261 results about "Latent variable" patented technology

The invention relates to a temperature field prediction and parameter inversion method based on a physical constraint generative adversarial network. The method comprises the following steps: acquiring sensor data and sampling random latent variables; constructing a generative adversarial model comprising a generator, a discriminator, an auxiliary parameter generator and an auxiliary variational encoder based on sensor data and a physical constraint partial differential equation; by minimizing reverse KL divergence and introducing physical consistency constraints, boundary condition constraints and information entropy regularization items, parameters of a generative adversarial model are jointly optimized; after offline training is completed, new space-time coordinates and random latent variables are input, a temperature field prediction result is obtained through a trained generator, and PDE parameter estimation of a corresponding position is obtained through an auxiliary parameter generator. Compared with the prior art, the method can perform unified modeling and prediction on the space-time dynamic system dominated by the random partial differential equation, and has generalization ability under uncertainty quantization, physical parameter estimation and sparse observation data.

The invention discloses an electricity meter adaptive error compensation method, system and device based on multi-physical coupling and hidden variable modeling and a storage medium, and relates to the technical field of electric energy metering and detection, and the method comprises the steps: constructing a multi-physical coupling basis vector containing a single variable and a cross variable based on a sensing output signal; hidden variable joint modeling is carried out based on the multi-physical coupling basis vector and parasitic capacitance change, magnetic conductivity drift and shunt resistance offset, and a compensation mapping relation of electromagnetic, thermal and stress coupling is formed; generating a compensation coefficient by adopting the compensation mapping relation, and applying the compensation coefficient in an ammeter metering link to complete an error compensation process; according to the method, voltage, current and phase compensation coefficients are generated through multi-physical coupling basis vector and hidden variable modeling, dynamic description and self-adaptive correction of error sources are achieved, dynamic accurate correction of voltage, current and phase is achieved, and it is ensured that the ammeter keeps high-precision metering for a long time under the complex environment and the device drift condition.

The invention relates to the technical field of sea wave height prediction, and provides a sea wave significant wave height prediction method and system fused with multi-source data, and the method comprises the following steps: correcting obtained satellite observation data based on obtained buoy observation data; carrying out preliminary fusion on the obtained reanalysis data and the corrected satellite observation data by adopting an optimal interpolation method; adding a layer of data mask to mark the position of the satellite data in the fused data; and inputting the marked initial fusion data into a VQ-VAE model, compressing the fusion data into discrete potential variables, inputting the discrete potential variables into a GPT model, and generating a prediction result of the significant wave height of the sea wave. According to the method, artificial intelligence, data assimilation and fine tuning technologies are combined, the data assimilation technology is introduced to fuse the multi-source time-space sparse marine observation data and the model simulation result, and the fine tuning technology is used in the training process to amplify the adjustment effect of the observation data, so that the precision and timeliness of sea wave prediction are improved.

The invention relates to the field of artificial intelligence and computer science, in particular to a causal structure and latent variable-based adaptive reinforcement learning reasoning migration method, which comprises the following steps of: constructing a causal world model fused with multi-modal observation and a decoupling latent variable space; establishing a hierarchical inference engine comprising an intuition layer, a conventional layer and a planning layer; pre-training a quick response and judicial planning dual-mode strategy and generating an interpretable fuzzy rule base; performing calculation level coarse tuning based on task identification and causal complexity; evaluating the real-time state criticality through an adaptive neural fuzzy system and dynamically switching a decision mode; after the action is executed, the threshold and the rule are subjected to closed-loop optimization, and cross-environment efficient migration is realized by utilizing a causal modularization characteristic. According to the technical scheme, consumption of computing resources is remarkably reduced on the premise that decision precision and safety are guaranteed, and the response speed and cross-scene adaptive capacity of a system on edge equipment are improved.

The invention discloses a multi-modal large model training data acquisition method and system, and relates to the technical field of data acquisition optimization, and the method comprises the steps: building a causal graph adjacency matrix based on a cleaning alignment data set, carrying out the anti-fact intervention after recognizing a prejudice variable, and generating an anti-fact sample set; combining the anti-fact sample set and the cleaning alignment data set into an enhanced data set; performing cross-modal anti-long-tail compensation on the enhanced data set to obtain a balanced data set; based on the image data and the text data in the balance data set, depth separable convolution feature extraction and BERT semantic coding are carried out respectively, cross-modal fusion is carried out, and fusion features are output; through the steps of depth separable convolution, BERT coding, quantum latent variable evolution and cross-modal semantic verification, the generalization ability, robustness and social adaptability of the multi-modal large model in a complex and real scene are significantly improved.

