi-modal data can be multi-source heterogeneous data containing geographic spatial information, meteorological evolution trend, power grid topology structure, economic cost parameter, for example, satellite remote sensing image, meteorological forecast model, SCADA system, cost database can be used for collection, and multi-dimensional data support is provided for disaster propagation modeling.

[0058] The geographic data can include terrain, topography, elevation, water system, geological structure, key infrastructure location and the like. The geographic data can be used to assess the probability, path and intensity of disasters (such as floods, landslides, fires). The meteorological data can include historical and forecasted wind speed, wind direction, rainfall, temperature, humidity and the like. The meteorological data can be used to predict disaster weather events and their impact. The power grid data can include power network topology (line, substation, generator, load node connection relationship), equipment parameter, real-time / historical operation state and the like. The power grid data can describe the physical structure and state of the power system itself. The economic data can include user types (residents, businesses, industries), load importance level, outage loss cost estimation and the like. The economic data can be used to quantify the economic losses caused by disasters and optimize the economy of resource allocation.

[0059] Exemplarily, geographic spatial data, meteorological observation data, power grid topology data and economic operation data can be collected from GIS platform, meteorological API, power dispatch center and market database respectively.

[0060] Optionally, the geographic data provides static parameters such as terrain elevation and vegetation coverage that affect disaster propagation. The meteorological data contains dynamic disaster factors such as wind speed and rainfall. The power grid data reflects line capacity and topology connection relationship. The economic data involves equipment cost and outage loss.

[0061] S102, cross-modal fusion is performed on the multi-modal data to obtain high-dimensional fusion features.

[0062] Cross-modal fusion refers to the dynamic association of data from different modalities through spatiotemporal attention mechanisms. For example, this can be achieved using spatiotemporal coupled tensor decomposition and a multi-head attention mechanism, addressing the inability of traditional fixed-weight fusion to adapt to the dynamic changes in disaster scenarios. High-dimensional fusion features refer to feature vectors or feature maps with high dimensionality (number of features) obtained through cross-modal fusion. These features integrate information on geographic environmental risks, meteorological threat intensity, power grid topology vulnerability, and the severity of economic consequences.

[0063] For example, the terminal can dynamically weighted fuse multimodal data through the spatiotemporal attention mechanism to generate high-dimensional fusion features containing disaster propagation paths and power grid status, which can be subsequently used to calculate disaster diffusion parameters or economic penalty coefficients.

[0064] S103. Input the high-dimensional fusion features into a pre-built differentiable disaster penetration gradient model, and obtain disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results through the differentiable disaster penetration gradient model.

[0065] Among them, the differentiable disaster penetration gradient model may refer to modeling the disaster diffusion process as a differentiable graph convolution operation. For example, it can be implemented using the adjacent node penetration gradient propagation algorithm, so that the disaster simulation results can guide resource allocation optimization through gradient back propagation.

[0066] Disaster penetration data may refer to the specific results output by the model simulation on how the disaster penetrates and spreads and its impact on the spatial distribution of power facilities. It may include spatiotemporal information such as the list of affected equipment, geographical location, impact start time, duration, intensity level, etc. Line capacity attenuation results may refer to the degree of damage caused by the disaster to the transmission capacity of a specific transmission line or distribution line as calculated by the model, which can usually be expressed as a percentage of the original capacity. Differentiable disaster penetration loss results may refer to the estimated value of total economic losses calculated by the model due to line capacity attenuation caused by disaster penetration. This loss value may be a differentiable function with respect to the model input (especially the grid decision variables), so that its gradient can be directly used to optimize grid decisions.

[0067] Exemplarily, the terminal can input high-dimensional fusion features into a pre-built differentiable disaster penetration gradient model. Through the differentiable disaster penetration gradient model, the dynamic propagation process of the disaster in space is simulated to generate disaster penetration data. By evaluating the degree to which each line on the propagation path is affected by the disaster, the line capacity attenuation result is generated. By comprehensively considering line attenuation, load value, economic parameters, etc., the differentiable total expected economic loss is calculated to obtain the differentiable disaster penetration loss result.

[0068] Optionally, the fused features are input into a differentiable disaster penetration gradient model, graph convolution operations are performed based on the power grid topology, the disaster penetration intensity of each node is iteratively calculated, and the line capacity attenuation degree and the differentiable disaster loss gradient signal are synchronously output.

[0069] S104. Determine multi-objective optimization indicators based on disaster penetration data and high-dimensional fusion features; and determine power constraints based on the results of differentiable disaster penetration losses.

[0070] The multi-objective optimization index can refer to a comprehensive evaluation system that considers economic costs, power supply reliability, and expected disaster losses. For example, it can be constructed using linear weighting or constraint transformation methods to ensure that the optimization direction meets actual project requirements. Power constraints can refer to the physical and operational power constraints that need to be met during the optimization process.

[0071] For example, by analyzing disaster penetration data (to understand the scope and extent of the disaster's impact) and combining it with information from fused features, such as economic data and grid topology, we can define objective functions that need to be simultaneously minimized / maximized in the optimization problem, such as minimizing total cost and minimizing power outages on critical loads. This allows us to determine multi-objective optimization metrics. Differentiable disaster penetration loss results and underlying grid data can be used to derive the physical and operational constraints that must be met during the optimization process, such as power flow equations and voltage limits for the post-disaster network state.

[0072] S105. Optimize the power system decision variables according to the multi-objective optimization index and the power constraint conditions to obtain the target decision variables of the power system. The target decision variables are used to indicate the pre-disaster configuration result of the power system.

