380 results about "Linear relationship" patented technology

The invention discloses a diesel generator turn-to-turn short circuit fault protection method and system based on multi-parameter characteristics, and relates to the technical field of intelligent fault diagnosis of power equipment, and the method comprises the steps: carrying out the feature extraction of operation state monitoring data, carrying out the integration according to a time sequence, and generating a real-time feature sequence set; performing spatio-temporal feature coding and fusion on the real-time feature sequence set to generate a multi-dimensional spatio-temporal feature map; mapping the comprehensive matching score set into fault confidence according to a linear relation, obtaining a fault confidence sequence, judging a fault level in combination with a characteristic amplitude overrun condition, and generating a fault diagnosis result set; and executing hierarchical response based on the fault diagnosis result set, generating a hierarchical protection action execution record, and generating a short-circuit fault protection report in combination with the fault diagnosis result set. According to the method, quantitative similarity analysis of real-time features and standard modes is realized, the fault probability is accurately evaluated through a multi-parameter weighted fusion mechanism, and the reliability and interpretability of a diagnosis result are improved.

The present application relates to the technical field of storage, and provides a method and apparatus for determining an optimal read voltage parameter of a flash memory, and a flash memory device. The method comprises: on the basis of a linear relationship model between a read retry voltage offset value and a bit flip count, acquiring a new read retry voltage offset value corresponding to a minimized predicted bit flip count of the flash memory; determining whether data can be successfully read under the new read retry voltage offset value; if yes, adding the new read retry voltage offset value to the header of a read retry offset table to form a new read retry offset table; and on the basis of the new read retry offset table, updating the linear relationship model by means of an adaptive learning model. By means of the efficient and adaptive voltage parameter determination method, the present application can improve the success rate of data recovery and reduce performance loss.

The invention relates to the technical field of data reasoning, and discloses an AI-based power load prediction method, which comprises the following steps: compensating a missing value in target power data with a unified timestamp to obtain a standardized data set of a target power grid, and adaptively extracting a multi-dimensional feature vector; predicting a historical power load based on a training set in the historical power data set, and when an error value between the historical power load and a verification set in the historical power data set is lower than a load threshold value, obtaining a load prediction weight; fitting a linear relationship between the temperature change in the historical power data set and the error value, and representing the slope of the linear relationship as a compensation coefficient; predicting the power load condition of the target power grid, and correcting the load fluctuation of the power load condition based on the compensation coefficient; encoding the corrected power load condition into a power data packet, and transmitting the power data packet to the historical power data set in S2 to obtain a target load prediction report; according to the invention, the accuracy of power load prediction can be improved.

The invention relates to the technical field of coupling alignment of single-mode optical fibers, in particular to an optical fiber coupling alignment method and system based on wavefront aberration and a sensitivity matrix. The system comprises an optical measurement module, a processing and control module and a multi-axis precision adjustment platform, and the alignment method comprises the steps that firstly, an interference fringe image is obtained and processed through an interference light path, continuous wavefront is recovered through Fourier transform and phase unwrapping, and a Zernike coefficient for quantizing wavefront distortion is obtained through Zernike polynomial fitting. A sensitivity matrix representing the linear relation between the degree-of-freedom adjustment amount and Zernike coefficient changes is established and serves as a physical guide item to introduce a speed updating formula of a particle swarm optimization algorithm, an SG-PSO algorithm is formed, a particle swarm is guided to conduct directional intelligent search and rapidly output the optimal adjustment amount, and finally a multi-axis precision adjustment platform is driven to complete alignment. According to the method, optimization of wavefront sensing and physical model guiding is effectively fused, and the convergence speed, the alignment precision and the global search capability are remarkably improved.

The invention discloses a low-voltage power distribution network line loss prediction method, a model training method and a related device, and the model training method comprises the steps: (1) employing a neural network structure combining a time sequence feature extraction module and a multi-scale feature extraction module, and capturing the long-term trend of photovoltaic output and load fluctuation, line loss change modes under different time scales are effectively modeled; (2) a deep feature extraction module introduces a deep feature extraction mechanism with mutually different architectures, and the expression ability and generalization performance of the model to a complex nonlinear relationship are improved; (3) a crown porcupine optimization algorithm is adopted to perform joint optimization on model parameters and structures, efficient search can be realized in a high-dimensional non-convex space, local optimum is avoided, and prediction precision and stability are improved; based on the line loss prediction model, the line loss prediction method provided by the invention can be directly used for power distribution network operation situation evaluation and energy efficiency analysis, and has good engineering application value under the background of wide distributed photovoltaic access.