The invention discloses a converter batching data generation and optimization method, which comprises the following steps: acquiring an original industrial data set in a converter batching process, preprocessing the original industrial data set, inputting preprocessed training data into an encoder, generating a mean vector and a variance vector of potential variable distribution, and optimizing the mean vector and the variance vector; constraining the similarity between the potential variable distribution and the standard Gaussian distribution through KL divergence loss; inputting the potential variable and the random noise vector into a generator to generate a simulation sample, and distinguishing a real sample from the generated sample through a discriminator; a forced discriminator module is introduced, and a discrimination result is converted into an additional loss item; and carrying out data generation on the missing batching parameters, outputting a complete batching data set conforming to process constraints, constructing a cost objective function, a quality objective function and a resource consumption objective function, constructing constraint conditions of chemical element content of a target finished product, optimizing the converter batching data, and generating optimized converter batching data.

The invention relates to the technical field of artificial intelligence, in particular to a diffusion model preference optimization method and system based on adaptive gradient adjustment, and the method comprises the steps: carrying out the de-noising processing of an initial data sequence, generating a generation sample similar to real distribution, and carrying out the calculation of the preference of a diffusion model according to the sample generated by the initial data sequence in combination with human feedback information; a diffusion model is directly optimized through an adaptive gradient adjustment mechanism, a human preference data sample is constructed, the preference of a user to generated content is reflected, the gradient update weight of the denoising process is dynamically adjusted through a gradient weight function, the function is adaptively adjusted according to the time step and the hidden variable frequency characteristic, and by adjusting the gradient update of the denoising process, the user experience is improved. According to the method, the initial data sequence parameters are optimized, the optimized data sequence is obtained, the optimal hidden variable is calculated according to the optimized data sequence, the optimal hidden variable is adopted to generate the data sample conforming to the human preference, an explicit reward function does not need to be constructed, optimization is achieved only through indirect modeling of the human preference sample, and the model optimization process is simplified.

The invention provides a privacy protection multi-energy load day-ahead probability prediction method based on a diffusion model, and belongs to the field of multi-energy load prediction of an integrated energy system. Comprising the following steps: S1, acquiring respective energy consumption load data by a local main body, and preprocessing and normalizing the data; s2, dividing the energy consumption data into historical loads and labels, and respectively training two groups of self-encoders-self-decoders; s3, the local main body uploads the latent variables obtained after coding to the cloud; s4, grouping and integrating latent variables by the diffusion model of the cloud, and constructing a joint probability prediction model; and S5, during model reasoning, after a latent variable generated by the cloud is downloaded locally, a final predicted value is obtained through decoding by a self-decoder. According to the method, a prediction framework based on longitudinal federated learning is designed for a multi-energy load probability prediction scene related to a cross-energy form in an integrated energy system, and a diffusion model capable of predicting multi-energy load joint probability distribution is constructed by considering a coupling relationship among multi-energy loads.

The invention belongs to the field of intelligent transportation, and relates to a future feature enhanced vehicle trajectory prediction generative adversarial method, which comprises the following steps: a conditional information learning step: completing joint modeling of historical and future features; a generative adversarial training step: performing multi-modal trajectory prediction by adopting a CVAE-GAN hybrid architecture; wherein the CVAE-GAN hybrid architecture comprises an encoder, a generator and a discriminator; the encoder comprises a regression encoder and a prior encoder, and the regression encoder is responsible for mapping a real track to a submerged space, learning compact representation of the track and only being used during model training; the priori encoder performs network sampling to obtain a latent variable; the generator reconstructs a multi-modal future trajectory according to latent variables and condition information; the discriminator receives the trajectory and the condition information, and evaluates the authenticity of the trajectory and the consistency of the trajectory and the condition information. By designing a double-stage strategy, stable generation and optimization of multi-modal trajectory distribution are realized, and the problems that traditional GAN training is unstable and CVAE output is too smooth are solved.

The invention discloses a distillate oil property prediction method based on deep learning feature extraction and partial least squares regression. The method comprises the following steps: firstly, carrying out classification training on a near infrared spectrum through a convolution-attention double-branch fusion network, and extracting high-dimensional spectral features with local and global information; then, historical samples are retrieved from a database based on prediction categories, a plurality of most similar samples are selected by adopting cosine similarity measurement to construct a correction set, and the spectral features and property labels are subjected to standardization processing; and finally, carrying out partial least squares regression modeling on the correction set, extracting latent variables to maximize covariance between spectral features and physicochemical properties, and inputting feature vectors of an oil sample to be detected into the trained PLS model to obtain a corresponding property prediction result. According to the method, the modeling requirement and the category specificity characteristics under the small sample condition are considered while the prediction precision is guaranteed, and the method is suitable for rapid property detection and intelligent analysis in the refining process.