[0073] Power system decision variables can refer to system parameters that can be adjusted or configured before a disaster and require optimization. These power system decision variables can serve as the "independent variables" of the optimization problem. The target decision variables can refer to the specific set of optimal (or Pareto optimal) values ​​of the power system decision variables ultimately obtained through optimization.

[0074] For example, a suitable multi-objective optimization algorithm (such as the gradient-based optimizer NSGA-II, MOEA / D, or a gradient descent method using the gradient of the loss function) can be used to search for values ​​of the power system decision variables that can achieve the best trade-off (Pareto optimality) among multiple optimization objectives while satisfying all power constraints.

[0075] Alternatively, a multi-objective function encompassing economics, reliability, and safety can be constructed in conjunction with disaster penetration data, using differentiable loss gradients to generate constraints such as line capacity and node penetration thresholds. Ultimately, a Pareto front compression algorithm is used to search for optimal configuration solutions that meet these constraints in the solution space, achieving optimal resource allocation in disaster scenarios.

[0076] In this embodiment, it is possible to dynamically adapt to the multimodal data fusion needs under different disaster scenarios, realize real-time collaborative optimization of the disaster propagation process and resource allocation plan, significantly reduce the computational complexity while ensuring the diversity of the solution set, and ultimately generate a pre-disaster configuration plan that takes into account both economy and safety, thereby improving the system resilience of the power system's pre-disaster configuration in the face of disasters.

[0077] In an exemplary embodiment, cross-modal fusion is performed on multimodal data to obtain high-dimensional fusion features, including:

[0078] Perform spatiotemporal registration processing on each modal data in the multimodal data to obtain multi-dimensional spatiotemporal coupling tensor data;

[0079] The spatiotemporal coupling tensor data is weightedly fused to obtain high-dimensional fusion features.

[0080] Among them, spatiotemporal registration processing refers to the operation of aligning the spatiotemporal dimensions of different modal data. For example, it can be achieved by using downsampling, cubic spline interpolation, Z-score normalization and outlier removal methods. By unifying the geographic grid units and time slice benchmarks, the spatiotemporal resolution differences between modalities can be eliminated.

[0081] Among them, weighted fusion processing refers to the operation of dynamically adjusting the fusion weight according to the nonlinear relationship between modalities. For example, it can be achieved by using the attention mechanism to generate dynamic weight coefficients, and by encoding the correlation between modal vectors, capturing the coupling effect of disaster impact in multimodal data.

[0082] For example, the terrain elevation data obtained from the geographic information system and the wind speed data collected by the meteorological observation station may have inconsistent spatial resolutions. By downsampling, the geographic data is gridded to the same 1km×1km resolution as the meteorological data. Then, cubic spline interpolation is performed on the line load information in the power grid topology data to complete the time slice to the minute level. At the same time, the electricity price fluctuations in the economic data are normalized by Z-score, and finally a four-dimensional tensor structure with a unified time and space reference is formed. In the weighted fusion stage, by calculating the attention weights between the modal vectors, for example, geographic data is given a higher weight in the typhoon path area, and the economic data is given an enhanced weight in the load center area, so as to achieve adaptive fusion of modal features.

[0083] Optionally, the resolution difference can be resolved by downsampling, cubic spline interpolation is used to fill in the missing time series, Z-score normalization is applied to eliminate the dimension effect, and after removing the outliers, the spatiotemporal registration algorithm is used to ensure that all modal data are strictly aligned on the grid unit and time slice. For example, it can be shown in Table 1 below; the final output contains a four-channel feature tensor with dimensions such as terrain elevation, wind speed irradiation, line load, and electricity price demand, forming a spatiotemporal coupled tensor structure , which can provide standardized input for subsequent modules.

[0084] Table 1 Summary of multimodal data processing methods

[0085]

[0086] In this embodiment, by performing spatiotemporal registration on multimodal data, the problem of feature correlation breakage caused by inconsistent spatiotemporal benchmarks in multimodal data can be solved, so that the fused high-dimensional features can accurately characterize the cross-modal impact of disasters on the power system, thereby improving the system resilience of the power system's pre-disaster configuration in the face of disasters.

[0087] In an exemplary embodiment, Figure 2 A flowchart of a step for determining high-dimensional fusion features provided in an embodiment of the present application is shown as follows: Figure 2 As shown, it can be Figure 1 On the basis of the above, the steps of the method for pre-disaster configuration of a power system by fusing multimodal data are exemplarily described. In step S102, cross-modal fusion of multimodal data to obtain high-dimensional fusion features may include weighted fusion processing of spatiotemporal coupling tensor data to obtain high-dimensional fusion features. For example, the steps may include S201 to S203, wherein:

[0088] S201 . Encode each modal vector contained in the spatiotemporal coupling tensor data to obtain a query vector, a key vector, and a value vector corresponding to each modal vector.

[0089] S202. Determine the attention weight corresponding to each modal vector based on the query vector and the corresponding key vector corresponding to each modal vector.

[0090] S203. Based on the attention weights corresponding to the modal vectors, weighted fusion is performed on the value vectors corresponding to the modal vectors to obtain high-dimensional fusion features.

[0091] Encoding processing can refer to the nonlinear transformation of each modality vector into a query vector, a key vector, and a value vector. For example, this can be achieved using a multilayer perceptron, where the mapping of modality vectors to the three vectors is achieved through a neural network layer with learnable parameters. The query vector can refer to the feature vector of the "requirements" or "focus" of the current modality. The key vector can refer to the feature vector of the "retrievable information" of the current modality. The value vector can refer to the feature vector of the original semantic information of the current modality.

