Power transmission network ice disaster toughness optimization method based on remote sensing

Abstract

The invention belongs to the technical field of power transmission network toughness improvement, and particularly relates to a power transmission network ice disaster toughness optimization method based on remote sensing, which comprises the steps of S1, adopting an ice disaster scene generation process to generate a scene set reflecting an actual evolution rule of an ice disaster in a region; s2, constructing a source-network-load multi-dimensional system toughness evaluation index system; s3, constructing a two-stage robust toughness improvement planning model of the power transmission network; S4, solving the two-stage robust planning model in the S3, and outputting an optimal energy storage configuration scheme and an emergency maintenance scheduling strategy; and S5, deploying the energy storage configuration scheme output in the step S4 in actual operation of the power grid, and dynamically adjusting energy storage charging and discharging and unit output according to the emergency maintenance scheduling strategy when an ice disaster occurs, thereby realizing toughness improvement and economic loss minimization of the power transmission network. According to the method, the whole-process toughness improvement and the total cost minimization of the power transmission network under the slow-development ice disaster can be realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power grid resilience improvement, and particularly relates to a power grid ice disaster resilience optimization method based on remote sensing. BACKGROUND

[0002] As a typical extreme weather disaster, ice disaster often leads to icing, dancing and even breaking of power lines and tower collapse, which poses a serious threat to the safe and stable operation of power grids. Unlike sudden natural disasters such as typhoons and earthquakes, ice disasters have the characteristics of long duration, slow development process, and high modulation by micro-meteorological conditions and complex terrain, and the spatio-temporal evolution process presents significant non-uniformity and uncertainty. This characteristic makes the influence of ice disaster on the power system not an instantaneous impact, but a long-term cumulative effect and dynamic coupling with the system state, forming the so-called "relaxation effect" - that is, the system performance gradually deteriorates with the duration of icing, and there is still an intervention window during the disaster process.

[0003] However, the current research on the resilience of power grid ice disaster still has obvious limitations. On the one hand, ice disaster scenario modeling is heavily dependent on historical statistics or simplified assumptions, and lacks a physical-data fusion type time sequence generation method based on high-resolution remote sensing observations that can depict the dynamic expansion and intensity evolution of icing areas. The existing model cannot accurately reflect how key spatial distribution characteristics such as icing thickness and coverage range evolve over time, resulting in a scenario set that is detached from reality and cannot support refined decision-making. On the other hand, existing resilience improvement strategies often adopt a binary and fragmented approach of "pre-disaster prevention" and "post-disaster repair": pre-disaster focuses on reinforcement or redundancy configuration, and post-disaster relies on passive repair. It ignores the "active regulation" opportunity window provided by the "slow development" characteristics of ice disaster - that is, during the duration of the disaster, through flexible scheduling of energy storage, adjustment of unit output and dynamic optimization of maintenance resources, the system performance degradation can be significantly alleviated.

[0004] Therefore, how to realize the whole-process resilience improvement and total cost minimization of the power grid under the slow development of ice disaster has become a problem to be solved. SUMMARY

[0005] In view of the above technical problems of the prior art, the application provides a power grid ice disaster resilience optimization method based on remote sensing, which can realize the whole-process resilience improvement and total cost minimization of the power grid under the slow development of ice disaster.

[0006] In order to solve the above technical problems, the application adopts the following technical solutions:

[0007] A power grid ice disaster resilience optimization method based on remote sensing, comprising the following steps:

[0008] S1, based on the target area of multispectral satellite remote sensing data, using ice disaster scene generation process, generate a set of reflecting the actual evolution law of the region ice disaster scene set, each scene in the scene set contains line ice caused fault state, wind light output loss and load disturbance information, and its occurrence probability is associated;

[0009] S2, construct source-grid-load multi-dimensional system resilience evaluation index system, the index system includes power generation recovery ratio, power transmission recovery ratio and load recovery ratio, and the system comprehensive recovery ratio is obtained by proportionally weighting based on the three indexes, as the resilience loss index Rs, to quantify the loss of system operation performance;

[0010] S3, based on the scene set reflecting the actual ice disaster characteristics of the target area and its occurrence probability generated by S1, combined with the resilience loss index Rs defined by S2, a two-stage robust resilience improvement planning model of power transmission network is constructed:

[0011] The first stage takes the configuration position, capacity and power of the fixed energy storage before the disaster as the decision variable, and minimizes the investment cost;

[0012] The second stage takes the charging and discharging power of the energy storage, the output of the thermal power unit and the maintenance resource scheduling as the decision variable, and minimizes the operation loss under the given ice disaster scene;

[0013] The objective function is to minimize the total cost, including the energy storage configuration cost, the maintenance resource configuration cost and the expected resilience loss item under the worst probability distribution;

[0014] S4, solve the two-stage robust planning model of S3, output the optimal energy storage configuration scheme and emergency maintenance scheduling strategy;

[0015] S5, deploy the energy storage configuration scheme output by S4 in the actual operation of the power grid, and dynamically adjust the charging and discharging of the energy storage and the unit output according to the emergency maintenance scheduling strategy when the ice disaster occurs, to realize the resilience improvement and the minimization of economic loss of the power transmission network.

[0016] Compared with the prior art, the present application has the following beneficial effects:

[0017] 1, realize high-fidelity ice disaster scene generation, significantly improve the authenticity and pertinence of disaster modeling. Unlike the existing ice disaster modeling methods which rely on historical statistics or static assumptions, this scheme is based on multispectral satellite remote sensing data, integrates the dynamic evolution law of icing area and the "source-grid-load" disturbance response mechanism, generates a structured scene containing line fault state, wind light output loss and load disturbance, and gives its occurrence probability. This method breaks through the limitations of traditional scene "separation from physical process, lack of space-time correlation", and provides a more realistic input basis for subsequent resilience evaluation and optimization.

[0018] 2. Construct a quantifiable "source-network-load" multi-dimensional resilience evaluation system to realize the comprehensive characterization of system recovery capability. Existing researches mostly focus on a single dimension (such as power supply recovery time) or qualitatively describe the resilience level. This scheme proposes to take the recovery proportion of three key elements of power generation, power transmission and load as the core indicators, and form a unified resilience loss index Rs through weighted synthesis, so that the system performance degradation degree is quantifiable and comparable, and a clear evaluation target is provided for the optimization model.

[0019] 3. Create a collaborative optimization mechanism of pre-disaster configuration and disaster dispatching to fully tap the regulation potential under the "slow development" characteristics of ice disaster. Unlike the fragmented mode of traditional "pre-disaster prevention + post-disaster repair", this scheme takes advantage of the long duration and relaxation effect of ice disaster, and jointly optimizes the pre-disaster energy storage investment decision and disaster operation dispatching (including energy storage charging and discharging, thermal power output adjustment, and maintenance resource allocation) under the two-stage robust planning framework. This whole-process collaborative strategy effectively connects the "prevention-response-recovery" chain and significantly enhances the active adaptation capability of the system in the disaster process.

[0020] 4. Introduce the expected resilience loss term under the worst probability distribution to improve the robustness and risk resistance of the planning scheme. In view of the high uncertainty of ice disaster scenarios, the model embeds the distribution robust optimization idea in the objective function, considers the most unfavorable but possible probability distribution scenario, and avoids the failure of the scheme due to scenario deviation. Compared with deterministic optimization or simple stochastic programming, this method greatly improves the reliability in dealing with extreme or unforeseen ice disaster scenarios while ensuring economic efficiency.

[0021] In summary, this method can realize the whole-process resilience improvement and total cost minimization of the power transmission network under slow development ice disaster.