The invention provides a virtual sample generation method and system based on a non-stationary neural network Gaussian process, and the method comprises the steps: obtaining original data, judging the non-stationarity, and judging whether a statistical characteristic changes with an input position or not; if not, a hidden variable model fusing the neural network and the Gaussian process is constructed, and a hidden variable space representing non-stationary distribution is obtained through training; sampling based on hidden variable space probability distribution; using the model to generate = (, Z) (Z, Z)-1X through non-stationary high-dimensional mapping, which is a non-stationary kernel function, Z is a low-dimensional hidden variable, and X is an input variable; checking the consistency of the high-dimensional virtual samples and the original data, and screening a virtual sample set meeting statistical consistency; according to the method, high-quality and high-diversity virtual samples can be generated in small sample and non-stationary scenes.

The invention belongs to the technical field of physical ability analysis and diagnosis, and particularly relates to a latent variable fusion modeling method and system for old people weakness trajectory prediction based on biomechanics. According to the method, the plantar pressure, the joint angle and the electromyographic signal in the gait process of the elderly are collected through wearable sensing equipment, multi-dimensional structured feature parameters are extracted, and a key index group reflecting the weakness degree is generated based on preset weight fusion. And in combination with time sequence fitting and variation trend analysis, dynamic evaluation and risk level judgment of the weakness state of the old people are realized. The system comprises a data acquisition module, a feature extraction module, an index fusion module, a trend analysis module and a result output module, has the characteristics of high accuracy, strong real-time performance, good interpretability and the like, and is suitable for weakness risk screening and rehabilitation intervention guidance.

Disclosed are apparatuses, systems, and techniques that may use machine learning for implementing generative text-to-speech models. The techniques include identifying a mapping of speech characteristics (SC) on a target distribution of a latent variable using a non-linear transformation for at least a subset of the SC. Parameters of the non-linear transformation are determined using a neural network that approximates a statistics of the SC with a statistics predicted for the SC based on the identified mapping and the target distribution of the latent variable.

The invention discloses a quality evaluation method and system for a composite board, and relates to the technical field of quality detection, and the method comprises the steps: constructing a latent variable feature space representing interface binding energy abnormity based on an initial response data field through a physical constraint relation between interface local response features and binding energy distribution; extracting a topological correlation structure between local latent variable features in the latent variable feature space through a graph convolutional network to obtain a local binding energy anomaly initial candidate region; active response mode enhanced excitation is carried out on the local area of the initial candidate area, the dynamic enhanced response of each local area is obtained, and a binding energy abnormal area is determined; determining the specific position and range of the hidden pseudo solder strip according to the correlation topological structure difference between the local feature in the binding energy abnormal region and the normal region; according to the method, precise identification and high-precision positioning of the titanium-steel composite plate interface hidden cold solder strip are realized.

The invention discloses a causal variation inference-based confounding factor-removing multi-behavior recommendation method, which comprises the following steps of: encoding potential uncertainty in multi-behavior interaction by adopting a variation graph automatic encoder so as to capture heterogeneity among different behaviors; in order to efficiently deduce potential confounding factors, a confounding factor reasoning mechanism is designed, and the potential confounding factors are generated through variational reasoning. In a forward diffusion stage of the conditional diffusion module, the model gradually injects noise into potential variables of users and articles to simulate dynamic evolution of user preferences along with time and different situations. In the back diffusion stage, the model uses the deduced confounding factors and combines causal reasoning to guide the denoising process, the influence of potential confounding factors is relieved, and the real causal effect of multi-behavior interaction is captured. Experiments are carried out on two public data sets by aiming at ten recommendation algorithms of four different research emphasis, and experimental results show that the performance of the method is superior to that of existing algorithms of the same type.

The present invention discloses an AI-based intelligent building operation analysis method and system. The method includes: collecting environmental system data, energy consumption system data, and personnel flow system data of a building through sensors, and performing time alignment on the three types of system data through time axis mapping to obtain an alignment matrix; performing adaptive wavelet packet decomposition on the alignment matrix to obtain extended features, and constructing a feature tensor based on the extended features; using the features in the feature tensor as observation variables to construct a latent variable model, where the latent variable model calculates the influence weight of the latent variable on the observation variable through a measurement equation, and performs dynamic coupling degree calculation based on the influence weight; using the features in the feature tensor as nodes, and using the product of transfer entropy and dynamic coupling degree as edge weights to construct a causal network, and performing root cause analysis on the causal network through the PrefixSpan algorithm. The present invention solves the problem that existing building operation analysis technologies cannot comprehensively consider the dynamic interaction between multi-system data.

