CN-121981534-A - Storm tide and embankment risk dynamic assessment method integrating multisource remote sensing data

CN121981534ACN 121981534 ACN121981534 ACN 121981534ACN-121981534-A

Abstract

The invention provides a storm surge and embankment risk dynamic assessment method integrating multisource remote sensing data, and relates to the technical field of risk prediction. The method comprises the steps of collecting multi-source observation data of storm tide stages, uniformly preprocessing the multi-source observation data to construct a standardized multi-dimensional time sequence state vector, introducing a disturbance seed function to construct a characteristic enhancement module, extracting local response changes in a disturbance window to generate a disturbance response sequence, carrying out frequency domain offset reconstruction to establish spectrum stability index enhancement characteristic robustness, combining a disturbance response path with a non-return enhancement factor, guiding context characteristic mapping decoding to realize risk region response focusing, designing a disturbance feedback dynamic strategy optimization mechanism, fusing disturbance amplitude modulation, channel adjustment and threshold control strategies to improve risk score prediction accuracy, constructing a joint loss function, and jointly optimizing spectrum disturbance consistency and risk score error to remarkably enhance identification and dynamic prediction performance of potential flood bank risks in a complex environment.

Inventors

WU ZHIHONG
DONG WENLONG
HU JINGWEN
WANG QIXIANG
SUN SHUNA
DENG YE
ZHAO YANFEI
FU YANDA

Assignees

山东省海洋预报减灾中心
山东五牛技术服务有限责任公司

Dates

Publication Date: 20260505
Application Date: 20260109

Claims (8)

1. A storm surge and embankment risk dynamic assessment method integrating multisource remote sensing data is characterized by comprising the following steps: acquiring original remote sensing image data in different states of a storm surge area, performing preprocessing, and constructing a first data set; Based on a first data set, extracting a multi-dimensional eigenvalue vector of a region, calculating response deviation of each dimension of the eigenvalue vector and a neighborhood, constructing a frequency disturbance sensitive factor by combining the mean value, introducing normal distribution difference to adjust disturbance amplitude, constructing a disturbance sensitivity tensor, and performing frequency transformation operation on the disturbance sensitivity tensor in adjacent dimensions to obtain a high-order spectrum structure diagram; Combining the disturbance sensitivity tensor with a high-order spectrum structure chart to construct a disturbance collaborative guide factor, and adopting a cross modulation structure and channel direction adjustment to obtain an enhanced regional characteristic representation; fusing the regional characteristic characterization and disturbance sensitivity tensor, extracting a wind factor disturbance response map, and embedding frequency disturbance amplitude and direction information to obtain a state vector; Constructing a multi-factor joint action space based on the state vector, defining channel adjustment, disturbance amplitude and response threshold action vector, and generating an action execution factor set by adopting a reinforcement learning strategy of consistency residual feedback; Performing gating fusion according to the state vector and the response threshold motion vector, constructing fusion feature mapping, introducing regional gating parameters by combining the motion execution factor set, and constructing a risk representation vector; And introducing disturbance amplitude window constraint, and combining optimization strategy and scoring mechanism to realize dynamic evaluation of storm tide embankment risk.
2. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion according to claim 1 is characterized by comprising the steps of obtaining multi-source original remote sensing image data of a storm surge and embankment area under different observation time sequences, performing unified cutting, spatial resampling, time alignment, missing test completion and normalization processing, constructing a standardized remote sensing grid data set with spatial resolution of 100m and time sequence of 365 days, extracting five-dimensional feature vectors at each pixel position, generating a supervision labeling sample by combining area label information, and constructing a first data set with label information, wherein the first data set is covered in a full period and has a unified structure.
3. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion is characterized by comprising the steps of firstly dividing a target area into a plurality of space units and extracting multi-source characteristic value vectors of each area through a hierarchical tensor graph modeling process guided by spectrum response based on structured area sample data constructed by a first data set, then calculating response deviation of each area and adjacent areas of each area on characteristic dimensions, further obtaining mean value and standard deviation information of the deviation to construct disturbance sensitive factors, introducing a normal distribution difference weight function to conduct weighted modulation on disturbance amplitude to generate disturbance sensitivity tensors describing the characteristic fluctuation of the areas on the basis, further conducting frequency transformation operation on the tensors on the adjacent dimensions, extracting multi-scale spectrum response components of the tensors, selecting representative frequency channels to construct multi-layer characteristic response graph tensors guided by frequency domains, and finally forming a high-order spectrum structure diagram for subsequent risk structure modeling and propagation calculation.
4. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion according to claim 1 is characterized in that a frequency response diagram between a region and wind factor nodes is obtained from a high-order frequency spectrum structure diagram according to disturbance sensitivity tensor, disturbance collaborative guide factors are constructed to reflect joint disturbance trends between the region nodes and the wind factor, a cross modulation structure is further constructed based on bidirectional embedding representation, a frequency spectrum attention path containing a nonlinear modulation function is introduced to perform channel direction adjustment on wind factor characteristics, and disturbance perception enhanced region characteristic characterization is obtained.
5. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion according to claim 1 is characterized in that based on disturbance perception enhanced regional feature characterization, a joint frequency response map is constructed to generate a spectrum state expression tensor by combining a connection relation between a region and wind factor nodes in a spectrum structure diagram, a structure enhancement is performed on frequency disturbance response through a channel dimension unfolding strategy, a disturbance guiding rule is constructed by introducing frequency coupling gradients, a mapping relation between a frequency domain signal and state expression is established, and finally a structure coding result is used as a state input of a disturbance guiding strategy network to form a state vector with disturbance response structure perception capability.
6. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion is characterized by constructing a multi-factor action space with disturbance adjustment capability based on state vectors, wherein the multi-factor action space comprises three action dimensions of channel adjustment factors, disturbance amplitude parameters and response threshold parameters, designing reinforcement learning strategies, taking disturbance response residual errors as feedback basis, constructing a fusion boundary constraint reward function for guiding network learning, and updating operator iteration optimization parameters by adopting strategies with disturbance direction response capability to realize cooperative improvement of dynamic adjustment capability and regional risk prediction performance of the strategies under spectrum disturbance perception guidance.
7. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion is characterized by combining a state vector generated by disturbance driving and a combined action representation, adopting a channel adjustment mechanism to conduct weighted fusion on various channel characteristics to generate an enhanced state representation with regional response capability, constructing a multi-element feature vector which fuses disturbance sensitive characteristics, channel modulation results and response mutation indexes, inputting the multi-element feature vector into a nonlinear mapping network to obtain a risk assessment representation with discriminant and stability, and introducing a channel-region combined gating mechanism to conduct adjustment on response strength in the fusion process to finally construct a characterization vector for risk score prediction as an input standard of strategy-supervision integrated optimization.
8. The storm surge and embankment risk dynamic assessment method based on multi-source remote sensing data fusion according to claim 1 is characterized in that a joint loss function is designed as an optimization target of a scoring network and a strategy network, the loss function consists of a scoring error term and a consistency penalty term, wherein the scoring error term is constructed based on square errors between real risk scores and predicted scores in multiple time steps, the consistency term is constructed based on a difference matching relation between representation changes after regional disturbance and action disturbance amplitudes, disturbance amplitude dynamic constraint factors are introduced into the joint loss to control strategy fluctuation, model stability is improved, overfitting is avoided, and stable risk response capability and disturbance adjustment consistency of output results are ensured.