[0022] Preferably, in S1, the ice disaster scenario generation process comprises:

[0023] (1) Collect and screen multispectral satellite remote sensing data, and generate high-resolution, continuous time-series ice-covered area images using image fusion technology to improve the spatial distribution identification accuracy of ice-covered areas;

[0024] (2) Based on the continuous time-series ice-covered area images generated in step (1), a spatio-temporal evolution model is constructed to analyze the change law of the ice-covered mode and predict the spatial distribution of the ice-covered area and its dynamic evolution trend in the future period to obtain the ice-covered area prediction result and its spatial distribution characteristics; the spatial distribution characteristics include ice thickness and coverage range;

[0025] (3) Based on the ice-covered area prediction result and its spatial distribution characteristics provided in step (2), the fault state of the power transmission line, the output loss of the wind-solar power generation system, and the disturbance of the power grid load are calculated to obtain structured disturbance data;

[0026] (4) Combine the historical source-grid-load operation data and the structured disturbance data obtained in step (3), learn the joint distribution law thereof by using a probabilistic generation model, and generate an ice disaster scene set conforming to the actual distribution characteristics.

[0027] Such a setting, 1, converts multispectral remote sensing data into high-resolution, continuous time series of icing area images through image fusion technology, significantly improving the ability to capture the boundaries, intensity and expansion trend of icing compared to traditional single sensors or low-frequency observation methods. This provides more accurate and continuous input basis for subsequent dynamic prediction, avoiding modeling bias caused by sparse data.

[0028] 2, Realize the structured quantification of ice disaster impact, support system-level response decision. Based on the icing prediction results, calculate key indicators such as line failure rate, wind and light output loss and load disturbance to form structured disturbance data. This process converts "geospatial information" into "power system observable variables", breaking the semantic gap between remote sensing data and power system models, so that the impact of ice disasters can be directly used for system evaluation and optimization.

[0029] 3, Combine historical "source-grid-load" operation data and structured disturbance data, learn the joint distribution law thereof by using a probabilistic generation model (such as CVAE), and generate an ice disaster scene set with statistical consistency. This method overcomes the distribution distortion problem caused by traditional scene enumeration or simple random sampling, ensuring that the generated scene not only reflects the operation characteristics of the real system, but also covers a variety of possible ice disaster evolution paths, thereby providing high-quality input for resilience optimization.

[0030] Preferably, step (2) said constructing a spatio-temporal evolution model, analyzing the change law of the icing mode, predicting the spatial distribution of the icing area and its dynamic evolution trend in the future period, including:

[0031] A convolutional neural network based on partial differential characteristics is constructed as a spatio-temporal evolution model, which includes a convolutional layer embedding a partial differential operator and a time recursive feedback structure; wherein the convolutional layer embedding the partial differential operator is used to simulate the spatial diffusion characteristics of the icing area, and the time recursive feedback structure is used to input the prediction results of the current period as the input of the next period in the training and prediction stages, realizing multi-step time series prediction;

[0032] Under the rolling prediction training framework, the continuous time series of icing area images generated in step (1) are used as the initial input, and the spatio-temporal evolution model is used to generate prediction images from t+1 to t+c-1 time periods in turn, and the prediction images from t+1 to t+c-1 time periods and the real icing area image of t time period are jointly used as input, and the spatio-temporal evolution model is trained to predict the icing area image of t+c time period;

[0033] The latest continuous time sequence icing area image generated in step (1) is input into the trained spatio-temporal evolution model, and the future c time period icing area prediction results are recursively output through the time recursive feedback structure of the model; the prediction results include the icing thickness and coverage range of each time period.

[0034] Such a setting, 1, fusion of physical laws and data-driven, improves the rationality and accuracy of icing evolution prediction. By embedding partial differential operators in the convolutional neural network, the model can explicitly capture the spatial diffusion behavior of icing (such as gradient propagation, edge expansion, etc.), rather than relying solely on pure data fitting. This "physically guided" modeling approach significantly enhances the model's generalization ability and interpretability, avoiding the "hallucination" predictions that may occur in traditional deep learning models.

[0035] 2, realize multi-step time series prediction, support long-term evolution trend deduction. The introduction of time recursive feedback structure enables the model to maintain memory and recursive ability to historical state in both training and prediction phases. This not only improves the response accuracy of the model to short-term changes, but also enables it to continuously deduce the future long-term icing development trend, meeting the long-term prediction needs in the slow development process of ice disasters.

[0036] 3, adopt rolling prediction training strategy, enhance the adaptability of the model to the real evolution path. In the training process, the prediction image of the first c-1 time period (rather than the true value) is used as input to predict the c time period, simulating the self-recurrence reasoning process in actual application. This approach ensures that the model learns how to handle its own prediction errors during training, thereby improving its robustness and stability during deployment, avoiding the training-inference inconsistency problem caused by "teacher forcing".

[0037] 4, output complete spatial distribution information, support subsequent system-level impact analysis. The model output not only includes the binary mask of the icing area, but also includes the quantitative spatial distribution characteristics such as icing thickness and coverage range at each time period. These structured information can be directly used to calculate the line fault probability, wind and light output loss, and load disturbance, providing high-precision input for "source-grid-load" collaborative resilience assessment.

[0038] Preferably, in step (3), the fault state of the power transmission line is calculated using a failure rate model, and the construction process of the failure rate model includes:

[0039] The predicted icing area is grid processed using a geographic information system, and each icing area is equivalent to a circular area with a radius of the icing area Θ at time t; Θ,t

[0040] According to the distance R between the coordinate point of the power transmission line and the center of the icing area Θ​Θ,(a,b),t The severity of the ice storm, Y, at that point is determined using a piecewise function. Θ,(a,b),t ;

[0041] Based on ice thickness R l,m,t and the maximum anti-icing thickness d of the line l A piecewise exponential function is used to establish the failure rate model of the line within the grid:

[0042]

[0043] In the formula, p l,m,t Let be the failure rate of line l in the m-th grid region at time t.

[0044] This setup, 1) transforms continuous icing areas into regular geographic units through GIS gridding, precisely matching them with the coordinates of transmission lines, thus avoiding errors caused by "global thresholds" or "coarse-grained division" in traditional methods. This refined spatial alignment mechanism allows line fault risks to be assessed segment by segment based on geographical location, significantly improving the accuracy and spatial consistency of fault prediction.

[0045] 2. A distance-based piecewise function is introduced to calculate the severity of ice disasters at each line location, realizing a mathematical expression of the physical law that "the closer to the center of icing, the greater the impact." This breaks away from the traditional approach of judging faults solely based on whether the area is in an icing zone, making fault assessment more closely aligned with actual operating conditions, and is particularly suitable for localized heavy icing areas in complex terrain.

[0046] Preferably, in step (4), the process of generating the ice disaster scene set includes:

[0047] A conditional variational autoencoder (CVAE) is constructed, using structured perturbation data as conditional input and source-grid-load operation data as target variable.

[0048] The CVAE is trained unsupervised using multiple sets of historical paired data. Each set of historical paired data includes historical source-grid-load operation data and its corresponding structured disturbance data, so that the CVAE can learn the conditional probability distribution of source-grid-load operation data under given disturbance conditions.

[0049] The structured perturbation data generated in step (3) is used as a conditional input to the trained CVAE, and the latent variables are sampled from the standard Gaussian distribution through its decoder to generate source-network-load operation data that matches the perturbation conditions.

[0050] Based on the generated "source-grid-load" operation data, a set of ice storm scenarios is constructed.

[0051] This setup achieves the following: 1. It enables the generation of conditions for "disturbance → operating status," improving the physical consistency of the generated scenario. By employing a CVAE model and using structured disturbance data as conditional input, the generated "source-grid-load" operating data is forced to maintain logical consistency with the input disturbance. Compared to traditional random sampling or simple interpolation methods, this approach ensures that the generated scenario is causally reasonable, avoiding situations that violate physical laws, such as "no faults but a significant drop in output."

[0052] 2. Output structured and quantifiable system operation data to support subsequent quantitative assessment and optimization. The generated "source-grid-load" operation data includes key indicators such as power generation output, transmission power, and load level, which can be directly used to calculate resilience indicators such as power generation recovery ratio, transmission recovery ratio, and load recovery ratio, forming a complete "sensing-modeling-assessment-optimization" closed loop.