The invention provides an insect pest prediction method based on PSO-LSTM, belongs to the technical field of agricultural information, and aims to solve the problems that a single machine learning model is adopted in a traditional insect pest prediction method, the long-term modeling capability of time sequence data is insufficient, the complex nonlinear relation between the number of insect pests and meteorological factors is difficult to capture, and the prediction precision is low. Comprising the following steps: S1, collecting insect pest data and establishing a multivariable time sequence data set; s2, data preprocessing; s3, feature selection; and S4, establishing an insect pest prediction model, optimizing the insect pest prediction model by using a PSO module, and obtaining a prediction value of the number of insect pests at the next moment based on the optimized insect pest prediction model.

The invention discloses a design method of a pressure control throttle valve with the pressure drop linearly changing along with the stroke, and belongs to the technical field of automation of oil and gas exploitation equipment.The design method comprises the following steps that the linear relation between the pressure drop and the stroke of the throttle valve serves as a target; a mathematical model of a valve element contour is established by establishing a minimum throttling area equation of a cylindrical valve and a wedge-shaped valve, correcting a flow pressure drop characteristic equation and deducing a differential pressure linear adjustment equation, a valve element contour curve meeting the linear requirement is deduced through variable-diameter contour optimization design, and therefore the valve element structure of the throttling valve is obtained. By means of the design method, the valve element contour curve with the wider linearity range of pressure control adjustment and the better precision can be obtained, and the design method is more efficient and universal.

The invention discloses a crop yield prediction method based on remote sensing inversion, and the method comprises the steps: obtaining historical remote sensing image data and historical ground actual measurement data, and building a standard value library based on the historical crop yield data, the vegetation features of different crops, and the linear relation between the vegetation features in the same year and the crop yield data; constructing a crop yield estimation model, and training the crop yield estimation model to obtain a trained crop yield estimation model; obtaining current-year remote sensing image data, and inputting vegetation features obtained by the current-year remote sensing image data into the trained crop yield estimation model to obtain a preliminary yield estimation value; and obtaining agricultural meteorological data of the current year, correcting the preliminary yield estimation value according to the agricultural meteorological data, and obtaining a target yield estimation value, thereby being beneficial to improving the precision of crop yield estimation.

The invention provides a security analysis method for a post-quantum NTRU cryptographic algorithm, and the method comprises the steps: an attacker carries out the sampling of a public key, and obtains a public key sample, different private keys g with a public Hamming weight, and a fixed target private key; according to each public key sample and information disclosed by the Hamming weight of the private key, establishing a modular equation, and solving the modular equation by combining a linear technology and a preset initial value; obtaining a complete value of the second polynomial based on a linear relation between the equation set solution and other parts except the constant term in the second polynomial and a linear relation between a preset initial value and the constant term of the second polynomial; and according to the public key sample and the complete value of the second polynomial, utilizing a GS algorithm to obtain a speculative value of the target private key, and according to the relationship between the speculative value and the actual value of the target private key, carrying out security analysis on the password security system. The private key is recovered by using the relation between the solution of the modular equation and the private key, and the bottleneck of the existing technical means is broken through.

The invention relates to a wind noise modeling method and system based on machine learning and application, and belongs to the field of marine acoustics and environmental noise modeling, and the method comprises the steps of data preprocessing and feature extraction, hybrid network model building, hybrid network model training and verification. According to the method, a hybrid network model of multiple linear regression and a multi-layer perceptron is constructed based on a known physical mechanism of wind noise and two main noise generation mechanisms of surface turbulence and bubble oscillation, a linear relation and a non-linear relation are modeled respectively, outputs of the two models are integrated through a weighted fusion strategy, and a multi-layer perceptron model is constructed. Smooth transition modeling from a low-wind-speed linear relation to a high-wind-speed nonlinear relation is achieved.