The invention discloses an aircraft aerodynamic configuration optimization design method based on manifold learning dimensionality reduction mapping, and relates to the technical field of aerodynamic configuration design, and the method comprises the steps: defining design variable parameters of an aircraft aerodynamic configuration, and determining a design variable parameter sample and pressure distribution sample data according to a variable value range; training a preset Bayesian manifold learning dimensionality reduction mapping model, and determining a latent variable by using the trained target Bayesian manifold learning dimensionality reduction mapping model and the pressure distribution sample data; training a preset multi-output deep neural network model based on the design variable parameter sample and the latent variable; and performing pressure distribution prediction based on the trained target multi-output deep neural network model, the target Bayesian manifold learning dimensionality reduction mapping model and the variable value range, optimizing target pressure distribution in the variable value range by using a pressure distribution prediction result, and determining a target design variable parameter value. The problem of high-dimensional mapping precision in existing related schemes can be solved, and the design accuracy is improved.

A method may include: accepting, by a data transformation module, an original dataset as input to a first and a second neural network and outputting a transformed dataset; accepting, by a sensitive attribute suppression module, the transformed dataset as input to a third neural network and calculating a sensitive attribute suppression loss; accepting, by an annotated useful attribute preservation module, the transformed dataset as input to a fourth neural network and calculating a useful attribute preservation loss; accepting by a generic feature suppression module, parameters of a distribution of a latent variable from the first neural network and calculating, for an unannotated generic attribute, a generic feature suppression loss; combining the sensitive attribute suppression loss, the useful attribute preservation loss, and the generic feature suppression loss into a total loss; and training the first neural network and the second neural network with the total loss.

The invention discloses a material performance prediction method and system based on deep learning, and relates to the technical field of material performance prediction, and the method comprises the steps: building a cross-level message passing network based on a double-layer graph data structure, extracting atomic-level and functional group-level feature embedding, and generating a molecular global latent variable through bidirectional cross-scale attention interaction; fusing molecular global latent variables with atomic-scale and functional group-scale features, applying physical constraints in combination with a chemical bonding rule base and a group contribution theory, and constructing a physical property prediction model; obtaining candidate molecules through autoregression generation and molecular force field verification based on a molecular global latent variable and a physical property prediction model; and screening candidate molecules by using the physical property prediction model, and outputting a final molecule set through molecular dynamics simulation and synthesis feasibility evaluation verification. Through multi-stage verification of molecular dynamics simulation and synthesis feasibility evaluation, a high-tension ring or a non-synthesized structure is effectively eliminated, so that the generated molecule has high performance and manufacturability.

The invention discloses a latent variable diffusion three-dimensional physical model reconstruction method and device, and an identification method, and the reconstruction method comprises the steps: obtaining a simulation data set of a three-dimensional fluid, carrying out the dimension reduction processing of simulation data at each time point, determining the noise adding amount corresponding to each time point, and carrying out the recognition of the three-dimensional physical model. And determining a more accurate target potential variable of the initial potential variable at the target time point and under the target condition variable, and finally reconstructing a more accurate target state of the three-dimensional fluid of the simulation information of the first number dimension at the target time point and under the target condition variable. Therefore, the rebuilding accuracy of the latent variable diffusion three-dimensional physical model is improved.

A method for generating a principal predictor model from multi-dimensional time series data, comprises: receiving, from a plurality of sensors, multi-dimensional time series data corresponding to a plurality of original variables; transforming the multi-dimensional time series data to a lower dimension to define reduced-dimensional time series data; extracting, by a controller, a plurality of latent variables from the reduced-dimensional time series data and determining values of the plurality of latent variables in a first time period; initializing, by the controller, a loadings matrix corresponding to a set of latent variables of the plurality of latent variables; and determining, by the controller, one or more principal predictor model parameters, by performing an iterative process. The iterative process comprises (a) predicting values of the plurality of latent variables based on the reduced-dimensional time series data and the loadings matrix, by using an estimation process which maximizes a covariance between the values of the plurality of latent variables and the predicted values of the plurality of latent variables; (b) calculating a new loadings matrix from the loadings matrix and the predicted values of the latent variables; (c) updating the loadings matrix based on the calculated new loadings matrix; and (d) iteratively repeating (a) to (c) until the one or more principal predictor model parameters reach convergence.