[0053] Preferably, in S2, the power generation recovery ratio R s,wp for:

[0054]

[0055] In the formula, P s,i,wp,t ΔP represents the active power generation of the wind and solar turbines connected to node i at time t in scenario s; s,i,wp,t Ω represents the reduction in active power generation of the wind and solar turbines connected to node i at time t in scenario s; T represents the total planning time; N For the set of power transmission network nodes;

[0056] Power transmission recovery ratio R s,ij for:

[0057]

[0058] In the formula, Ω NL For a collection of power transmission lines; P s,ij,max,t Let ΔP be the maximum active power that transmission line ij can transmit at time t under scenario s; s,ij,max,t The reduction in the maximum active power that transmission line ij can transmit at time t under scenario s;

[0059] Load recovery ratio R s,L for:

[0060]

[0061] In the formula, P s,i,L,t Let ΔP be the load at node i at time t in scenario s; s,i,L,t This represents the load reduction at node i at time t in scenario s.

[0062] System overall recovery rate R s for:

[0063] R s =w wp R s,wp +w ij R s,ij +w L R s,L ;

[0064] In the formula, w wp w ij and w L These are the weighting coefficients, and w wp +w ij +w L =1.

[0065] This setup enables the quantification of resilience across the entire power generation-grid-load chain, enhancing the systematic and comprehensive nature of the assessment. Traditional resilience assessments often focus on a single aspect (such as outage duration or recovery speed), while this scheme defines the recovery ratios for generation, transmission, and load at three levels, comprehensively characterizing the multidimensional response characteristics of the power system under ice storms. This hierarchical assessment mechanism avoids overgeneralization, making resilience evaluation more comprehensive and scientific.

[0066] Preferably, in S3, the optimization problem of the two-stage robust resilience enhancement planning model is expressed as:

[0067]

[0068] In the formula, C Inv For fixed energy storage configuration costs; C Main λ represents the investment cost for emergency maintenance; λ is the investment decision variable for the first stage, including the installation location, configured capacity, and configured power of stationary energy storage; p s Ω represents the probability of ice disaster scenario s occurring; S represents the total number of ice disaster scenarios; s To meet And p s The set of probability distributions ≥ 0; The second phase is defined by investment decision-making and system operation decision-making variables under ice storm scenarios, including the charging and discharging power of stationary energy storage and the output of thermal power units; C R Penalty for unit load reduction; R s This is a resilience loss index under the ice storm scenario s;

[0069] The optimization problem also includes resource constraints of stationary energy storage devices, operational constraints of stationary energy storage, DC power flow constraints of the transmission network, and node power balance constraints.

[0070] This setup achieves two key advantages: 1) It enables synergistic optimization between pre-disaster investment and in-disaster operation, breaking through the traditional fragmented planning model. Unlike existing studies that separate "pre-disaster reinforcement + post-disaster repair," this scheme jointly models energy storage configuration (phase one) and emergency dispatch (phase two), allowing pre-disaster resource allocation to proactively adapt to in-disaster operational needs. This integrated design fully utilizes the control window created by the "slow development" characteristic of ice storms, significantly enhancing the overall system's response capability.

[0071] 2. A distributed bar optimization approach is introduced to enhance the model's resilience to uncertainties in ice storm scenarios. Unlike deterministic optimization or simple stochastic programming, this approach maximizes the expected resilience loss term under the worst-case probability distribution, thus considering situations where probabilistic information about ice storm scenarios is incomplete or biased. This avoids the risk of the solution failing due to distorted probability assumptions, improving the robustness and reliability of the planning results.

[0072] 3. Quantify resilience loss into an optimizable objective, promoting the shift from qualitative to quantitative resilience assessment. Use the overall system recovery ratio Rs as the resilience loss indicator, and reduce the penalty cost C per unit load. R This translates into economic costs, making the abstract concept of "resilience" measurable, comparable, and optimizable. This transforms resilience enhancement from a subjective judgment into a scientific decision-making process based on clear cost-benefit analysis.

[0073] 4. Integrating physical constraints of the power system to ensure the technical feasibility of the optimization scheme. The model includes key constraints such as energy storage operation, DC power flow, and power balance, ensuring that the output energy storage configuration scheme and emergency dispatch strategy are not only economically optimal but also executable in the actual power grid. This dual guarantee mechanism of "economic efficiency + feasibility" enhances the engineering application value of the scheme.

[0074] Preferably, the fixed energy storage configuration cost C Inv for:

[0075]

[0076] In the formula, Ω N For the set of power transmission network nodes; It is a 0-1 variable. If it is 1, it means that a fixed energy storage device is installed at node i; otherwise, it is 0. and These represent the basic configuration cost, unit configuration capacity cost, and unit configuration power cost of fixed energy storage at node i in the transmission network, respectively. and P i SES These represent the configured capacity and power of stationary energy storage at node i in the transmission network, respectively.

[0077] Emergency repair investment cost C Main for:

[0078]

[0079] In the formula, Ω M Assemble the maintenance team; C m and C S These are the unit maintenance team cost and the unit maintenance resource cost, respectively; N m The total number of maintenance teams; ω m The maximum maintenance resources that can be allocated to maintenance team m.

[0080] This setup enables refined modeling and collaborative optimization of key investment elements in the process of enhancing the resilience of power transmission networks during ice storms, providing reliable technical support for balancing economic benefits and system resilience.

[0081] Preferably, the resource constraints of stationary energy storage equipment include:

[0082]

[0083] In the formula, For fixed energy storage devices, α i It is a 0-1 variable. If it is 1, it means that a fixed energy storage device is installed at node i; otherwise, it is 0.

[0084] Fixed energy storage configuration capacity at node i and configuration power P i SES The following constraints must be satisfied:

[0085]

[0086] In the formula, and These represent the maximum and minimum fixed energy storage configuration capacity at node i, respectively. Configure the maximum power of the fixed energy storage at node i; and These are the maximum and minimum charging rates for fixed energy storage, respectively.

[0087] Stationary energy storage operation constraints include:

[0088]

[0089] In the formula, and The variable is 0-1. If it is 1, it means that the fixed energy storage i is in charging or discharging state at time t under scenario s, respectively; otherwise, it is 0. and Let represent the charging and discharging power of fixed energy storage i at time t under scenario s; This represents the stored energy of fixed energy storage i at time t under scenario s; This represents the stored energy of fixed energy storage i at time t+Δt under scenario s; and These are the charging and discharging efficiencies of the fixed energy storage i; and These are the upper and lower limits of the fixed energy storage state of charge, respectively.

[0090] Preferably, the DC power flow constraints of the transmission network include:

[0091] -M(1-S s,ij,t )≤B ij θ s,ij,t -P s,ij,t ≤M(1-S s,ij,t );

[0092] -S s,ij,t P s,ij,max,t ≤P s,ij,t ≤S s,ij,t P s,ij,max,t ;

[0093] θ s,i,min ≤θ s,i,t ≤θ s,i,max ;

[0094] In the formula, B ij S represents the susceptance value of line ij; s,ij,t The variable is 0-1, representing the state of transmission line ij at time t under scenario s. It takes the value 0 when the line is in a fault or under maintenance state, and 1 otherwise; M is a number greater than a preset value; θ s,ij,t θ represents the voltage phase angle difference between node i and node j at time t in scenario s. s,i,max and θ s,i,min These represent the upper and lower limits of the voltage phase angle of node i in scenario s, respectively; P s,ij,t The active power transmitted by line ij at time t under scenario s; P s,ij,max,t Let be the maximum permissible active power transmitted by line ij at time t under scenario s;

[0095] Node power balance constraints include:

[0096]

[0097] P s,i,g,min,t ≤P s,i,g,t ≤P s,i,g,max,t ;

[0098] P s,i,wp,min,t ≤P s,i,wp,t ≤P s,i,wp,max,t ;

[0099] 0≤ΔPs,i,L,t ≤P s,i,L,t ;

[0100] In the formula, Ω Gi Ω WPi These are the sets of thermal power and wind / solar turbine units at node i; Ω to(i) Ω is the set of transmission lines whose last node is i. from(i) The set of transmission lines whose first node is i; P s,i,g,t P represents the active power generation of the thermal power unit connected to node i at time t in scenario s; s,i,wp,t P represents the active power generation of the wind and solar turbines connected to node i at time t in scenario s; s,i,L,t P represents the load at node i at time t in scenario s; s,i,g,min,t P s,i,g,max,t P s,i,wp,min,t and P s,i,wp,max,t These represent the minimum and maximum active power generation of the thermal power unit and wind and solar power unit connected to node i at time t in scenario s, respectively. Attached Figure Description

[0101] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will now be described in further detail with reference to the accompanying drawings, wherein:

[0102] Figure 1 This is a flowchart of the method;

[0103] Figure 2 This is a schematic diagram of the image fusion process based on Laplacian pyramid decomposition in Example 1;

[0104] Figure 3 This is a schematic diagram of the dynamic prediction model for icy region images based on partial differential properties of a convolutional neural network in Example 1.

