58 results about "Residual correction" patented technology

The invention discloses a tunnel or mine water gushing space-time prediction method and system coupled with a hydrodynamic numerical model, and the method comprises the steps: outputting multi-source data based on an identified and verified underground water numerical model, complementing the missing of measured data, quantifying the difference between the permeability characteristics of a fault and a normal stratum, and coupling the difference to a data system, and tunnel or mine excavation space data are merged. And constructing an LSTM-isolated forest-K neighbor regression coupling model, and configuring a multifunctional module to realize multi-scene data co-training. The preprocessed multivariate time series data is divided into a training set and a test set, hidden features are extracted through a coupling model, anomaly detection results are fused, a residual error correction model is synchronously trained, and hyper-parameters and weights are adaptively optimized according to multi-engineering prediction error feedback. And based on the trained coupling model, carrying out synchronous water gushing space-time prediction by adopting a window rolling strategy, and outputting prediction data meeting engineering precision in combination with residual correction. And reliable technical support is provided for safety prevention and control of engineering construction.

The invention relates to the technical field of energy prediction, in particular to an energy short-term load prediction method and system based on SE-Block improved Transform. The method comprises the steps of performing reversible normalization preprocessing based on acquired multi-element load sequence data; carrying out feature extraction and fusion on the preprocessed data by utilizing improved cross-scale interaction Patching, wherein the feature extraction and fusion comprise multi-scale feature extraction, cross-scale interaction alignment, residual error correction and dynamic fusion; and performing feature screening on the fused features based on a channel attention mechanism, wherein the feature screening comprises feature response based on improved SE-Block and non-linear interaction of context vectors. Aiming at the non-stationarity of the actual load caused by the influence of meteorological conditions and user behaviors, the model accurately depicts the fluctuation details of the load curve by automatically eliminating the noise interference among multiple variables, and the robustness of the model in the multi-element load prediction of the integrated energy system is reflected.

The invention provides an effective wave height real-time prediction method and device based on dynamic residual correction, and relates to the field of ocean engineering. The method comprises the following specific steps: acquiring wave height data and performing multi-dimensional feature screening; constructing an integrated filter fusing L1 trend filtering and variational mode decomposition, optimizing parameters by using a sea image optimization algorithm, introducing a causal sliding window to extract features so as to construct a time sequence input tensor, and inputting the time sequence input tensor into a stacked bidirectional long-short-term memory network based on an attention mechanism after noise addition standardization so as to obtain a basic predicted value; calculating a manifold coherent structure, PID dynamics and physical statistical characteristics, and cascading with the basic prediction characteristics to construct a comprehensive element characteristic vector; a LightGBM architecture is constructed, and a prediction residual error is fitted after optimization is carried out through a sea image optimization algorithm; and finally, executing linear reconstruction based on the dynamic safety threshold constraint, and outputting a real-time correction result. According to the method, the error evolution rule is deeply mined by using manifold geometric features, and the real-time precision and robustness of significant wave height prediction are remarkably improved.

The invention discloses a flood peak enhanced physical base flow and residual error correction collaborative runoff prediction method, which belongs to the field of hydrological prediction, and comprises the following steps of: dividing a training set and a verification set according to a proportion, performing oversampling processing on flood peak samples, and constructing a time sequence window; a Xinanjiang model is discretized and expressed by adopting an ordinary differential equation, rainfall and potential evaporation data are input, and intermediate variables are obtained. A physical base flow and residual error correction dual-channel module is constructed, and physical base flow and residual error correction is calculated through two full-connection networks. And calculating a final runoff predicted value by adopting a residual connection structure, taking basic NSE loss as a core, superposing a flood peak sample error weighted item, strengthening flood peak fitting precision, and updating physical parameters and neural network weight through a back propagation algorithm. And verifying the model, and respectively calculating prediction indexes of the training set and the verification set. According to the method, fusion of a traditional hydrological model and a deep learning method is realized, the physical interpretation of the model is enhanced, and the basin runoff prediction precision is improved.