The invention discloses a humidity self-adaptive calibration method of a resistance type gas sensor. The method comprises the following steps: firstly, measuring basic resistance values of a gas sensor under different humidities and response resistance values of target test gases with different humidities and different concentrations; fitting a linear relation between a gas response resistance change value and a basic resistance value under a fixed target test gas concentration to obtain a proportionality coefficient and a bias coefficient; fitting a function relationship among the target test gas concentration, the proportionality coefficient and the bias coefficient through a linear regression method, and calculating the actual concentration value of the target test gas under different humidity conditions by using the function relationship, so as to adaptively calibrate the humidity drift of the gas sensor. By revealing the linear relation between the basic resistance value and the gas response resistance change value when the humidity changes, the self-calibration method which does not need external hardware and is low in calculation complexity is achieved, and a reliable low-cost solution is provided for actual deployment of the sensor.

The invention belongs to the technical field of wind power prediction, and provides a double-layer cascade wind power prediction method and system based on TransformerEncoder, and the method comprises the steps: a fusion variable selection module comprises a Pearson correlation coefficient and an XGBoost regression model; the data processing module preprocesses the input original data to obtain normalized data; a fusion variable selection module screens normalized data, a Pearson correlation coefficient quantifies a linear relation of the normalized data, an XGBoost regression model captures a nonlinear relation, and a correlation feature sequence is obtained through fusion; and the model construction and prediction module processes the related feature sequence, predicts and optimizes a power value, and outputs an optimized power prediction value. According to the method, efficient screening of key variables is realized, change rules of meteorological data, fan speed data and power data are effectively identified, redundant variables are effectively reduced, the complexity of the model is reduced, dynamic characteristics of a wind power system can be better understood and learned, and higher precision and stability are shown in an actual prediction task.

The invention provides a deep water gravity flow gas field reservoir water saturation earthquake prediction method based on TDCN-TransNet. The method comprises the following specific steps: a TDCN-TransNet network architecture comprises an input processing layer, a dual TWT position code, a TDCN module, a Transformer encoder layer and an output layer; mapping the features to a hidden dimension in an input processing layer through linear full connection; a dynamic Gaussian convolution kernel and a CBAM attention mechanism are embedded in the TDCN, and modeling of a nonlinear relation between an elastic parameter and saturation is enhanced through multi-scale time sequence feature extraction; tCBlock is integrated on a Transform encoder layer, and long-distance sequence dependence is captured by using attention pooling and an SE channel excitation mechanism; a TWT position code is introduced between the TDCN and the Transform, and sequence position information is enhanced; constructing a combined loss function of a mean square error and L2 regularization; finally, training and reasoning are conducted based on the model, and high-precision water saturation prediction is achieved. The method provided by the invention has high generalization and accuracy, and shows excellent prediction performance in a complex geological environment of a deepwater gravity flow reservoir.

The invention provides a rectifying tower parameter intelligent control method for waste liquid treatment, and relates to the technical field of rectifying tower intelligent control. Comprising the steps of collecting influence data in a waste liquid treatment process in real time; analyzing energy utilization efficiency and mass transfer efficiency in the waste liquid rectification and purification process according to the influence data, calculating a reliability factor of key equipment, and constructing a linear relation; constructing a rectification purification mode according to the linear relation and the reliability factor of the equipment, and optimizing the running parameter ratio of the equipment by using an improved self-adaptive genetic algorithm; the optimal operation parameter ratio is converted into a control instruction, and the control instruction is executed; product quality data and equipment operation state data in the waste liquid treatment process are monitored in real time, the deviation value between actual data and an expected target is analyzed, and the operation parameter ratio of all equipment is adjusted according to the actual situation. According to the invention, a rectification purification mode is constructed, and an improved self-adaptive genetic algorithm is used to optimize the operation parameter ratio of each device, so that the intelligent control of waste liquid rectification is realized.

The invention relates to the technical field of river ice condition early warning, and discloses an ice plug prediction method based on fusion of a physical model and machine learning, and the method comprises the steps: obtaining and preprocessing the monitoring data of a to-be-predicted river reach; carrying out parallel calculation on a physical risk index based on a physical mechanism and a machine learning prediction probability based on data driving; performing weighted fusion on the index and the probability to obtain a basic risk index; then, performing multi-stage correction on the basic risk index based on the critical jamming condition and the river channel curvature to obtain a final risk index; and finally, determining and outputting an ice plug risk grade according to a preset threshold value. The mechanism definition of a physical model is combined with the ability of a machine learning model to capture a complex nonlinear relationship. By introducing critical jamming condition correction and river channel geometric feature reinforcement, the reliability of the prediction method under extreme working conditions is ensured, the applicability and prediction precision of the model to different river channel conditions are enhanced, and the comprehensive accuracy of a prediction result is improved.