[0092] The attention weight can refer to a measure of the strength of association between vectors of different modalities. For example, this can be achieved by performing a dot product operation on the query vector and the key vector combined with normalization, dynamically reflecting the importance differences in feature interactions between modalities. Weighted fusion can refer to the linear combination of value vectors based on the attention weights. For example, this can be achieved using an element-by-element multiplication and accumulation approach, strengthening the feature contribution of key modalities through weighted distribution.

[0093] For example, each modal vector in the spatiotemporal coupling tensor can be input into three independent multi-layer perceptrons to generate corresponding query vectors, key vectors, and value vectors respectively. For the target modal vector, the similarity between its query vector and the key vectors of other modalities is calculated, and converted into an attention weight in the form of a probability distribution through a normalized exponential function. This weight characterizes the correlation strength of different modalities in the current spatiotemporal scenario, and is dynamically adjusted through differentiable operations. Subsequently, the value vector of each modality is weightedly summed with the corresponding attention weight to generate a fused high-dimensional feature. In order to further enhance the feature expression capability, a multi-head attention mechanism can be used to execute multiple sets of linear transformations in parallel, and the attention results of different subspaces can be spliced ​​to form the final fused feature.

[0094] Optionally, perform deep fusion on the above four modal data, and use cross-modal attention to mine associations from the spatial, temporal and semantic levels to obtain a unified high-dimensional fusion feature. This process reflects the innovation of cross-modal dynamic coupling: it no longer simply splices multimodal information, but uses learnable attention weights to highlight key modalities and weaken secondary modalities.

[0095] Here, the input information is the feature tensor obtained in the input layer. Assume that each Multimodal vector of Encoded (query, key, value):

[0096] (1);

[0097] In an attention head, the attention weight of (b, t) to (b', t') is:

[0098] (2);

[0099] in Can be defined as The neighborhood in space and time. Then the value vector Weighted accumulation:

[0100] (3);

[0101] Multi-head attention is to transform several different linear transformations Execute in parallel and then perform dynamic splicing. That is the node ,time The fused features at the ST-SJE output layer can be used to calculate disaster diffusion parameters or economic penalty coefficients.

[0102] In this embodiment, a dynamic interaction model between modalities is established through a learnable attention mechanism, and the feature fusion weights are adaptively adjusted in the spatiotemporal dimensions, which can accurately reflect the cross-modal coupling effects in scenarios such as typhoon path deviation or geological mutations. The multi-head attention design further enhances the model's expressive power in different feature subspaces, overcoming the local optimality problem that may exist in single-head attention. It can automatically enhance the correlation strength between meteorological data and geographic data based on the direction of disaster propagation, and accurately capture the synergistic effects of sudden temperature drops and wind speed changes in equipment icing prediction scenarios. This can improve the system resilience of the power system's pre-disaster configuration in the face of disasters.

[0103] In an exemplary embodiment, Figure 3 A flowchart of a step for determining high-dimensional fusion features provided in an embodiment of the present application is shown as follows: Figure 3 As shown, it can be Figure 1 Based on the above, the steps of the method for pre-disaster configuration of a power system integrating multimodal data are exemplarily described. In step S103, disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results are obtained through a differentiable disaster penetration gradient model. For example, the method may include S301 to S302, wherein:

[0104] S301: Determine, based on multimodal data, an initial value of disaster penetration of a current power system node and an initial value of disaster penetration of an adjacent power system node; the adjacent power system node is directly connected to the current power system node in a power grid diagram;

[0105] S302 , using a differentiable disaster penetration gradient model, perform convolution iteration on the disaster penetration initial value of the current power system node and the disaster penetration initial value of the adjacent power system node to obtain the disaster penetration data of the current power system node.

[0106] The current power system node may refer to a node in the power grid topology whose disaster penetration status is to be calculated. The adjacent power system node may refer to a node in the power grid topology that is directly connected to the current power system node via a line (edge).

[0107] The differentiable disaster penetration gradient model refers to a disaster propagation computational framework capable of updating parameters via a backpropagation algorithm. For example, it can be implemented using a graph neural network architecture, where convolution kernel weights are tied to the physical properties of power grid lines. This model leverages its differentiability to enable end-to-end joint training of the disaster propagation process and resource allocation optimization.

[0108] Convolution iteration can refer to a dynamic update process based on the states of adjacent nodes. For example, it can be implemented using matrix multiplication of the adjacency matrix and the node states combined with a nonlinear activation function. This iterative process can simulate the gradient propagation characteristics of disasters along the power grid topology, quantifying the disaster impact on adjacent nodes as the penetration increment of the current node. The power grid diagram can refer to a topological structure diagram that reflects the connection relationships between power system nodes. For example, the adjacency list data structure from graph theory can be used to store node connection information. The topological characteristics of the power grid diagram provide a physical constraint foundation for modeling disaster propagation paths.

[0109] For example, the initial disaster penetration value (initial disaster penetration value) of the current node can be calculated using the altitude information in the geographic data and the wind speed information in the meteorological data. At the same time, based on the connection relationship in the power grid data, the set of adjacent nodes directly connected to the node is obtained. After the initial penetration value of each node is input into the differentiable model, a convolution operation is performed to extract the state characteristics of the adjacent nodes, such as superimposing the penetration value of the adjacent node to the current node by weighted summation. The weighting coefficient is dynamically determined by power grid parameters such as line impedance and transmission capacity. During the iterative process, a corrected linear unit is used to limit the reasonable range of the penetration value to prevent numerical overflow. After multiple iterations, the penetration value of the current node converges to a stable state. At this time, the output disaster penetration data contains the propagation characteristics associated with the power grid topology.