The invention discloses a significant wave height two-stage space-time prediction method and system based on diffusion residual correction, and relates to the technical field of sea wave prediction, historical significant wave height data and corresponding historical wind field data are input into a significant wave height space-time prediction model for processing, and a preliminary significant wave height prediction result is obtained; and inputting the initial prediction error, historical significant wave height data and synchronous forecast wind field data into a residual error correction module based on a diffusion model, and compensating the initial prediction error to obtain a final significant wave height prediction result. A diffusion model based on an EDM framework is innovatively introduced to carry out refined reconstruction on a prediction residual error, and a basic prediction field, a synchronous wind field dynamic factor and a prediction aging code are used as multi-dimensional physical strong conditions to be injected into a diffusion generation process. According to the method, high-frequency texture details and non-linear fluctuation components missed by the first-stage model can be certainly recovered, long-term deviation accumulation is relieved to a certain extent, the high-wave region prediction quality is improved, and the prediction stability is improved.

The invention discloses a deep learning method for uniform background light and shadow based on a Transform model. The method comprises the steps that an input image and a corresponding portrait main body mask are acquired, and the input image comprises an area with uneven background light and shadow; preprocessing the input image and the portrait main body mask, including image standardization and feature fusion, to obtain a fused feature map; based on a patch segmentation method, converting the fusion feature map into sequence features; encoding the sequence features to output enhanced sequence features, modeling a light and shadow distribution dependency relationship of full image pixels through a global attention mechanism, and distinguishing portrait main body features and background features based on the portrait main body mask; and based on the inverse logic of patch segmentation, recovering the enhanced sequence features into a spatial feature graph and the like. According to the method, semantic mask prior, global context modeling and adaptive residual correction are organically integrated, and a solution is provided for solving the core problem in background light and shadow homogenization.

The invention belongs to the technical field of intersection of inertial navigation and deep learning, and discloses a missing data interpolation method of an inertial measurement unit, which comprises the following steps: constructing a baseline interpolation sequence based on an original IMU observation sequence with missing, and observing by taking the difference between the two as a residual error; constructing missing perception information based on an original IMU observation sequence, and splicing the missing perception information with residual observation to form an enhanced feature vector; constructing a parallel bidirectional LNN branch and a bidirectional LSTM branch which are respectively used for extracting an enhanced feature vector to obtain a double-path residual prediction item and generating a hidden state; and taking the hidden state, uncertainty estimation and an original IMU observation sequence as input, calculating an adaptive fusion weight and fusing a double-path residual prediction item to obtain a residual correction item, and superposing the residual correction item to a baseline interpolation sequence to output an interpolation result. The invention also discloses a missing data interpolation device of the inertial measurement unit. When IMU missing data is processed, the interpolation physical consistency can be enhanced, the perception capability is improved, and the interpolation precision is effectively improved.

The invention discloses a stress detection method and device for a pipeline and a medium, and relates to the technical field of structural health monitoring, and the method comprises the steps: collecting the geometric dimensions and material parameters of the pipeline, building a digital twin model, completing the zero calibration and residual strain baseline setting, and obtaining a parameter baseline set; performing synchronization, noise filtering and restoration on the multi-source observation data by using the parameter baseline set to generate a data frame with a quality mark; the initial calculation stress is obtained based on temperature compensation and residual correction, the initial calculation stress is compared with a digital twin model to recognize the load and disturbance, corrected stress is obtained, the corrected stress and the initial calculation stress are subjected to fusion estimation, and a fusion stress result and uncertainty are output; risk indexes are calculated and graded according to the stress anomaly and the acoustic emission event, residual life and maintenance suggestions are given, and an audit record is formed. According to the invention, synchronization, noise filtering and deletion repairing are carried out on the multi-source observation data based on the parameter baseline set, so that high reliability of stress detection data and improvement of anti-interference capability are realized.