The invention relates to the technical field of oil content measurement, in particular to a method for measuring the oil content of oil-based drilling cuttings and application of the method. The method comprises the following steps: querying a dynamic extraction curve database according to sample lithology and a processing method, and determining an optimal extraction time point by calculating a concentration change rate; multi-point sampling is carried out in the extraction process, the environment temperature and pressure are measured at the same time, an environment influence model comprising a temperature influence sub-model and a pressure influence sub-model is established, and an environment correction factor is obtained; performing infrared spectrum measurement on the extract sample, performing data correction, and calculating an average value and a standard deviation to remove an abnormal value; and finally, calculating the final oil content by adopting a piecewise linear equation aiming at different concentration ranges. By considering the lithologic characteristics, the environmental influence and the segmented linear relation, the measurement precision and reliability are remarkably improved.

The invention discloses a cracking zone tunnel blasting construction safety risk grading evaluation method and system, and belongs to the technical field of tunnel engineering safety monitoring. According to the method, dimensionality reduction and conflict resolution are performed on the original indexes by adopting the PCA-EWM method, and the mutually independent composite indexes among the indexes are obtained, so that the independence and effectiveness of the indexes are improved, and the problem that the final evaluation result is inaccurate due to the fact that the information among the original indexes is overlapped and the weight distribution is inevitably caused by repeated distribution is avoided; the weight is dynamically adjusted through the offset of the criterion layer so as to adapt to the complex engineering environment, and it is ensured that the evaluation result can reflect the current risk state; due to the fact that the composite index of tunnel blasting and the construction risk degree are in a nonlinear relation, a normal cloud model is adopted to reconstruct a membership function and process nonlinear risk evolution of the index so as to better adapt to nonlinear characteristics of the tunnel blasting construction data of the cracked zone, the accuracy of risk assessment is improved, and the safety of construction is guaranteed.

The invention relates to a method and a device for determining a signal detection network. The method comprises the following steps: determining sample communication parameters of a sample transmitting end and a sample receiving end communicating through a sample intelligent reflecting surface (IRS) (S101); and training an initial neural network based on a training sample set formed by the sample communication parameters to obtain a signal detection network, the input of the initial neural network being the sample communication parameters, and the output of the initial neural network being an estimated value of a transmission signal of the sample transmitting end (S102). According to the method and the device, the initial neural network can be trained based on the training sample set to obtain the signal detection network, and the neural network is a nonlinear algorithm and is not limited to operation of a linear relation, so that the method and the device can be effectively used for signal detection even in a complex communication environment with more nonlinear factors; and a high-precision estimation result is obtained for a transmitted signal.

The invention discloses a block identification-based low-voltage transformer area topology modeling method and device and a storage medium, and the method comprises the steps: carrying out the correlation analysis of a target voltage time sequence of a branch box and a user based on the target voltage time sequence of the branch box and the user, obtaining the affiliation of the user and the branch box, and completing the preliminary blocking; constructing a data matrix used for branch box-electric meter box topology identification, and based on the target power time sequence of the branch box and the electric meter box, calculating to obtain a relation matrix representing a topological relation between branch box nodes and electric meter box nodes; a data matrix used for electricity meter box-user topology identification is constructed, and a relation matrix representing the topological relation between electricity meter box nodes and user nodes is obtained through calculation based on the power time sequences of electricity meter boxes and users in all branches. According to the method, the power sequences serve as main data, the data matrix is constructed, the linear relation of the power sequences of the adjacent hierarchical nodes is solved integrally, node-by-node similarity comparison is not needed, and the recognition convenience and the calculation efficiency are improved.

The invention relates to the technical field of industrial equipment monitoring, in particular to an industrial equipment remote monitoring and early warning method based on data fusion, which comprises a multi-dimensional parameter dynamic correlation analysis module, an edge side distributed data processing unit, a self-adaptive threshold generator and an abnormal state early warning mechanism. Through capturing a multi-parameter nonlinear relationship, distributed data processing, dynamic threshold adjustment and a hierarchical early warning strategy, the monitoring real-time performance and accuracy are significantly improved, and the false alarm rate and the missing report rate are reduced. The method can effectively solve the problems of long monitoring period, high abnormity identification delay and the like, is suitable for large-scale industrial scenes, and has a wide application prospect.