[0110] Alternatively, let Representation node At the moment The disaster penetration value under . Adjacent convolution iteration can be used:

[0111] (4);

[0112] in, is the node-level disaster penetration value, is the disaster penetration value after iteration, is the disaster penetration value before iteration.

[0113] Expression (4) can be rewritten as a differentiable operator (such as vector operation + approximate boundary), which can be updated during back propagation , when a node When it increases, it may "infect" adjacent nodes ; can also be at different times The coupling between them causes the disaster to spread over time.

[0114] In this embodiment, by constructing a convolution iteration mechanism under the constraints of the power grid topology, the state changes of adjacent nodes are incorporated into the calculation process of the current node, so that the disaster penetration process can accurately reflect the physical connection characteristics of the power system. Through the above technical solution, the problem of the disconnection between disaster propagation path modeling and the physical structure of the power grid can be solved. By quantifying the node connection relationship into learnable convolution kernel parameters, a dynamic simulation of the propagation of disasters along the power grid lines is achieved. In this way, the cascade propagation path of disasters in the power network can be accurately predicted, providing accurate disaster impact assessment data for subsequent resource allocation optimization, thereby improving the system resilience of the pre-disaster configuration of the power system in the face of disasters.

[0115] In an exemplary embodiment, Figure 4 A flowchart of a step for determining high-dimensional fusion features provided in an embodiment of the present application is shown as follows: Figure 4 As shown, it can be Figure 1 and Figure 3 Based on this, the steps of the method for pre-disaster configuration of a power system integrating multimodal data are exemplarily described. In step S103, disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results are obtained through a differentiable disaster penetration gradient model. Specifically, the method may further include S401 to S403, wherein:

[0116] S401. Determine a line capacity attenuation result based on disaster penetration data of a current power system node, disaster penetration data of adjacent power system nodes, and a preset line capacity.

[0117] S402. Through the differentiable disaster penetration gradient model, according to the high-dimensional fusion characteristics, the differentiable disaster diffusion coefficient is obtained.

[0118] S403 : Determine a differentiable disaster penetration loss result based on the differentiable disaster diffusion coefficient, the disaster penetration data of the current power system node, and the disaster penetration data of the adjacent power system nodes.

[0119] Among them, the differentiable disaster diffusion coefficient can refer to a learnable parameter of the disaster propagation rate across nodes in the power grid topology, which can be dynamically generated by high-dimensional fusion features.

[0120] For example, a differentiable disaster penetration gradient model can be used to couple the disaster diffusion process with resource allocation optimization end-to-end. Line capacity attenuation can be calculated based on the disaster penetration data of adjacent nodes. For example, a hyperbolic tangent function can be used to perform a nonlinear transformation on node penetration values ​​to accurately reflect the cumulative impact of disaster propagation on line capacity.

[0121] Multi-layer neural networks can be used to process cross-modal fusion features and generate a disaster diffusion coefficient that dynamically changes over time and space. For example, activation functions can be used to constrain the numerical range of the diffusion coefficient. Furthermore, the diffusion coefficient can be combined with node penetration data to construct a differentiable loss function. For example, a weighted summation of the disaster propagation differences between adjacent nodes and the economic cost gradient can be used to form constraints that can be embedded in optimization algorithms. By maintaining the differentiability of all computational steps, this process allows the gradient information output by the disaster diffusion model to directly guide iterative adjustments to resource allocation decisions.

[0122] Optionally, when the node or The penetration value or Higher, line Available capacity can be attenuated as shown in expressions (6) and (7):

[0123] (6);

[0124] (7);

[0125] in, is the line capacity attenuation result (line capacity after attenuation), is the originally designed line capacity, is the attenuation sensitivity. This function is differentiable, so node decisions can be modified during backpropagation due to disaster distribution or capacity constraints.

[0126] Furthermore, the differentiable disaster penetration loss can be defined as:

[0127] (8);

[0128] in, is the result of the microscopic disaster penetration loss, Represents the difference in disaster diffusion between nodes (corresponding to adjacency convolution or morphological gradient), represents the gradient of the economic cost affected by the disaster. All terms are differentiable and can be solved together with the grid constraints during deep learning or iterative optimization.

[0129] In this embodiment, by establishing a differentiable calculation link, the disaster diffusion coefficient can be dynamically adjusted according to the multimodal data. For example, the diffusion intensity parameter is automatically updated when the meteorological conditions change. At the same time, the line capacity attenuation calculation is converted into a differentiable function, so that the resource allocation decision can reversely affect the modeling process of the disaster propagation path, and realize the two-way coupling of disaster simulation and decision optimization. In this way, the coordinated optimization of the disaster propagation path and the resource allocation strategy can be achieved. The multimodal data is dynamically injected into the disaster diffusion process through the differentiable calculation mechanism, so that the resource allocation plan can respond to the changes in the geographical environment and the fluctuations in meteorological conditions in real time. The differentiable calculation of line capacity attenuation provides a gradient adjustment direction for the optimization algorithm, effectively avoiding the decision lag caused by the model splitting of the traditional method, thereby improving the anti-destruction capability of the power system in extreme disaster scenarios, that is, improving the resilience of the power system in the face of disasters.