The invention relates to the technical field of machine vision and optical measurement mechanics, and discloses a deformation identification and measurement method based on machine vision and deep learning, and the method comprises the following steps: S1, system calibration and base library construction; s2, offline training of the hybrid model; s3, real-time decoupling and coefficient regression are carried out; and S4, final assembly and physical quantity calculation of the deformation field. According to the method, a deformation field is decoupled into a global nonlinear deformation field () and a residual deformation field () through S3, and the global nonlinear deformation field () and the residual deformation field () are superposed. The physical significance and the stability of a deformation main body are ensured by utilizing the physical deformation base library constructed in S1 and model reconstruction; meanwhile, a luminosity residual image () generated in the S3 is used for driving module reconstruction, local high-frequency disturbance which is not covered by a physical model is accurately compensated, a high-dimensional deformation field is decoupled into low-dimensional global physical prior regression and residual correction, and efficient real-time measurement is achieved; meanwhile, numerical difference is replaced by analytic derivation, so that the signal-to-noise ratio and the measurement precision of the strain field are remarkably improved.

The application discloses a tunnel or mine gushing water space-time prediction method and system coupled with a water power numerical model, and the method comprises the following steps: based on the identified and verified underground water numerical model, outputting multi-source data, complementing the missing measured data, quantifying the difference between fault and normal stratum permeability characteristics and coupling to the data system, and incorporating the tunnel or mine excavation space data. An LSTM-isolation forest-K nearest neighbor regression coupled model is constructed, and a multifunctional module is configured to realize common training of multi-scene data. The pretreated multivariate time series data is divided into a training set and a test set, hidden features are extracted through the coupled model, abnormal detection results are fused, a residual correction model is trained synchronously, and the hyperparameters and weights are adaptively optimized according to the multi-project prediction error feedback. Based on the trained coupled model, a window rolling strategy is adopted to carry out synchronous gushing water space-time prediction, residual correction is combined, and prediction data meeting the engineering precision is output. Reliable technical support is provided for engineering construction safety control.

This invention belongs to the fields of new energy technology and artificial intelligence technology, and provides a high-resolution long-term photovoltaic power prediction method and system based on a dual-branch architecture. By constructing a parallel dual-branch architecture, the baseline branch extracts multi-scale features, and the ramp branch identifies power ramping events. The features from both branches are embedded through multiple channels and then fed into a BiGRU-Enhanced Transformer model with shared weights. By strategically deploying BiGRU layers at the input and between the encoder and decoder, local instantaneous dependencies and global long-term trends are captured simultaneously. A multi-task joint optimization loss function is designed for the ramp branch, using weighted cross-entropy and... L The probability, temporal location, and fluctuation amplitude of slope occurrence are jointly optimized using the L1 norm. Finally, nonlinear residual correction is applied to the outputs of the two branches to achieve the final prediction. This invention effectively overcomes the oversmoothing effect of deep learning models, demonstrating extremely high capture accuracy and robustness in long-term prediction of one month's worth of data at 1-minute resolution.

The invention relates to the technical field of artificial intelligence model optimization, in particular to a large model reasoning acceleration method and device for packet sensing quantization and residual correction, and the method comprises the steps: carrying out the statistical analysis of the weight and activation of each layer of a large model, and generating a channel feature matrix; constructing a learnable grouping mapping matrix, dividing channels into different groups, and dynamically distributing quantized bit width to obtain a grouping weight matrix; on the basis of the channel weight-activation joint sensitivity, calculating each grouping error contribution on line by using a small prediction model, and adjusting a grouping weight matrix to generate an optimized weight matrix; constructing an error control matrix to dynamically adjust the quantization error along the propagation path, and generating a correction matrix; dynamically adjusting the sparse rate and the quantization precision according to the channel feature matrix and the hardware constraint by combining a structured sparse strategy, and generating a sparse quantization matrix; in the reasoning process, model reasoning is carried out according to the correction matrix and the sparse quantization matrix, and the weight matrix and the correction matrix are updated and optimized in a closed-loop mode.