The invention discloses a dam multi-point deformation prediction method and system, and relates to the technical field of dam safety monitoring, and the method comprises the steps: constructing a finite element analysis model, calculating a hydraulic component displacement feature set, introducing measuring point space coordinates to construct a dam deformation feature factor set, carrying out the feature screening through employing an improved BorutaShap algorithm, and building a simple and efficient deformation prediction BorutaShap model. And capturing a complex nonlinear relationship contained in a residual error of a model predicted value and a real value by using an iTransform deep learning model, predicting a residual error value, superposing the residual error value with a BorutaShap model predicted value, outputting a final deformation predicted value, and establishing a BorutaShap-iTransform model. According to the dam multi-point deformation prediction method, through feature screening and residual error correction, the model complexity is remarkably reduced, effective information in residual errors is effectively mined, and the dam multi-point deformation prediction precision is greatly improved.

The invention discloses a tea garden soil moisture physical simulation and data driving fusion prediction method and device, and relates to the field of soil moisture prediction.The method comprises the steps that based on soil basic attributes and multi-depth soil moisture time sequence observation data, a soil moisture physical baseline sequence is obtained in combination with boundary conditions and initial conditions; respectively extracting data of a target soil layer in a preset time window from the soil water content physical baseline sequence, the multi-source driving data and the multi-depth soil water content time sequence observation data so as to splice the data into an input feature set; obtaining a water content prediction value of the target soil layer on the prediction day based on the input feature set and a preset soil water content prediction model; according to the method, the HYDRUS-1D simulation result is taken as a baseline, the time sequence dependence of soil moisture, especially the lag response of different soil layers, is effectively learned, and the modeling capability of a complex nonlinear relationship is improved.

The invention discloses a method and device for evaluating the susceptibility of rainfall-induced landslide based on side slope geometrical characteristics, and belongs to the technical field of landslide geological disaster prevention and control. The method comprises the following steps: based on a side slope to be evaluated, establishing a standard three-dimensional side slope model, establishing simulated three-dimensional slope models with different gradients and different fluctuation degrees based on the standard three-dimensional slope model; the slope roughness coefficient of each simulated three-dimensional slope model and the safety coefficient of each simulated three-dimensional slope model in different rainfall stages are calculated, and the slope roughness coefficients are obtained based on slope geometric feature calculation and the standard three-dimensional slope model; and fitting the linear relationship between the slope roughness coefficient and the safety coefficient of each simulated three-dimensional slope model in the same rainfall stage to obtain the corresponding relationship between the slope roughness coefficient and the safety coefficient in different rainfall stages. According to the method, the rainfall-induced landslide susceptibility of the slope can be accurately and efficiently evaluated.

The invention discloses a linear variation stiffness matrix method for suspension bridge space cable forming form finding, relates to the technical field of bridge numerical simulation, and solves the defects of a stiffness matrix in the process of solving the suspension bridge forming form finding through a segmented catenary method. The method comprises the steps that a suspension bridge is simplified into a space cable system composed of a main span, a side span and an anchor span, and the main span, the side span and the anchor span are sequentially subjected to a bridge forming and shape finding process; the linear change stiffness matrix solved in the bridge forming and shape finding process of the main span is used for representing the linear relation between the displacement change of the control point and the initial force change of the main cable; according to the method, an analytical expression of a linear change stiffness matrix is deduced, and an efficient iterative algorithm is established. According to the algorithm, the initial force correction is directly calculated according to the displacement error of the control point in each iteration.

A system and method for neural network confidence regularization is disclosed. A classification system uses a neural network training model to generate prediction. The confidence of the neural network training model is adjusted based on feature prevalence. The system processes mixed data types (binary, categorical, continuous, and date) through type-specific transformations and tensor construction. A tempering factor is calculated from the unweighted sum of features and applied to intermediate neural network outputs. This tempering mechanism reduces model confidence when several low-weight features are present, enabling faster convergence, better generalization, and improved classification accuracy compared to standard neural networks, particularly for complex non-linear relationships in tabular data domains.