[0105] Figure 4 This is a schematic diagram of the gridding process for the icing area in Example 1;

[0106] Figure 5 This is a schematic diagram illustrating the changes in the icing area in a certain location as shown in Example 2;

[0107] Figure 6 This is a diagram of the improved IEEE RTS-79 power transmission system in Example 2;

[0108] Figure 7 This is a schematic diagram showing the temporal variation of the severity of the ice storm at node 24 in Example 2;

[0109] Figure 8 This is a schematic diagram of the overall system recovery ratio curves under different resilience enhancement strategies in Example 2. Detailed Implementation

[0110] The following detailed explanation illustrates the specific implementation methods:

[0111] Example 1

[0112] like Figure 1 As shown in the figure, this embodiment discloses a remote sensing-based method for optimizing the resilience of power transmission networks during ice storms, including the following steps:

[0113] S1. Based on multispectral satellite remote sensing data of the target area, an ice disaster scene generation process is adopted to generate a set of scenes that reflect the actual evolution of ice disasters in the area. Each scene in the scene set includes information on fault status caused by line icing, wind and solar power output loss and load disturbance, and associates their occurrence probability.

[0114] The ice disaster scene generation process includes:

[0115] (1) Collect and screen multispectral satellite remote sensing data, and use image fusion technology to generate high-resolution, continuous time series images of icing areas to improve the accuracy of spatial distribution identification of icing.

[0116] (2) Based on the continuous time series of icing area images generated in step (1), construct a spatiotemporal evolution model, analyze the changing patterns of icing patterns, predict the spatial distribution and dynamic evolution trend of icing areas in future periods, and obtain the prediction results of icing areas and their spatial distribution characteristics; the spatial distribution characteristics include icing thickness and coverage area.

[0117] (3) Based on the icing area prediction results and spatial distribution characteristics provided in step (2), calculate the fault status of transmission lines, the output loss of wind and solar power generation systems and the disturbance of grid load to obtain structured disturbance data.

[0118] (4) Combining historical source-grid-load operation data and structured disturbance data obtained in step (3), a probabilistic generation model is used to learn their joint distribution law and generate a set of ice disaster scenarios that conform to the actual distribution characteristics.

[0119] In specific implementation, step (2) involves constructing a spatiotemporal evolution model, analyzing the changing patterns of icing patterns, and predicting the spatial distribution and dynamic evolution trend of icing areas in future periods, including:

[0120] A convolutional neural network based on partial differential characteristics is constructed as a spatiotemporal evolution model. The spatiotemporal evolution model includes convolutional layers with embedded partial differential operators and a temporal recursive feedback structure. The convolutional layers with embedded partial differential operators are used to simulate the spatial diffusion characteristics of the icing area, and the temporal recursive feedback structure is used to use the prediction results of the current time period as the input of the next time period during the training and prediction stages to achieve multi-step temporal prediction.

[0121] Under the rolling prediction training framework, the continuous time series icing area image generated in step (1) is used as the initial input. The spatiotemporal evolution model is used to generate the predicted images for the time period from t+1 to t+c-1. The predicted images for the time period from t+1 to t+c-1 are combined with the real icing area image for time period t as input to train the spatiotemporal evolution model to predict the icing area image for time period t+c.

[0122] The latest continuous time series icing area image generated in step (1) is input into the spatiotemporal evolution model after training, and the icing area prediction results for the next c time periods are recursively output through its time recursive feedback structure; the prediction results include the icing thickness and coverage area for each time period.

[0123] By embedding partial differential operators into the convolutional neural network, the model can explicitly capture the spatial diffusion behavior of ice cover (such as gradient propagation and edge expansion), rather than relying solely on pure data fitting. This "physically guided" modeling approach significantly enhances the model's generalization ability and interpretability, avoiding the "illusionary" predictions that may occur in traditional deep learning models. Furthermore, the introduction of a temporal recursive feedback structure allows the model to maintain its memory and recursive ability regarding historical states during both the training and prediction phases. This not only improves the model's response accuracy to short-term changes but also enables it to continuously extrapolate the development trend of ice cover over longer timescales, meeting the long-term prediction needs during the slow development of ice disasters. Moreover, during training, predicted images (rather than true values) from the previous c-1 time periods are used as input to predict the c-th time period, simulating the autoregressive inference process in real-world applications. This approach ensures that the model learns how to handle its own prediction errors during training, resulting in stronger robustness and stability during deployment and avoiding training-inference inconsistencies caused by "teacher-mandated" approaches.

[0124] In specific implementation, step (3) uses a fault rate model to calculate the fault state of the transmission line. The process of constructing the fault rate model includes:

[0125] The predicted icing areas were gridded using a geographic information system, with each icing area being equivalent to a radius. Circular region; where S Θ,t Predict the area of ​​the icing region Θ at time t;

[0126] Based on the distance R between the coordinate point of the transmission line and the center of the icing area Θ Θ,(a,b),t The severity of the ice storm, Y, at that point is determined using a piecewise function. Θ,(a,b),t ;

[0127] Based on ice thickness R l,m,t and the maximum anti-icing thickness d of the line lA piecewise exponential function is used to establish the failure rate model of the line within the grid:

[0128]

[0129] In the formula, p l,m,t Let be the failure rate of line l in the m-th grid region at time t.

[0130] By using GIS gridding, continuous icing areas are transformed into regular geographic units and precisely matched with the coordinates of transmission lines, avoiding errors caused by "global thresholds" or "coarse-grained division" in traditional methods. This refined spatial alignment mechanism allows line fault risks to be assessed segment by segment based on geographical location, significantly improving the accuracy and spatial consistency of fault prediction. Furthermore, a distance-based piecewise function is introduced to calculate the severity of ice-related disasters at each line location, providing a mathematical expression of the physical law that "the closer to the center of icing, the greater the impact." This breaks away from the traditional approach of judging faults solely based on whether the area is within an icing zone, making fault assessment more closely aligned with actual operating conditions, especially suitable for localized heavy icing areas in complex terrain.

[0131] In specific implementation, step (4) involves generating a set of ice disaster scenarios, including:

[0132] A conditional variational autoencoder (CVAE) is constructed, using structured perturbation data as conditional input and source-grid-load operation data as target variable.

[0133] The CVAE is trained unsupervised using multiple sets of historical paired data. Each set of historical paired data includes historical source-grid-load operation data and its corresponding structured disturbance data, so that the CVAE can learn the conditional probability distribution of source-grid-load operation data under given disturbance conditions.

[0134] The structured perturbation data generated in step (3) is used as a conditional input to the trained CVAE, and the latent variables are sampled from the standard Gaussian distribution through its decoder to generate source-network-load operation data that matches the perturbation conditions.

[0135] Based on the generated "source-grid-load" operation data, a set of ice storm scenarios is constructed.