The application relates to the technical field of data processing, and discloses an application control method considering data collaborative prediction and adaptive correction, which collaboratively predicts the data amount of a prediction period through a baseline prediction model and a residual correction model, can significantly improve the data amount prediction accuracy, and then determines the deviation feature of a target prediction period according to the predicted data amount of the target prediction period and the actual data amount of the target prediction period, so that the correction decision basis can be obtained, the deviation scene adapted by the current prediction period can be accurately identified based on the deviation feature, and the first prediction sequence is adaptively corrected according to the deviation scene, so that the second prediction sequence is obtained. It can be seen that the layered correction mechanism not only avoids frequent global correction, prevents overcorrection or undercorrection, but also ensures that the actual prediction scene change condition can be quickly responded, has strong real-time adaptive capacity, and thus the control accuracy can be effectively improved.

The application discloses an effective wave height double-stage space-time prediction method and system based on diffusion residual correction, relates to the technical field of sea wave prediction, and inputs historical effective wave height data and corresponding historical wind field data into an effective wave height space-time prediction model for processing to obtain a preliminary effective wave height prediction result, and then inputs the preliminary effective wave height prediction result, the historical effective wave height data and synchronous forecast wind field data into a residual correction module based on a diffusion model to compensate for the preliminary prediction error and obtain a final effective wave height prediction result. The diffusion model based on the EDM architecture is innovatively introduced to finely reconstruct the prediction residual, the basic prediction field, the synchronous wind field dynamic factor and the prediction time limit are taken as multi-dimensional physical strong conditions to inject the diffusion generation process, the application can deterministically restore the high-frequency texture details and nonlinear wave components missed by the first-stage model, to a certain extent, the application can relieve the long-term deviation accumulation, improve the prediction quality in high wave areas and improve the prediction stability.

This invention discloses a method for designing the composition of nickel-containing high-strength material systems based on transfer learning and the barrel theory optimization. The method includes: first, using a recursive feature elimination algorithm to screen key features across source domain benchmark data and target domain experimental data containing sparse components; second, constructing a deep neural network for pre-training on the source domain dataset, extracting nonlinear feature representations between components and performance through hidden layers; third, freezing the weights of the feature extraction layers of the pre-trained model, and for sparse samples in the target domain, introducing Bayesian ridge regression to construct a residual correction model to achieve cross-domain knowledge transfer; and finally, combining multiple metallurgical constraint criteria to perform an evolutionary search within a limited composition space. This invention solves the problem of effectively modeling sparse data in new material systems. By combining deep feature extraction with Bayesian residual correction, it significantly improves the robustness of high-strength material performance prediction and enables precise optimization of industrial-grade composition formulations.

The invention provides a salinization type quantitative inversion method based on feature weighting and residual error correction, and relates to the field of remote sensing technology and machine learning, and the method comprises the steps: collecting actual measurement concentration data of soil salt ions in a target region, and obtaining a remote sensing image of a corresponding region; extracting feature information data from the remote sensing image; constructing a salinization type quantitative inversion data set through the feature information data and the measured data; evaluating the correlation between each feature and the soil salinity by using a feature selection method to obtain a feature importance weight vector; the first stage is used for preliminary prediction of the soil ion concentration, the second stage is used for introducing a residual learning mechanism, the first stage residual error is fitted to improve the prediction precision, and ion concentration data of the target area are predicted and generated; and according to an inversion result, obtaining a two-dimensional space-time distribution diagram of soil salinization type quantitative classification. The technical scheme of the invention is suitable for high-precision identification and classification management of regional scale soil salinization types.