The invention provides a rocket aircraft flow field correction method based on multi-fidelity data fusion, and the method comprises the following steps: 1, collecting low-fidelity data, and carrying out the construction of a low-fidelity data set: simulating the change process of an unsteady flow field around an object in a finite time step through a Reynolds time-average simulation method or a large vortex simulation method; 2, data correlation analysis is carried out, wherein a linear relation model yH = rho (x) yL + delta (x) of a non-viscous flow field yL and a viscous flow field yH of the rocket aircraft is established; step 3, neural network architecture design: constructing a low-fidelity data approximation network NNL; 4, performing hyper-parameter learning and optimization; defining a loss function; 5, performing data acquisition and preprocessing: performing high-fidelity viscous flow field data acquisition, and preprocessing the acquired high-fidelity and low-fidelity flow field data; and step 6, model training and verification.

The invention belongs to the technical field of time modulation arrays, and relates to a direction and polarization joint estimation method based on a time modulation array, which comprises the following steps: establishing an alternating polarization time modulation array, obtaining a receiving signal of each array element, a time modulation function and a polarization vector of each array element, and further obtaining an actual receiving signal of each array element, establishing a linear relation among a harmonic component, a polarization parameter and an incident direction of a received signal of each sub-array; according to the linear relation, the relation between direction disturbance and direction function disturbance is obtained, a vector formed by subarray harmonic components is rewritten into a form containing a noise item, error disturbance between a truth value of a direction function and a corresponding estimator is obtained, and then a direction estimation value is obtained; according to the linear relation, the relation between the polarization matching coefficient of the array manifold vector estimation value and antenna polarization and radiation source polarization is obtained, and then the polarization estimation value is obtained. According to the invention, joint estimation of direction and polarization can be realized.

An embodiment of the present invention comprises: a first step of filtering data collected by a data collection unit and tracking an effective rolling radius (ERR) in an operation unit that performs an operation; a second step of correcting the effective rolling radius on the basis of the contact ratio in the calculation unit; a third step in which the calculation unit confirms whether the effective rolling radius increases; a fourth step in which the calculation unit generates a wear rate during the period in which the effective rolling radius increases by using a graph; a fifth step in which the calculation unit outputs wear when the effective rolling radius increases; a sixth step in which the calculation unit corrects the effective rolling radius on the basis of the contact ratio when the effective rolling radius decreases; a seventh step in which the calculation unit estimates the wear of the tire using the linear relationship between the effective rolling radius and the amount of wear of the tire in the portion where the effective rolling radius is reduced; and an eighth step in which the calculation unit outputs wear when the effective rolling radius is reduced.

The present application belongs to the technical field of ship rescue, and provides a fishing boat rescue method based on a Beidou intelligent navigation positioning system, which comprises the following steps: comparing and analyzing the Beidou terminal calculated position and the real position of the rescued fishing boat in historical rescue, dividing the historical rescue into positioning error rescue and normal rescue, then analyzing the spatial consistency and periodic consistency of the positioning error vector, identifying target rescue caused by the multi-path effect, fitting the wave height-positioning error linear relationship according to the satellite elevation angle grouping based on the target rescue data, constructing a multi-path error compensation model combined with the wave direction-azimuth angle difference, and constructing a multi-path effect risk prediction model using the sea wave parameters and the fishing boat position parameters of normal / target rescue, if there is a risk, correcting the Beidou positioning result using the compensation model to solve the positioning error problem caused by the sea wave multi-path effect, improving the positioning accuracy to within the nominal accuracy range of Beidou, and being suitable for offshore fishing boat emergency rescue.

The invention discloses an ultra-sparse CBCT (cone beam computed tomography) reconstruction method, system and equipment based on an ordinary differential equation, belongs to CBCT reconstruction in the field of artificial intelligence, and aims to solve the technical problem of low quality of CBCT reconstructed images. The method comprises the following steps: acquiring sample data, preprocessing the data, constructing and training a CBCT-CT nonlinear relation reconstruction model, and performing real-time reconstruction; during preprocessing, converting the three-dimensional image volume data into simulated X-ray projection data, and reconstructing the simulated X-ray projection data by adopting an FDK reconstruction algorithm to obtain an FDK-CBCT image; the CBCT-CT nonlinear relation reconstruction model comprises an encoder, a NODE module and a decoder; in the training process, the CBCT-CT nonlinear relation reconstruction model is trained through the obtained CT sample image and the FDK-CBCT image. In the reconstruction model, through continuous evolution of NODE modeling image features, the model can model a continuous evolution mapping process from a sparse low-quality image to a high-quality CT image during training, so that stripe artifacts and structural distortion do not easily exist in the reconstructed image, and the reconstruction quality is high.