[0136] By employing a CVAE model and using structured disturbance data as input, the generated "source-grid-load" operational data is forced to maintain logical consistency with the input disturbance. Compared to traditional random sampling or simple interpolation methods, this approach ensures that the generated scenarios are causally reasonable, avoiding situations that violate physical laws, such as "no faults but a significant drop in output." Furthermore, the generated "source-grid-load" operational data includes key indicators such as power generation output, transmission power, and load levels, which can be directly used to calculate resilience indicators such as power generation recovery rate, transmission recovery rate, and load recovery rate, forming a complete "perception-modeling-evaluation-optimization" closed loop.

[0137] To facilitate a better understanding of the technology in S1 of this invention by those skilled in the art, the following description is provided.

[0138] The Laplacian pyramid is used to perform multi-scale decomposition of multispectral remote sensing images. A Gaussian filter is used to decompose the original remote sensing images of different band combinations after grayscale processing into multiple detail and approximate components at different scales.

[0139]

[0140] In the formula, D l and Represents the details and approximate components of the l-th layer of the pyramid; L is the number of pyramid layers; A l Image of the first level of the Gaussian pyramid; D b This represents the original remote sensing image with absorption type band b; ups(·) is the upsampling function; A 5×5 The filter is a Gaussian filter. After decomposition, the details and approximate components of different scale decomposition layers are fused separately, and finally the fused image is obtained through inverse Laplacian transform. Figure 2 The process of fusion of remote sensing images of an icy region with b=2 and L=1 is shown. The white area in the fused image is the icy region.

[0141] In image fusion, the weights of the fusion components are determined by the saliency level of the original image. To reduce aliasing due to spatial feature mismatch while preserving the advantages of each spectral feature, guided filtering is used to correct the saliency level of the original image during image fusion. The fusion calculation method is as follows:

[0142]

[0143] In the formula, D * To merge images; D is the fusion component of the l-th layer of the pyramid; b,l The pyramid layer l-level fusion component of band b; W b,l S represents the weights of the pyramid layer l-level fusion components for band b; b f represents the significance level of band b;l (;r l ,ε l ) represents the guided filter function for the l-th layer of the pyramid; r l The radius of the parametric guided filter; ε l D is the regularization coefficient. base This serves as the basis for guided filtering.

[0144] Based on the partial differential characteristics, dynamic prediction of icing area images is performed. Temporal modeling is constructed according to the development of ice disaster to obtain the power transmission system fault model and the power output loss model of wind turbine generator.

[0145] like Figure 3 As shown, a convolutional neural network is used to construct an image prediction model for icing areas, predicting changes in icing areas based on historical images. Furthermore, by combining the dynamic characteristics of partial differential equations (PDAs), a dynamic prediction model for icing area images based on PDAs is constructed using a convolutional neural network.

[0146] When the convolution kernel of a convolutional neural network satisfies the moment matrix, partial differential operations can be used as convolution operations.

[0147]

[0148] In the image scrolling prediction model, the predicted image within the time interval t+1 to t+c-1 is used. Remote sensing images of the actual icing area during time period t Image of icing area used as model input to predict time period t+c The prediction model is as follows:

[0149]

[0150] The damage caused by ice storms to power transmission networks mainly lies in the damage to transmission network equipment. The probability of equipment failure is related to the severity of the ice storm, Y. k Directly related, with a wide power grid coverage area and similar failure probabilities of adjacent equipment, to ensure the practicality of assessing the severity of ice storms, the area S of a predicted icing region at time t is used in ArcGIS software. Θ,t The icing area is equivalent to a radius of R. Θ,t A circular area.

[0151]

[0152] Improving the prediction accuracy of ice disaster development time series models by gridding ice-covered areas, such as... Figure 4 As shown.

[0153] The severity of the ice storm at coordinate point A(a, b) is Y. Θ,(a,b),t The distance R between it and the center of the icing area Θ Θ,(a,b),t The decision is as follows:

[0154]

[0155] Based on the icing thickness, the vulnerability model of the transmission line is as follows:

[0156]

[0157] In the formula, p l,m,t Let d be the failure rate of line l in the m-th grid region at time t; l The maximum anti-icing thickness of transmission line l.

[0158] The failure rate p of line l at time t l,t and transmission capacity S l,t as follows:

[0159]

[0160] In the formula, N l,G S represents the total number of grids in which line l is located; l,0 This refers to the transmission capacity under normal operating conditions of the line.

[0161] During ice storms, icing at the tips of wind turbine blades and on the surfaces of photovoltaic modules leads to a significant decrease in the output of wind and solar power units. The icing thickness R wp,m,t The prediction model is shown below. The power output loss P of the wind and solar turbines in the m-th grid region at time t is... wp,t as follows:

[0162]

[0163] In the formula, α wp and β wp d is the wind and light loss coefficient; wp This represents the maximum anti-icing thickness for wind and solar turbine units.

[0164] Leveraging the powerful fitting capabilities of deep neural networks, the variational autoencoder (VAE) learns the statistical patterns of historical icing regions unsupervised by inputting "source-network-load" data. Based on the trained probability distribution model, it generates new "source-network-load" data that conforms to the characteristics of the data distribution.

[0165] S2. Construct a multi-dimensional system resilience assessment index system for source-grid-load, which includes the power generation recovery ratio, transmission recovery ratio and load recovery ratio; and obtain the system comprehensive recovery ratio based on the weighted average of these three indicators, which serves as the resilience loss index Rs to quantify the loss of system operating performance.

[0166] In practice, the power generation recovery ratio Rs,wp for:

[0167]

[0168] In the formula, P s,i,wp,t ΔP represents the active power generation of the wind and solar turbines connected to node i at time t in scenario s; s,i,wp,t Ω represents the reduction in active power generation of the wind and solar turbines connected to node i at time t in scenario s; T represents the total planning time; N For the set of power transmission network nodes;

[0169] Power transmission recovery ratio R s,ij for:

[0170]

[0171] In the formula, Ω NL For a collection of power transmission lines; P s,ij,max,t Let ΔP be the maximum active power that transmission line ij can transmit at time t under scenario s; s,ij,max,t The reduction in the maximum active power that transmission line ij can transmit at time t under scenario s;

[0172] Load recovery ratio R s,L for:

[0173]

[0174] In the formula, P s,i,L,t Let ΔP be the load at node i at time t in scenario s; s,i,L,t This represents the load reduction at node i at time t in scenario s.

[0175] System overall recovery rate R s for:

[0176] R s =w wp R s,wp +w ij R s,ij +w L R s,L ;

[0177] In the formula, w wp w ij and w L These are the weighting coefficients, and w wp +w ij +w L =1.

[0178] Traditional resilience assessments often focus on a single aspect (such as outage duration or recovery speed), while this scheme defines the recovery ratios at three levels: generation, transmission, and load, comprehensively characterizing the multidimensional response characteristics of the power system under ice storms. This hierarchical assessment mechanism avoids overgeneralization and makes resilience evaluation more comprehensive and scientific.

[0179] S3. Based on the scenario set generated by S1, reflecting the actual ice disaster characteristics of the target area and its occurrence probability, and combined with the resilience loss index Rs defined in S2, a two-stage robust resilience improvement planning model for the power transmission network is constructed:

[0180] The first phase uses the location, capacity, and power of pre-disaster fixed energy storage as decision variables to minimize investment costs;

[0181] The second stage uses the energy storage charging and discharging power, thermal power unit output, and maintenance resource scheduling during the disaster as decision variables to minimize operational losses under a given ice disaster scenario.

[0182] The objective function is to minimize the total cost, which includes the energy storage configuration cost, the maintenance resource configuration cost, and the expected resilience loss term under the worst probability distribution.