The application discloses a kind of based on contrast learning's weld pipe unit abnormality detection method, to solve the time scale drift of multiple sensing channels, cross-station alignment and the lack of physical consistency, transient interference causes detection instability and inaccurate positioning problem, the present application is through the time scale correction, time to cumulative length mapping and encoder boundary constraint, neural residual correction and alignment confidence evaluation, fixed length grid constructs physical consistent contrast sample, memory bank and dynamic threshold detection and causal online fine-tuning rollback, realizes cross-station unified alignment and station level positioning, improves the technical effects of abnormality detection accuracy and robustness and maintains online stability.

The invention discloses a facility damage identification and evolution prediction system based on physical information generation, and belongs to the technical field of structural health monitoring and digital twinning. In order to solve the problems that in the prior art, a data-driven model depends on a large amount of real labeled data, and a physical model is difficult to capture a nonlinear effect and lacks self-adaptive ability, the system comprises a physical information generation module for generating a synthetic data set based on the physical model to train an artificial intelligence model; the real-time state diagnosis module is used for receiving a real observation image and diagnosing a current internal state field; and the double-flow evolution prediction module is used for generating prediction of the physical flow and the data flow at the next moment in parallel based on the current state. According to the method, the problem of data sparsity is solved through physical information generation, prediction accuracy is improved through dynamic residual correction, long-term self-adaption of the model is achieved through a closed-loop self-evolution mechanism, and robustness and interpretability of the system are enhanced.

The application discloses a surface roughness prediction method based on a diffusion model and neural structure search, performs quantile normalization on process parameters, and performs Z-score standardization on surface roughness values; synthetic process parameter data meeting physical consistency and distribution fidelity is generated; a balanced, high-variety mixed training set is constructed to alleviate the data scarcity problem in a small sample scene; a Gaussian process-based Bayesian optimization strategy is adopted to automatically search for an optimal prediction network architecture matching the data complexity; then residual correction and structure fine-tuning are performed to improve the generalization ability and prediction robustness of the model under a small sample, and a surface roughness prediction value is output. The generalization ability and robustness of the model are improved, higher-precision surface roughness prediction can be realized, thereby providing a general solution for small-sample, high-dimensional and strong nonlinear process modeling in the field of precision manufacturing, and the surface quality prediction of other precision machining processes can be further popularized.

This invention discloses a method for dynamic path optimization of an inspection robot based on reinforcement learning, specifically including: S1, constructing a set of path structural units; S2, constructing a path structural state vector; S3, constructing a set of structural actions; S4, performing continuous-time mapping on the path structural state vector using a liquid time constant neural network that introduces a structural residual correction time constant; S5, selecting structural actions based on the output of the liquid time constant neural network and reconstructing the path structural units; S6, collecting information on structural cost changes after path structural unit reconstruction and constructing a reward vector; S7, performing perturbation sampling and weighted update of the gate parameter subset of the liquid time constant neural network using a natural evolution strategy that introduces hierarchical noise in structural actions. This invention uses structural-level reinforcement learning for dynamic path optimization.

The application discloses an internet information system integrated service system and belongs to the technical field of internet services; the system is used for solving the technical problems of low prediction accuracy and incapability of supporting dynamic scheduling caused by single model and insufficient data preprocessing in the prior art; through full-link collaborative design of multi-source data acquisition, singular spectrum analysis multi-scale decomposition, gradient boosting decision tree main model and bidirectional LSTM residual correction model cooperation and multi-step prediction, the defects of single feature dimension, insufficient data non-stationary processing, single model limitation and single prediction dimension in the prior art are solved, and the collaborative effect of all links makes the model prediction accuracy, real-time performance and decision scientificity significantly better than those of the traditional scheme, thereby providing reliable technical support for performance optimization of the internet information system integrated service.