[0130] In an exemplary embodiment, Figure 5 A flowchart of a step for determining high-dimensional fusion features provided in an embodiment of the present application is shown as follows: Figure 5 As shown, it can be Figure 1 Based on the above, the steps of the method for pre-disaster configuration of a power system integrating multimodal data are exemplarily described. In step S105, the power system decision variables are optimized according to the multi-objective optimization index and the power constraint conditions to obtain the target decision variables of the power system. For example, the method may include S501 to S503, wherein:

[0131] S501. Optimize power system decision variables according to multi-objective optimization indicators and power constraints to obtain multiple initial decision variables.

[0132] S502: Perform Pareto front compression on multiple initial decision variables to obtain multiple candidate decision variables.

[0133] S503: Obtain a target decision variable based on multiple candidate decision variables.

[0134] Pareto front compression refers to the process of mathematically screening out non-dominated solution sets. For example, this can be achieved by mapping a high-dimensional solution space into a low-dimensional feature space using gradient projection. This reduces the redundancy of the solution set while maintaining solution diversity through dimensionality reduction. Candidate decision variables can be the set of solutions that satisfy the non-dominated condition after Pareto front compression. For example, these can be obtained through a joint screening mechanism that combines electrical, economic, and spatial constraints. The synergy of multi-dimensional constraints ensures the physical feasibility and economic viability of the solution.

[0135] For example, the multi-objective optimization process can first generate an initial set of decision variables covering different objective weights, for example by traversing the solution space using a genetic algorithm or particle swarm optimization algorithm. This initial set of solutions is then subjected to Pareto front compression, for example by calculating the dominance relationship of each solution in the objective function space and eliminating dominated solutions to form a non-dominated solution set, thereby reducing subsequent computational complexity. Finally, candidate decision variables are screened based on spatial constraints and electrical and economic constraints, for example by using the analytic hierarchy process to rank candidate solutions for multi-attribute decision making, and the solution with the highest overall score is selected as the final objective decision variable.

[0136] Optionally, when obtaining the disaster diffusion field and fusion features Finally, it is necessary to consider the resilience index (with related) and economic indicators (related to Related) and load reduction for multi-objective optimization. The power system decision variables can be made into , the node disaster penetration value is The goal can be expressed as (9):

[0137] (9);

[0138] And meet the following power constraints: power balance constraints, line capacity limitations, unit / energy storage / load related constraints (including unit output upper limit, ramp rate, energy storage SOC range, load interruptibility upper limit, etc.)

[0139] When solving, a number of mutually non-dominated solutions can be obtained In order to reduce the redundancy of solutions, this embodiment can introduce Pareto frontier search and its compression technology, and obtain a batch of candidate solutions that can be used for deployment by searching for several optimal solutions that do not dominate each other in the high-dimensional target space.

[0140] Multi-objective optimization problems are often expressed as:

[0141] (10)

[0142] in, represents the candidate solution set, is the number of solutions in the solution set, It measures the dispersion between solutions, is a total loss function that combines disasters and economic goals. and compress , we can get a Pareto solution with compact structure and diverse distribution. θ is the key parameter (or decision variable) that affects the total loss function J, Θ represents the feasible range of θ, Can be a preset small positive threshold that defines acceptable loss sensitivity.

[0143] In order to quickly locate high-value solutions in the solution space, the attention-guided Monte Carlo tree search strategy (AG-MCTS) is introduced:

[0144] (11);

[0145] in is the attention weight based on the fusion feature, The gain (or loss reduction) brought by the action, is the number of times the action is sampled. AG-MCTS can focus on the most promising search path during the sampling process, The reflected disaster risk areas and multimodal information are combined to reduce blind exploration.

[0146] After the multi-objective search is completed, a batch of Pareto optimal solutions are obtained , where each solution It corresponds to a compromise configuration between toughness and economy.

[0147] In this embodiment, Pareto front compression is used to effectively reduce the size of the solution set. A multi-constraint joint screening mechanism is also employed to ensure convergence of solutions, thus avoiding the computational burden of high-dimensional optimization problems. This reduces computational resource consumption during multi-objective optimization, addresses the issue of dispersed optimization results, and improves the efficiency and reliability of pre-disaster configuration decisions for power systems, thereby enhancing their resilience.

[0148] In an exemplary embodiment, a target decision variable is obtained based on a plurality of candidate decision variables, including:

[0149] Determine spatial constraints based on the results of microscopic hazard infiltration losses;

[0150] Based on electrical constraints, economic constraints, and space constraints, a target decision variable is screened out from multiple candidate decision variables.

[0151] Among them, spatial constraints can refer to the quantitative constraint boundaries generated by disaster penetration loss results. For example, this can be achieved through analysis of geographic spatial penetration gradient characteristics to limit the deployment location of candidate solutions along the disaster diffusion path. Electrical constraints can refer to the compliance requirements of the grid topology and operating parameters. For example, node voltage limits and line capacity threshold verification can be used to ensure the stable operation of the power system. Economic constraints can refer to economic indicators of resource allocation. For example, levelized cost of electricity generation or initial capital investment cost assessment can be used to control the budget range of pre-disaster configuration.