[0183] In practical implementation, the optimization problem of the two-stage robustness enhancement planning model is expressed as:

[0184]

[0185] In the formula, C Inv For fixed energy storage configuration costs; C Main For emergency repair investment costs; The investment decision variables for the first phase include the installation location, configured capacity, and configured power of stationary energy storage; p s Ω represents the probability of ice disaster scenario s occurring; S represents the total number of ice disaster scenarios; s To meet And p s The set of probability distributions ≥ 0; The second phase is defined by investment decision-making and system operation decision-making variables under ice storm scenarios, including the charging and discharging power of stationary energy storage and the output of thermal power units; C R Penalty for unit load reduction; R s This is a resilience loss index under the ice storm scenario s;

[0186] The optimization problem also includes resource constraints of stationary energy storage devices, operational constraints of stationary energy storage, DC power flow constraints of the transmission network, and node power balance constraints.

[0187] In practical implementation, the fixed energy storage configuration cost C Inv for:

[0188]

[0189] In the formula, Ω N For the set of power transmission network nodes; It is a 0-1 variable. If it is 1, it means that a fixed energy storage device is installed at node i; otherwise, it is 0. and These represent the basic configuration cost, unit configuration capacity cost, and unit configuration power cost of fixed energy storage at node i in the transmission network, respectively. and P i SES These represent the configured capacity and power of stationary energy storage at node i in the transmission network, respectively.

[0190] Emergency repair investment cost C Main for:

[0191]

[0192] In the formula, Ω M Assemble the maintenance team; C m and C S These are the unit maintenance team cost and the unit maintenance resource cost, respectively; N m The total number of maintenance teams; ω m The maximum maintenance resources that can be allocated to maintenance team m.

[0193] Resource constraints for stationary energy storage equipment include:

[0194]

[0195] In the formula, For fixed energy storage devices, α i It is a 0-1 variable. If it is 1, it means that a fixed energy storage device is installed at node i; otherwise, it is 0.

[0196] Fixed energy storage configuration capacity at node i and configuration power P i SES The following constraints must be satisfied:

[0197]

[0198] In the formula, and These represent the maximum and minimum fixed energy storage configuration capacity at node i, respectively. Configure the maximum power of the fixed energy storage at node i; and These are the maximum and minimum charging rates for fixed energy storage, respectively.

[0199] Stationary energy storage operation constraints include:

[0200]

[0201]

[0202] In the formula, and The variable is 0-1. If it is 1, it means that the fixed energy storage i is in charging or discharging state at time t under scenario s, respectively; otherwise, it is 0. and Let represent the charging and discharging power of fixed energy storage i at time t under scenario s; This represents the stored energy of fixed energy storage i at time t under scenario s; This represents the stored energy of fixed energy storage i at time t+Δt under scenario s; and These are the charging and discharging efficiencies of the fixed energy storage i; and These are the upper and lower limits of the fixed energy storage state of charge, respectively.

[0203] DC power flow constraints in power transmission networks include:

[0204] -M(1-S s,ij,t )≤B ij θ s,ij,t -P s,ij,t ≤M(1-S s,ij,t );

[0205] -S s,ij,t P s,ij,max,t ≤P s,ij,t ≤S s,ij,t P s,ij,max,t ;

[0206] θ s,i,min ≤θ s,i,t ≤θ s,i,max ;

[0207] In the formula, B ij S represents the susceptance value of line ij; s,ij,t The variable is 0-1, representing the state of transmission line ij at time t under scenario s. It takes the value 0 when the line is in a fault or under maintenance state, and 1 otherwise; M is a number greater than a preset value; θ s,ij,t θ represents the voltage phase angle difference between node i and node j at time t in scenario s. s,i,max and θ s,i,min These represent the upper and lower limits of the voltage phase angle of node i in scenario s, respectively; P s,ij,t The active power transmitted by line ij at time t under scenario s; P s,ij,max,t Let be the maximum permissible active power transmitted by line ij at time t under scenario s;

[0208] Node power balance constraints include:

[0209]

[0210] P s,i,g,min,t ≤P s,i,g,t ≤P s,i,g,max,t ;

[0211] P s,i,wp,min,t ≤P s,i,wp,t ≤P s,i,wp,max,t ;

[0212] 0≤ΔP s,i,L,t ≤P s,i,L,t ;

[0213] In the formula, Ω Gi Ω WPi These are the sets of thermal power and wind / solar turbine units at node i; Ω to(i) Ω is the set of transmission lines whose last node is i. from(i) The set of transmission lines whose first node is i; P s,i,g,t P represents the active power generation of the thermal power unit connected to node i at time t in scenario s; s,i,wp,t P represents the active power generation of the wind and solar turbines connected to node i at time t in scenario s; s,i,L,t P represents the load at node i at time t in scenario s; s,i,g,min,t P s,i,g,max,t P s,i,wp,min,t and P s,i,wp,max,t These represent the minimum and maximum active power generation of the thermal power unit and wind and solar power unit connected to node i at time t in scenario s, respectively.

[0214] Unlike existing studies that separately address "pre-disaster reinforcement + post-disaster repair," this scheme jointly models energy storage configuration (phase one) and emergency dispatch (phase two), enabling pre-disaster resource allocation to proactively adapt to operational needs during a disaster. This integrated design fully utilizes the control window created by the "slow development" characteristic of ice storms, significantly enhancing the overall system's response capability. Furthermore, unlike deterministic optimization or simple stochastic programming, this scheme maximizes the expected resilience loss term under the worst-case probability distribution, considering situations where probabilistic information in ice storm scenarios is incomplete or biased. This avoids the risk of scheme failure due to distorted probability assumptions, improving the robustness and reliability of the planning results.

[0215] The overall system recovery ratio Rs is used as a resilience loss indicator, and the penalty cost C is reduced based on the unit load reduction. RThis translates into economic costs, making the abstract concept of "resilience" measurable, comparable, and optimizable. This transforms resilience enhancement from a subjective judgment into a scientific decision-making process based on clear cost-benefit analysis. Furthermore, the model incorporates key constraints such as energy storage operation, DC power flow, and power balance, ensuring that the output energy storage configuration and emergency dispatch strategy are not only economically optimal but also feasible in a real power grid. This dual guarantee mechanism of "economic efficiency + feasibility" enhances the engineering application value of the solution.

[0216] S4. The two-stage robust programming model is iteratively solved using a column and constraint generation algorithm to output the optimal energy storage configuration scheme and emergency maintenance scheduling strategy.

[0217] S5. Deploy the energy storage configuration scheme output by S4 in the actual operation of the power grid, and dynamically adjust the charging and discharging of energy storage and the output of generating units according to the emergency maintenance and dispatch strategy when ice disasters occur, so as to improve the resilience of the transmission network and minimize economic losses.

[0218] Unlike existing ice disaster modeling methods that rely on historical statistics or static assumptions, this approach uses multispectral satellite remote sensing data to integrate the dynamic evolution of icing areas with the "source-grid-load" disturbance response mechanism. This generates structured scenarios that include line fault states, wind and solar power output losses, and load disturbances, assigning each scenario a probability of occurrence. This method overcomes the limitations of traditional scenarios that are "detached from physical processes and lack spatiotemporal correlation," providing a more realistic input basis for subsequent resilience assessment and optimization. Furthermore, existing research often focuses on a single dimension (such as power restoration time) or qualitatively describes resilience levels. This approach proposes using the recovery ratios of three key elements—generation, transmission, and load—as core indicators, and through weighted aggregation, forms a unified resilience loss index, Rs. This makes the degree of system performance degradation quantifiable and comparable, providing a clear evaluation target for the optimization model.

[0219] Unlike the traditional fragmented approach of "pre-disaster prevention + post-disaster repair," this solution leverages the long duration and relaxation effects of ice storms. Within a two-stage robust programming framework, it jointly optimizes pre-disaster energy storage investment decisions with in-disaster operational scheduling (including energy storage charging and discharging, thermal power output adjustment, and maintenance resource allocation). This end-to-end collaborative strategy effectively connects the "prevention-response-recovery" chain, significantly enhancing the system's proactive adaptability during disasters. Furthermore, addressing the high uncertainty of ice storm scenarios, the model incorporates distributed robust optimization into the objective function, considering the most unfavorable but possible probability distribution to prevent solution failure due to scenario deviations. Compared to deterministic optimization or simple stochastic programming, this method significantly improves the reliability of responding to extreme or unforeseen ice storm scenarios while ensuring economic efficiency.