The invention discloses a flow field prediction method and system based on wavelet alignment cross attention, and the method comprises the steps: enabling a computational domain structure grid coordinate and a working condition parameter to form a multi-channel input tensor, and constructing a flow field prediction model to achieve the end-to-end regression of a flow field variable. In a bottleneck layer of the flow field prediction model, a first branch generates a key / value based on a high-frequency sub-band of an input feature through discrete wavelet transform; the second branch generates a query based on an input feature through a linear self-attention aggregation result; the fusion branch executes linear cross attention, and band-by-band soft mask and lightweight residual error correction are applied to high-frequency sub-bands based on the linear cross attention; a hierarchical cross attention refinement module is introduced into a decoding sub-module, after query, keys and values are generated based on high-frequency coding features and high-frequency decoding features, band-by-band enhancement and inverse transformation reconstruction are carried out according to coefficient domain cross attention, and jump connection features are fused through symmetric gating.

This invention relates to a method and system for reconstructing airfoil flow fields based on boundary observation guidance and interpolation residual correction. The method includes the following steps: S1: Acquire airfoil geometry, operating parameters, flow field query points, and sparse field observations and / or sparse surface pressure observations; S2: Generate observation value channels and observation mask channels to form field point node features and boundary node features; S3: Construct a field point map and perform spatial neighborhood message passing; S4: Construct a boundary observation map and propagate airfoil geometry and surface pressure information; S5: Selectively inject boundary observation representations into field point representations through gated boundary-to-field point fusion; S6: Construct interpolation priors, predict residual correction amounts using graph networks, and obtain a complete high-resolution airfoil flow field through replacement. This invention can uniformly handle flow field super-resolution, surface pressure-assisted reconstruction, and hybrid observation reconstruction, reducing observation and computation costs and improving reconstruction accuracy in unobserved and near-wall regions.

This invention provides a method and related device for real-time prediction of rock mass strength in tunnel excavation advance drilling. By acquiring drilling parameters during the advance drilling process, a mechanism characteristic representing rock breaking strength and energy consumption is constructed based on these parameters. This mechanism characteristic is then input into a baseline prediction model, a residual correction branch, and a physical consistency branch. A fusion gating network generates a first fusion weight corresponding to the residual correction and a second fusion weight corresponding to the physical correction. Based on the baseline prediction value, the residual correction, the physical correction, and the first and second fusion weights, the final predicted value of the uniaxial compressive strength of the rock mass is determined. This invention uses drilling mechanism characteristics as input to construct a collaborative prediction framework consisting of a baseline prediction model, a residual correction branch, and a physical consistency branch. A dual-gating mechanism enables sample-level adaptive fusion and robust output, thereby simultaneously improving prediction accuracy, physical rationality, and engineering applicability.

The application discloses an NDVI downscaling method based on random forest and nonlinear residual correction, belongs to the field of ecological remote sensing and geoinformation technology, and comprises the following steps: 1, constructing a random forest model; 2, multi-source data preprocessing and feature grid construction, generating a total sample data set for random forest model training and prediction; 3, training the random forest model to obtain an optimal random forest model; 4, NDVI downscaling prediction, applying the trained random forest model to a high-resolution feature variable data set to obtain a high-resolution annual average NDVI prediction result; 5, performing nonlinear residual correction on the optimal random forest model based on GAM; and 6, applying the constructed nonlinear residual correction model to all downscaling NDVI data to obtain a corrected NDVI result. The application adopts the above method, and the spatial consistency, time continuity, overall precision and stability of the downscaling result are improved.

The invention discloses a self-adaptive double-layer simulation method and system based on intelligent optimization and residual correction, and belongs to the technical field of system simulation. The method comprises the following steps: receiving a simulation task and intelligently deciding a simulation strategy; when the simulation strategy decision is double-layer simulation, global scanning and automatic optimization are carried out through first-layer rapid simulation, and a key area is identified; performing refined simulation on the key area by fusing a physical information neural network through second-layer precise simulation; and finally fusing two layers of results. In addition, the system is provided with an outer ring residual error correction link, and model parameters are corrected online and model drift is resisted by comparing a simulation result with real data. Through double-layer cooperation and closed-loop correction, the contradiction between precision and efficiency in the simulation process is effectively solved, the self-adaptive capacity and long-term reliability of simulation are improved, and the method is particularly suitable for full-life-cycle simulation of complex systems such as a ship power system.