[0152] For example, after the Pareto front compression is completed, there may be a large number of solutions in the candidate decision variable set that meet the multi-objective optimization but do not meet the actual engineering needs. By introducing spatial constraints, the geographic spatial risk characteristics output by the disaster penetration gradient model can be converted into computable deployment constraints, such as prohibiting the deployment of key equipment in areas with a high probability of disaster spread. At the same time, the voltage stability and line capacity attenuation rate of the grid nodes are verified in combination with electrical constraints, and the equipment investment cost and operation and maintenance costs are screened using economic constraints. Finally, the solutions that do not meet any of the constraints are eliminated from the candidate solution set. This multi-dimensional joint screening mechanism effectively narrows the range of feasible solutions by quantifying the spatial risk of disasters and the actual limitations of engineering.

[0153] Optionally, after completing the Pareto front search, a "feasible solution" that meets a specific disaster risk threshold and economic budget is selected or filtered from the Pareto solution set. The corresponding information, such as device type (PV, energy storage, etc.), deployment location, access node, and capacity, is combined with the configuration constraints in the table to output the final device deployment strategy. Table 2 below lists exemplary deployment rules.

[0154] Table 2 Example deployment rules

[0155]

[0156] Among them, DHPG is a differentiable disaster diffusion gradient indicator, which is used to quantify the degree of diffusion of natural disasters in the target area; The per-unit value (PUV) of voltage is often used to standardize electrical quantities at different rated voltage levels. LCOE refers to the levelized cost of electricity, which integrates factors such as equipment investment, operation and maintenance, and fuel costs, facilitating comparisons between different power generation methods. SOC (State of Charge) represents the state of charge of the energy storage device, with a value range of [0, 1] or [0, 100%]. CAPEX (Capital Expenditure) refers to the initial capital investment and is often used to evaluate the fixed costs of equipment or projects.

[0157] In this embodiment, by introducing spatial constraints and dynamically linking disaster diffusion paths with equipment deployment locations, the screening process can simultaneously consider disaster risk distribution and project feasibility, improving the spatial rationality and disaster resilience of pre-disaster configuration plans. This enables rapid identification of power system configuration plans that combine disaster resilience, operational stability, and economic efficiency in complex disaster scenarios, significantly improving decision-making efficiency and plan feasibility, thereby enhancing the resilience of pre-disaster power system configurations in the face of disasters.

[0158] In an exemplary embodiment, Figure 6 As shown, Figure 6A schematic diagram of a process for inputting multi-source data into a final configuration strategy is provided in an embodiment of the present application. This process step is applied to a pre-disaster configuration method for a power system that integrates multimodal data. For example, it may include:

[0159] S601. Input multi-source information (Geo (geographic data), Weather (meteorological data), Grid (power grid data), Econ (economic data)).

[0160] S602, input layer: multimodal data feature extraction.

[0161] S603, spatiotemporal-semantic joint encoder: Feature fusion; wherein, a multi-head cross-modal attention mechanism is used to dynamically assign fusion weights for the four modalities of geography, weather, power grid, and economy, replacing traditional static weighting. This can be achieved through the embodiment corresponding to step S102.

[0162] S604: Differentiable disaster penetration gradient model. Based on node-level differentiable adjacency convolution and dynamic coefficient generation, the disaster diffusion process and economic constraints are unified into a differentiable loss function, achieving end-to-end joint optimization of disaster risk and grid operation constraints. This can be achieved through the embodiment corresponding to step S103.

[0163] S605: Pareto front compression search and multi-objective optimization. Attention-guided Monte Carlo tree search (AG-MCTS) can be used to focus on high-risk disaster nodes and economically sensitive periods, compressing the redundant solution set and accelerating the convergence of multi-objective optimization. This can be achieved through the embodiments corresponding to step S103. This can also be achieved through the embodiments corresponding to step S105.

[0164] S606, output layer: output device deployment strategy.

[0165] In practical applications, this embodiment can be used as a specific implementation method corresponding to the above embodiments. Therefore, the various steps of this embodiment can be implemented correspondingly through the above embodiments, and will not be repeated here.

[0166] In this example, a pre-disaster optimization scheme for power system configuration integrates multimodal data, employs a spatiotemporal attention mechanism, and employs differentiable disaster simulation techniques. This scheme aims to leverage cross-modal fusion and multi-objective optimization to achieve efficient deployment and improved resilience of distributed energy resources in complex and uncertain environments. The scheme consists of three main components: an input layer, a core algorithm layer, and an output layer. The core algorithm layer includes a spatiotemporal-semantic joint encoder (ST-SJE), a differentiable disaster penetration gradient (DHPG), and a Pareto front compression search module. Finally, the output layer formulates a device deployment strategy based on engineering constraints.

[0167] Optionally, multimodal data is first collected and preprocessed through the input layer, and then spatiotemporal attention fusion, differentiable disaster simulation and multi-target search are performed in sequence in the core algorithm layer, and finally the optimal solution is output to the deployment strategy.

[0168] This example proposes a comprehensive technical approach for pre-disaster optimization of power system configuration by leveraging a cross-modal attention mechanism and a differentiable disaster simulation method. This solution not only demonstrates advantages in integrating and understanding multi-source data, but also effectively balances economic benefits and system resilience through multi-objective Pareto optimization, forming a high-value, implementable strategy.

[0169] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0170] The following describes the pre-disaster configuration system for an electric power system that integrates multimodal data provided in an embodiment of the present application. The pre-disaster configuration system for an electric power system that integrates multimodal data has the same inventive concept as the pre-disaster configuration method for an electric power system that integrates multimodal data described above. The implementation solution for solving the problem provided by the system is similar to the implementation solution recorded in the above method. Therefore, the specific limitations in one or more embodiments of the pre-disaster configuration system for an electric power system that integrates multimodal data provided below can refer to the limitations of the pre-disaster configuration method for an electric power system that integrates multimodal data described above. The pre-disaster configuration system for an electric power system that integrates multimodal data described below and the pre-disaster configuration method for an electric power system that integrates multimodal data described above can be referred to each other in correspondence and will not be repeated here.