[0220] In summary, this method can improve the resilience of the power transmission network and minimize the total cost during the entire process of slow-developing ice storms.

[0221] Example 2

[0222] To better illustrate the effects of the present invention, the following numerical examples are provided.

[0223] A case study analysis was conducted using remote sensing data on icing in a certain area and an improved IEEE RTS-79 power transmission system as the test system. The changes in icing areas in this region are shown below. Figure 5 As shown.

[0224] Establish a coordinate system xOy with the maintenance center O as the origin, and a unit length of 10km, such as... Figure 6 As shown in the figure. The temporal variation of the severity of the ice storm at node 24 is as follows. Figure 7 As shown.

[0225] Thermal power units are connected at nodes 7, 18, and 22; wind farms with the same installed capacity are connected at nodes 2, 15, 21, and 23; photovoltaic power stations with the same installed capacity are connected at nodes 1, 13, 14, and 16; and loads are connected at nodes 3, 4, 19, and 20. Assume the ideal travel time for a maintenance team to 1 km is 0.1 hours, the ideal repair time for a faulty component is 2 hours, and the required maintenance resources for each faulty component are the same, 15. The investment parameters for fixed energy storage and maintenance resources are shown in Table 1.

[0226] Table 1 Investment parameters for stationary energy storage and maintenance resources

[0227]

[0228] The initial state of charge (SOC) of the energy storage is set to 0.6, with upper and lower limits of SOC of 0.9 and 0.1, respectively, and a charge / discharge efficiency of 0.95. The penalty for load shedding is set at 5000 yuan / (MW·h), with a time step of 1 hour.

[0229] To verify the advantages of the proposed two-stage robust resilience enhancement planning strategy for transmission networks, four resilience enhancement strategies were compared and analyzed. Strategy 1 took no measures; Strategy 2 considered only pre-disaster fixed energy storage configuration; Strategy 3 considered only emergency maintenance by maintenance teams during the disaster; and Strategy 4, the strategy proposed in this invention, considered both pre-disaster fixed energy storage configuration and emergency maintenance by maintenance teams during the disaster. The overall system recovery ratio under different resilience enhancement strategies is as follows: Figure 8 As shown.

[0230] Depend on Figure 8It is evident that without any measures, the system's power generation, transmission capacity, and load will gradually decrease due to the continuous erosion caused by the ice storm. Compared to individual resilience enhancement measures, Strategy 4, through the coordinated planning of fixed energy storage and emergency maintenance resources, can maximize the speed of overall system recovery and reduce the amount of power load reduction. In terms of improving overall system resilience, Strategy 4 is 90.97%, 62.69%, and 82.98% higher than Strategies 1, 2, and 3, respectively. Strategy 2 only considers the configuration of fixed energy storage before the disaster, ignoring the gradual decrease in transmission capacity of transmission lines during the disaster. The continuous accumulation of ice leads to a gradual increase in the failure rate of transmission lines, limiting transmission power and failing to fully utilize the power support role of energy storage. Strategy 3 only considers emergency maintenance during the disaster, lacking the necessary power support from energy storage. When maintenance teams are on the move or in emergency maintenance mode, the reduction in system power load is significant. Strategy 4, through the coordinated cooperation of fixed energy storage power supply and emergency maintenance, can effectively ensure the supply and transmission of electricity during the ice storm, achieving a system recovery rate of over 96% in the later stages of the ice storm.

[0231] Table 2 Total Costs under Different Resilience Enhancement Strategies

[0232]

[0233] As shown in Table 2, although the investment cost of Strategy 4 is higher, its total cost is reduced by 43.19%, 24.14%, and 20.16% compared to Strategies 1, 2, and 3, respectively. Considering the resilience enhancement planning strategy of pre-disaster fixed energy storage configuration and emergency maintenance during disasters, the system load reduction loss and total cost are the lowest, resulting in better economic benefits while ensuring the safe supply of electricity.

[0234] Analysis of resilience enhancement planning results in different ice storm locations:

[0235] Ice storms 2-5 occurred at nodes 23, 9, 2, and 12, respectively, and their evolution process and system parameter settings were the same as those for ice storm 1. Table 3 shows the investment cost, overall system resilience improvement ratio, and total cost reduction ratio of the resilience enhancement strategy proposed in this invention under different ice storm locations.

[0236] Table 3 Total Cost of Resilience Enhancement at Different Ice Disaster Locations

[0237]

[0238] As shown in 3, the method proposed in this invention can effectively ensure the restoration of power supply to the transmission system, improve system resilience, and reduce load reduction losses and total costs for ice disasters occurring in different locations.

[0239] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the technical solutions. Those skilled in the art should understand that any modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A remote sensing based method for ice storm resilience optimization of power transmission grid, characterized in that, Comprise the following steps: S1, based on the target area of multispectral satellite remote sensing data, using ice disaster scene generation process, generate a set of reflect the actual evolution of the region ice disaster scene set, each scene in the scene set contains line ice caused by fault state, wind power output loss and load disturbance information, and its occurrence probability; S2, build source-net-load multi-dimensional system resilience evaluation index system, the index system includes power generation recovery rate, power transmission recovery rate and load recovery rate; and based on the three indicators proportion weighted comprehensive system comprehensive recovery rate as a resilience loss index Rs, to quantify the loss of system performance; S3, based on S1 generated, reflect the actual ice disaster characteristics of the scene set and its occurrence probability, combined with S2 defined resilience loss index Rs, build a two-stage robust resilience enhancement planning model of power transmission network: The first stage with the configuration position, capacity and power of fixed energy storage before disaster as decision variables, minimize the investment cost; The second stage with the charging and discharging power of energy storage, thermal power output and maintenance resource scheduling as decision variables, minimize the operation loss under the given ice disaster scenario; The objective function is to minimize the total cost, including energy storage configuration cost, maintenance resource configuration cost and expected resilience loss under the worst probability distribution; S4, solve the two-stage robust planning model of S3, output the optimal energy storage configuration scheme and emergency maintenance scheduling strategy; S5, deploy the energy storage configuration scheme output by S4 in the actual operation of the power grid, and dynamically adjust the charging and discharging of energy storage and unit output according to the emergency maintenance scheduling strategy when ice disaster occurs, realize the resilience enhancement and economic loss minimization of power transmission network.

2. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 1, wherein: In S1, the ice disaster scene generation process comprises: (1) Collect and screen multispectral satellite remote sensing data, use image fusion technology to generate high-resolution, continuous time series of icing area image, to improve the identification accuracy of icing spatial distribution; (2) Based on the continuous time series of icing area image generated in step (1), build a spatio-temporal evolution model, analyze the change rule of icing mode, predict the spatial distribution and dynamic evolution trend of icing area in future period, get the prediction result and spatial distribution characteristics of icing area; the spatial distribution characteristics include icing thickness and coverage range; (3) Based on the prediction result and spatial distribution characteristics of icing area provided in step (2), calculate the fault state of power transmission line, the output loss of wind and solar power generation system and the disturbance of power grid load, get the structured disturbance data; (4) Combine the historical source-net-load operation data and the structured disturbance data obtained in step (3), use the probability generation model to learn the joint distribution rule, generate an ice disaster scene set consistent with the actual distribution characteristics.

3. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 2, wherein: Step (2) said to build a spatio-temporal evolution model, analyze the change rule of icing mode, predict the spatial distribution and dynamic evolution trend of icing area in future period, including: A convolutional neural network based on partial differential characteristics is constructed as a spatio-temporal evolution model, which includes a convolutional layer embedding a partial differential operator and a time recurrent feedback structure; the convolutional layer embedding the partial differential operator is used to simulate the spatial diffusion characteristics of the icing area, and the time recurrent feedback structure is used to take the prediction result of the current period as the input of the next period in the training and prediction stages, so as to realize multi-step time series prediction; In the rolling prediction training framework, the continuous time series icing area image generated in step (1) is taken as the initial input, and the spatio-temporal evolution model is used to generate the prediction images of the t+1 to t+c-1 periods in turn, and the prediction images of the t+1 to t+c-1 periods and the real icing area image of the t period are jointly taken as the input, and the spatio-temporal evolution model is trained to predict the icing area image of the t+c period; The latest continuous time series icing area image generated in step (1) is input into the trained spatio-temporal evolution model, and the future c-period icing area prediction results are recursively output through the time recurrent feedback structure of the spatio-temporal evolution model; the prediction results include the icing thickness and coverage range of each period.

4. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 2, wherein: In step (3), the fault state of the power transmission line is calculated using a failure rate model, and the construction process of the failure rate model includes: The icing area is gridded by using a geographic information system, and each icing area is equivalent to a circle with a radius of S Θ,t is the area of the predicted icing area Θ at time t. According to the distance R between the coordinate point of the transmission line and the center of the icing area Θ Θ,(a,b),t , the icing severity Y at the point is determined by a piecewise function Θ,(a,b),t ; Based on the ice thickness R l,m,t and the maximum ice thickness d of the line l The failure rate model of the line in the grid is established by using piecewise exponential function: In the formula, p l,m,t is the failure rate of line l in the mth grid area at time t.

5. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 2, wherein: In step (4), the process of generating the ice disaster scenario set includes: A conditional variational autoencoder (CVAE) is constructed, which takes the structured disturbance data as the conditional input and the source-grid-load operation data as the target variable; A plurality of sets of historical paired data are used to unsupervisedly train the CVAE, each set of historical paired data including historical source-grid-load operation data and corresponding structured disturbance data, so that the CVAE learns the conditional probability distribution of the source-grid-load operation data under the given disturbance condition; The structured disturbance data generated in step (3) is input into the trained CVAE as the conditional input, and the latent variables are sampled from the standard Gaussian distribution through the decoder of the CVAE to generate the source-grid-load operation data matched with the disturbance condition; Based on the generated "source-grid-load" operation data, an ice disaster scenario set is constructed.

6. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 1 wherein: In S2, the power generation recovery ratio R s,wp is: In the formula, P s,i,wp,t is the active power generation of the wind and solar generator connected to node i at time t under scenario s; ΔP s,i,wp,t is the active power generation reduction of the wind and solar generator connected to node i at time t under scenario s; T is the total planning time; Ω N is the set of power transmission network nodes; Power transmission recovery ratio R s,ij is: where Ω NL is the set of transmission lines; P s,ij,max,t is the maximum active power that transmission line ij can deliver at time t under scenario s; ΔP s,ij,max,t is the reduction of the maximum active power that transmission line ij can deliver at time t under scenario s. Load recovery ratio R s,L is: where P s,i,L,t is the load at node i at time t under scenario s; ΔP s,i,L,t is the load reduction at node i at time t under scenario s; Systematic recovery ratio R s is: R s = w wp R s,wp + w ij R s,ij + w L R s,L ; where w wp , w ij , and w L are weight coefficients, and w wp +w ij +w L = 1.

7. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 1 wherein: In S3, the optimization problem of the two-stage robust resilience enhancement planning model is represented as: where C Inv is the fixed energy storage configuration cost; C Main is the emergency maintenance investment cost; is the first-stage investment decision variable, including the installation location, configuration capacity and configuration power of fixed energy storage; p s is the probability of ice disaster scenario s occurring; S is the total number of ice disaster scenarios; Ω s is the set of probability distributions of p and p s ≥ 0; is the second-stage system operation decision variable under given investment decisions and ice disaster scenarios, including the charge and discharge power of fixed energy storage and the output of thermal power units; C R is the penalty cost of unit load reduction; R s is the resilience loss index under ice disaster scenario s. The optimization problem further includes fixed energy storage device resource constraints, fixed energy storage operation constraints, power transmission network DC power flow constraints, and node power balance constraints.

8. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 7, wherein: Fixed energy storage configuration cost C Inv is: where Ω N is the set of transmission network nodes; is a 0-1 variable, which equals to 1 if a fixed energy storage device is installed at node i, otherwise equals to 0; and are the fixed energy storage basic configuration cost, unit configuration capacity cost and unit configuration power cost of node i in the transmission network, respectively; and P i SES are the configuration capacity and power of fixed energy storage at node i in the transmission network, respectively. Emergency repair investment cost C Main is: where Ω M is the set of repair crews; C m and C S are the unit repair crew cost and the unit repair resource cost, respectively; N m is the total number of repair crews deployed; ω m is the maximum repair resource that repair crew m can deploy.

9. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 8, wherein: The fixed energy storage device resource constraints include: wherein is the upper limit of the fixed energy storage device; a i is a 0-1 variable, which is 1 if node i is installed with a fixed energy storage device, otherwise 0; Configured capacity of fixed energy storage at node i and configured power P i SES subject to the following constraints: wherein, and are the maximum and minimum values of the fixed energy storage configuration capacity at node i, respectively; is the maximum value of the fixed energy storage configuration power at node i; and are the maximum and minimum charge rates of the fixed energy storage, respectively. The fixed energy storage operation constraints include: wherein, and are 0-1 variables, taking 1 if the fixed energy storage i is in charging or discharging state at time t under scenario s, respectively, and 0 otherwise; and represent the charging and discharging power of the fixed energy storage i at time t under scenario s, respectively; represents the storage power of the fixed energy storage i at time t under scenario s; represents the storage power of the fixed energy storage i at time t + Δt under scenario s; and are the charging and discharging efficiencies of the fixed energy storage i, respectively; and are the upper and lower limits of the state of charge of the fixed energy storage, respectively.

10. The remote-sensing-based transmission-line ice storm robustness optimization method of claim 9, wherein: The power transmission network DC power flow constraints include: - M(1-S s,ij,t ) ≤ B ij θ s,ij,t - P s,ij,t ≤ M(1-S s,ij,t ) - S s,ij,t P s,ij,max,t ≤ P s,ij,t ≤ S s,ij,t P s,ij,max,t ; θ s,i,min ≤θ s,i,t ≤θ s,i,max ; where B ij is the susceptance value of line ij; S s,ij,t is a 0-1 variable, representing the state of transmission line ij at time t under scenario s, taking 0 when the line is in a fault or maintenance state, and 1 otherwise; M is a number greater than a preset value; θ s,ij,t is the phase angle difference of the voltage of node i and node j at time t under scenario s. θ s,i,max and θ s,i,min are the upper and lower limits of the voltage phase angle of node i under scenario s; P s,ij,t is the active power transmitted by line ij at time t under scenario s; P s,ij,max,t is the maximum allowed active power transmitted by line ij at time t under scenario s; The node power balance constraints include: P s,i,g,min,t ≤P s,i,g,t ≤P s,i,g,max,t ; P s,i,wp,min,t ≤P s,i,wp,t ≤P s,i,wp,max,t ; 0 < ΔP s,i,L,t ≤P s,i,L,t ; where Ω Gi , Ω WPi , Ω to(i) , Ω from(i) , P s,i,g,t , P s,i,wp,t , P s,i,L,t , P s,i,g,min,t , P s,i,g,max,t , P s,i,wp,min,t and P s,i,wp,max,t are the active power generation of thermal units, wind and solar units connected to node i, respectively; Ω Gi is the set of transmission lines with the end node i; Ω WPi is the set of transmission lines with the start node i; P to(i) is the active power generation of thermal units connected to node i at time t in scenario s; P from(i) is the active power generation of wind and solar units connected to node i at time t in scenario s; P s,i,g,t is the load at node i at time t in scenario s; P s,i,wp,t , P s,i,L,t , P s,i,g,min,t , P s,i,g,max,t , P s,i,wp,min,t and P s,i,wp,max,t are the minimum and maximum active power generation of thermal units, wind and solar units connected to node i at time t in scenario s, respectively.

Images