Description

Storm tide and embankment risk dynamic assessment method integrating multisource remote sensing data Technical Field The invention relates to the technical field of risk prediction, in particular to a storm surge and embankment risk dynamic assessment method integrating multisource remote sensing data. Background At present, storm surge is a meteorological marine disaster which is common in coastal areas and has extremely strong destructive power, and often causes the back flow of seawater and the overflow of dykes and dams under extreme weather such as typhoons, thereby seriously threatening the safety of urban infrastructure and residents. Traditional risk assessment methods rely on limited hydrologic observation and experience models, and it is difficult to characterize the dynamic evolution process and spatial heterogeneity. Along with the development of remote sensing and open source meteorological hydrological data, synthetic aperture radar, vegetation index, terrain elevation, wind field and water level multisource data are fused, and possibility is provided for improving timeliness and space analytic capability of storm surge risk assessment. In order to improve accuracy and generalization capability of storm surge and embankment risk assessment, it is highly desirable to construct a region graph structure fused with multi-source remote sensing data, introduce a depth graph learning mechanism to realize dynamic risk identification, fully mine space-time coupling relations in the remote sensing data, effectively improve response sensitivity and assessment precision of high risk regions, and be suitable for generalization modeling and early warning application of different coastal regions. Publication number CN115840975A provides a storm surge and water increasing and embankment pre-warning scheme based on real-time water level and wave data, the accuracy of wave-crossing prediction is improved by constructing a three-dimensional tide mathematical model and a wave quantity calculation formula, the method can be used for seawall overflow risk monitoring and pre-warning system deployment, publication number CN116929709A discloses a binocular vision-based wave-crossing quantity measuring technology, two-dimensional to three-dimensional space information is reconstructed through optical distortion correction and characteristic point matching, non-contact continuous monitoring of a wave peak wave-crossing process is achieved, and the wave-crossing quantity can be directly obtained and used for model verification. The existing storm surge and embankment risk assessment method has obvious defects, generally depends on static history data or single extremum index, is difficult to characterize dynamic evolution characteristics in the storm surge development process, cannot reflect real-time surge, storm water increasing and embankment top exceeding relations in extreme weather in time, and causes assessment lag and response delay. In addition, the partial method ignores nonlinear coupling factors such as the local difference of the seawall structure, wave height change, tide level resonance and the like, has insufficient prediction precision on the surging strength and spatial distribution when the seawall structure approaches to the threshold value of the spillover dike, is easy to cause early warning misjudgment, and affects the scientificity and timeliness of coastal flood control safety decisions. Disclosure of Invention The invention provides a storm surge and embankment risk dynamic assessment method fused with multisource remote sensing data, which provides a region prediction and risk scoring combined regulation mechanism under disturbance driving aiming at the problems of undefined disturbance influence, inaccurate strategy regulation and inconsistent risk feedback of a traditional model under a complex running state, the method comprises the steps of constructing a state vector integrating residual disturbance, spectrum response and historical trend, designing a motion space of a multidimensional motion vector containing disturbance channel adjustment factors, dynamically updating a combined strategy network and a scoring network, and finally outputting a predicted result and dynamic scoring after regional disturbance, thereby realizing unified modeling and output enhancement of multi-source remote sensing data structure characteristics and state evolution characteristics. A storm surge and embankment risk dynamic assessment method integrating multisource remote sensing data comprises the following steps: s1, acquiring original remote sensing image data in different states of a storm surge area, and performing preprocessing to construct a first data set; S2, extracting a regional multidimensional eigenvalue vector based on a first data set, calculating the response deviation of each dimension of the eigenvalue vector and the neighborhood, constructing a frequency disturbance sensitive factor by combi