[0171] In an exemplary embodiment, Figure 7 A schematic diagram of the structure of a power system pre-disaster configuration system integrating multimodal data provided in an embodiment of the present application is shown in FIG. Figure 7 As shown, the power system pre-disaster configuration system 70 integrating multimodal data includes: a multimodal acquisition module 710, a multimodal fusion module 720, a disaster data determination module 730, an indicator constraint determination module 740, and an indicator constraint determination module 740, wherein:

[0172] The multimodal acquisition module 710 is used to acquire multimodal data, where the multimodal data includes at least geographic data, meteorological data, power grid data, and economic data.

[0173] The multimodal fusion module 720 is used to perform cross-modal fusion on multimodal data to obtain high-dimensional fusion features.

[0174] The disaster data determination module 730 is used to input the high-dimensional fusion features into the pre-built differentiable disaster penetration gradient model, and obtain disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results through the differentiable disaster penetration gradient model.

[0175] The indicator constraint determination module 740 is used to determine multi-objective optimization indicators based on disaster penetration data and high-dimensional fusion features; and to determine power constraint conditions based on the differentiable disaster penetration loss results.

[0176] The pre-disaster configuration determination module 750 is used to optimize the power system decision variables based on multi-objective optimization indicators and power constraints to obtain the target decision variables of the power system. The target decision variables are used to indicate the pre-disaster configuration results of the power system.

[0177] In an exemplary embodiment, the multimodal fusion module is used to perform spatiotemporal registration processing on each modal data in the multimodal data to obtain multi-dimensional spatiotemporal coupling tensor data; and perform weighted fusion processing on the spatiotemporal coupling tensor data to obtain high-dimensional fusion features.

[0178] In an exemplary embodiment, the multimodal fusion module is used to encode the modal vectors contained in the spatiotemporal coupling tensor data to obtain the query vector, key vector and value vector corresponding to each modal vector; determine the attention weight corresponding to each modal vector based on the query vector and the corresponding key vector corresponding to each modal vector; based on the attention weight corresponding to each modal vector, perform weighted fusion on the value vector corresponding to each modal vector to obtain a high-dimensional fusion feature.

[0179] In an exemplary embodiment, the disaster data determination module is used to determine the disaster penetration initial value of the current power system node and the disaster penetration initial value of the adjacent power system node based on multimodal data; the adjacent power system node is directly connected to the current power system node in the power grid diagram; through the differentiable disaster penetration gradient model, the disaster penetration initial value of the current power system node and the disaster penetration initial value of the adjacent power system node are convolutionally iterated to obtain the disaster penetration data of the current power system node.

[0180] In an exemplary embodiment, the disaster data determination module is used to determine the line capacity attenuation result based on the disaster penetration data of the current power system node, the disaster penetration data of the adjacent power system node, and the preset line capacity; obtain the differentiable disaster diffusion coefficient according to the high-dimensional fusion characteristics through the differentiable disaster penetration gradient model; determine the differentiable disaster penetration loss result according to the differentiable disaster diffusion coefficient, the disaster penetration data of the current power system node, and the disaster penetration data of the adjacent power system node.

[0181] In an exemplary embodiment, the pre-disaster configuration determination module is used to optimize the power system decision variables based on multi-objective optimization indicators and power constraints to obtain multiple initial decision variables; perform Pareto front compression on the multiple initial decision variables to obtain multiple candidate decision variables; and obtain the target decision variable based on the multiple candidate decision variables.

[0182] In an exemplary embodiment, the pre-disaster configuration determination module is used to determine spatial constraints based on the results of microscopic disaster penetration losses; based on electrical constraints, economic constraints, and spatial constraints, a target decision variable is screened out from multiple candidate decision variables.

[0183] In an exemplary embodiment, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors execute the steps of the power system pre-disaster configuration method for integrating multimodal data as described in any of the above embodiments.

[0184] In an exemplary embodiment, the present application also provides a computer device having a computer program stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the method for pre-disaster configuration of a power system integrating multimodal data as described in any of the above embodiments.

[0185] In an exemplary embodiment, the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the method for pre-disaster configuration of a power system by fusing multimodal data as described in any of the above embodiments.

[0186] Schematically, as Figure 8 As shown, Figure 8 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 800 can be provided as a server or a terminal. Figure 8Computer device 800 includes a processing component 802, which further includes one or more processors, and memory resources represented by memory 801 for storing instructions executable by processing component 802, such as application programs. The application programs stored in memory 801 may include one or more modules, each corresponding to a set of instructions. Furthermore, processing component 802 is configured to execute the instructions to perform the method for pre-disaster configuration of a power system integrating multimodal data according to any of the aforementioned embodiments.