The application discloses a numerical control rotary table time grating position precision self-evolution compensation method combining a mechanism model and reinforcement learning, and comprises the following steps: collecting time grating sensor original signals and rotary table running states; performing mechanism rough compensation through a harmonic error model to obtain rough compensation position values; forming a state vector by combining the rough compensation position values and the running states, inputting the state vector into a deep Q network intelligent agent, and outputting compensation residual correction values; superimposing the rough compensation values and the correction values to obtain final compensation values; calculating a reward according to a deviation from a high-precision reference, updating network parameters, and realizing self-evolution; and providing a corresponding compensation system based on the method. The application adopts a double-loop architecture to combine mechanism knowledge and a data-driven method, realizes online adaptive compensation of the numerical control rotary table time grating position precision through a composite reward function and a priority experience playback mechanism, and has the advantages of high compensation precision, strong adaptability and good training safety.

The application discloses a deep reinforcement learning-based oil and gas pipeline corrosion prediction method and platform, and relates to the technical field of oil and gas pipeline corrosion prediction. The method comprises the following steps: collecting pipeline operation related time series data through a multi-dimensional sensor, reconstructing non-equidistant data through an improved signal processing technology, constructing a deep Q network prediction framework of a time series residual error corrosion state prediction algorithm, determining an optimal model through multi-group super parameter combination iterative training optimization, outputting a normalized corrosion rate estimation value combined with a residual error correction mechanism, and obtaining an actual corrosion rate prediction result through inverse transformation; and performing specific processes such as prediction framework modeling, network structure construction, residual error algorithm embedding, super parameter optimization, test data processing and prediction result correction, and the platform realizes the whole-process closed-loop operation of data acquisition, processing, modeling, prediction and analysis relying on each functional unit. The application significantly improves the accuracy and time series adaptability of corrosion prediction, and provides technical support for the safe operation of oil and gas pipelines.

This invention proposes a coarse prediction-guided conditional residual diffusion model channel prediction method. The method first obtains a coarse prediction of the future CSI based on historical Channel State Information (CSI) sequences. Then, using this coarse prediction as the prior mean of the reverse refinement process, a conditional residual diffusion forward process is constructed, transitioning from the true future CSI value to the noisy state in the coarse prediction neighborhood. Subsequently, a conditional reverse refinement network is trained to estimate residual correction terms and noise removal terms based on the diffusion state, diffusion step index, and relevant conditional information. Finally, a denoising diffusion implicit model (DDIM) is used for a few steps of reverse update to obtain the future CSI prediction result. This invention can recover fine-grained random variation information missed in the coarse prediction result while maintaining the overall evolution trend of the future channel, improving the prediction accuracy of complex time-varying channels. Furthermore, by constraining the reverse process within the coarse prediction neighborhood, it enhances generation stability, reduces the number of reverse sampling steps, and lowers inference latency.

The application relates to the technical field of image processing, and discloses a single-image super-resolution processing method and device of an adaptive residual correction network, which comprises the following steps: performing feature extraction on a to-be-processed image to obtain initial image features, wherein the resolution of the to-be-processed image is a first resolution; performing multi-level adaptive feature fusion on the initial image features to obtain multi-level intermediate features; in each level of adaptive feature fusion, residual features generated by non-linear inference are corrected, the residual features contain heterogeneous features caused by mapping errors, the correction process converts the heterogeneous features into homogeneous features through a residual correction submodule, and residual deviations of the residual features are eliminated based on the homogeneous features; and performing image recovery according to the multi-level intermediate features to obtain a target image, wherein the resolution of the target image is a second resolution, and the second resolution is greater than the first resolution. The application realizes a single-image super-resolution processing method which takes into account both high fidelity and low calculation cost.