[0187] The computer device 800 may further include a power supply component 803 configured to perform power management of the computer device 800, a wired or wireless network interface 804 configured to connect the computer device 800 to a network, and an input / output (I / O) interface 805. The computer device 800 may operate based on an operating system stored in the memory 801, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0188] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0189] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0190] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only 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 terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0191] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0192] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for pre-disaster configuration of a power system integrating multimodal data, characterized in that: The method comprises: Acquiring multimodal data, wherein the multimodal data includes at least geographic data, meteorological data, power grid data, and economic data; Performing cross-modal fusion on the multimodal data to obtain high-dimensional fusion features; Inputting the high-dimensional fusion features into a pre-built differentiable disaster penetration gradient model, and obtaining disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results through the differentiable disaster penetration gradient model; Determining multi-objective optimization indicators based on the disaster penetration data and the high-dimensional fusion features; and determining power constraints based on the differentiable disaster penetration loss results; Optimizing the power system decision variables according to the multi-objective optimization index and the power constraint conditions to obtain target decision variables of the power system, wherein the target decision variables are used to indicate the pre-disaster configuration result of the power system; The disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results obtained by the differentiable disaster penetration gradient model include: Determining, based on the multimodal data, an initial value of disaster penetration of a current power system node and an initial value of disaster penetration of an adjacent power system node; the adjacent power system node is directly connected to the current power system node in a power grid diagram; Performing convolution iteration on the disaster penetration initial value of the current power system node and the disaster penetration initial value of the adjacent power system node through the differentiable disaster penetration gradient model to obtain the disaster penetration data of the current power system node; The disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results obtained by the differentiable disaster penetration gradient model include: Determining the line capacity attenuation result based on the disaster penetration data of the current power system node, the disaster penetration data of the adjacent power system nodes, and the preset line capacity; Obtaining a differentiable disaster diffusion coefficient based on the high-dimensional fusion features through the differentiable disaster penetration gradient model; Determining a differentiable disaster penetration loss result according to the differentiable disaster diffusion coefficient, the disaster penetration data of the current power system node, and the disaster penetration data of the adjacent power system nodes; The step of optimizing the power system decision variables according to the multi-objective optimization index and the power constraint conditions to obtain the target decision variables of the power system includes: Optimizing the power system decision variables according to the multi-objective optimization index and the power constraint conditions to obtain a plurality of initial decision variables; Performing Pareto front compression on the multiple initial decision variables to obtain multiple candidate decision variables; The target decision variable is obtained based on a plurality of the candidate decision variables.

2. The method according to claim 1, characterized in that The cross-modal fusion of the multimodal data to obtain high-dimensional fusion features includes: Performing spatiotemporal registration processing on each modal data in the multimodal data to obtain multi-dimensional spatiotemporal coupling tensor data; The spatiotemporal coupling tensor data is subjected to weighted fusion processing to obtain the high-dimensional fusion feature.

3. The method according to claim 2, characterized in that The weighted fusion processing is performed on the spatiotemporal coupling tensor data to obtain the high-dimensional fusion feature, including: Encoding each modal vector included in the spatiotemporal coupling tensor data to obtain a query vector, a key vector, and a value vector corresponding to each modal vector; Determining the attention weight corresponding to each modal vector according to the query vector and the corresponding key vector corresponding to each modal vector; Based on the attention weights corresponding to the modal vectors, the value vectors corresponding to the modal vectors are weightedly fused to obtain the high-dimensional fusion features.

4. The method according to claim 1, wherein The step of obtaining the target decision variable based on the plurality of candidate decision variables includes: Determining spatial constraints based on the microscopic hazard seepage loss results; The target decision variable is selected from the plurality of candidate decision variables based on the electrical constraint condition, the economic constraint condition, and the space constraint condition.

5. A power system pre-disaster configuration system integrating multimodal data, characterized in that: The system comprises: A multimodal acquisition module, configured to acquire multimodal data, wherein the multimodal data includes at least geographic data, meteorological data, power grid data, and economic data; A multimodal fusion module, used to perform cross-modal fusion on the multimodal data to obtain high-dimensional fusion features; A disaster data determination module is used to input the high-dimensional fusion features into a pre-built differentiable disaster penetration gradient model, and obtain disaster penetration data, line capacity attenuation results, and differentiable disaster penetration loss results through the differentiable disaster penetration gradient model; An indicator constraint determination module is used to determine multi-objective optimization indicators based on the disaster penetration data and the high-dimensional fusion features; and to determine power constraint conditions based on the differentiable disaster penetration loss results; a pre-disaster configuration determination module, configured to optimize the power system decision variables according to the multi-objective optimization index and the power constraint conditions to obtain target decision variables of the power system, wherein the target decision variables are used to indicate the pre-disaster configuration result of the power system; in: The disaster data determination module is used to determine, based on the multimodal data, an initial value of disaster penetration of a current power system node and an initial value of disaster penetration of an adjacent power system node; the adjacent power system node is directly connected to the current power system node in a power grid diagram; and convolutionally iterate the initial value of disaster penetration of the current power system node and the initial value of disaster penetration of the adjacent power system node using the differentiable disaster penetration gradient model to obtain the disaster penetration data of the current power system node; The disaster data determination module is used to determine the line capacity attenuation result based on the disaster penetration data of the current power system node, the disaster penetration data of the adjacent power system node, and the preset line capacity; obtain a differentiable disaster diffusion coefficient based on the high-dimensional fusion feature through the differentiable disaster penetration gradient model; and determine a differentiable disaster penetration loss result based on the differentiable disaster diffusion coefficient, the disaster penetration data of the current power system node, and the disaster penetration data of the adjacent power system node; The pre-disaster configuration determination module is used to optimize the power system decision variables according to the multi-objective optimization indicators and the power constraints to obtain multiple initial decision variables; perform Pareto front compression on the multiple initial decision variables to obtain multiple candidate decision variables; and obtain the target decision variable based on the multiple candidate decision variables.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images