The invention relates to the technical field of speech synthesis, and particularly discloses a speech synthesis method and system for controllable latent variable modeling based on semantic distillation, and the method comprises the steps: converting a Mel spectrum into continuous latent variable distribution through a speech coding module, generating continuous latent variables through re-parameterization sampling, introducing a self-supervised model for semantic distillation, and carrying out the semantic distillation. According to the method, alignment of latent variables and semantic features is constrained through marginal cosine similarity and distance matrix structure loss, a text encoder maps a phoneme sequence into latent variable distribution, time sequence alignment of a text and the latent variables is achieved in combination with monotonic alignment search, and a decoder reconstructs the latent variables into a Mel spectrum. According to the method, waveform synthesis through a vocoder and total loss function joint optimization reconstruction, KL divergence, distillation, text alignment and confrontation loss are carried out, discrete information loss is avoided through continuous latent variable modeling, semantic consistency and text alignment efficiency are enhanced, the naturalness, coherence and real-time performance of synthesized voice are improved, and the method is suitable for scenes such as voice assistants and virtual anchors.

The invention relates to a doubly-fed induction generator (DFIG) fault detection and early warning method based on multi-scale latent variable regression (MSLVR), is suitable for multiple fault types and complex coupling effect scenes, and belongs to the technical field of fault detection. According to the method, firstly, key coupling variables influencing the DFIG are analyzed and selected from the angle of information theory, and the time scale is expanded through multivariate variational mode decomposition (MVMD), so that the data quality is optimized. Then, aiming at the defect that latent variable regression (LVR) is difficult to pay attention to the local influence of a sub-mode in multi-scale data processing, designing a serial architecture to gradually extract and eliminate interference of different components; and the residual subspace is processed by further combining principal component analysis (PCA), so that the online detection method capable of identifying early DFIG faults is developed. Through training and optimization, the model is finally stored for calling, the influence of the coupling effect on the DFIG can be effectively eliminated, efficient and accurate fault detection and early warning can be realized in various fault type scenes, and the detection accuracy is obviously superior to that of the existing mainstream method.

This invention discloses a stepwise data assimilation method for a set subspace in a nonlinear inverse problem, relating to the field of data processing technology. The invention constructs an initial prior physical set based on the physical state variables to be inverted and optimized, and historical observation data. It extracts the static subspace basis anomaly matrix, initializes the latent variable set and particle weights, calculates fractional-step incremental reweighting, evaluates the current likelihood mismatch penalty using predicted data, updates and normalizes the particle weights in the logarithmic domain, obtains the effective sample number, performs system resampling operations in conjunction with a preset resampling tolerance coefficient, eliminates low-weight particles and replicates high-weight particles, synchronously updates the latent variable set and predicted data, executes a dynamic MCMC mutation loop to generate proposed latent state vectors and affinely maps them to a high-dimensional physical space, and updates the latent variable set by evaluating the annealing target energy within the latent variable subspace until all fractional steps are traversed. Finally, it outputs the latent variable set and maps it back to the physical space.

The invention belongs to the technical field of intelligent shield construction, and particularly discloses a shield tunneling machine posture multi-step prediction and deviation correction control method based on latent variable enhanced Free Transform. Firstly, shield construction process data are obtained, and a control quantity set is established and preprocessed; error features, change rate features and unit mileage features are constructed based on the preprocessed data, and a feature sequence is formed after normalization; organizing the sequence into a token sequence with a single time step, and inputting the token sequence into a pre-trained Free Transform model after position and mileage coding; the model aggregates complete sequence information through a non-causal encoder dedicated block, calculates latent variable distribution parameters based on a hidden state sequence, samples to obtain latent variables, maps the latent variables into conditional vectors and broadcasts the conditional vectors to a decoding path; and finally, outputting a future shield attitude vector, an over-threshold probability and a recommended control quantity in parallel by a multi-task output head. According to the method, working condition self-adaptive multi-step prediction and deviation correction are realized, and the safety and intelligent level of shield construction can be improved.

The invention discloses a roadway driving face roof fall risk non-contact multi-modal fusion real-time sensing and early warning method, which comprises the following steps: acquiring multi-modal heterogeneous data in a non-contact monitoring mode, the multi-modal heterogeneous data being data of different formats from different data sources; performing feature extraction on the multi-modal heterogeneous data to obtain multi-modal features, performing key feature screening and dimensionality reduction optimization on the multi-modal features, and calculating roof fall risk latent variables according to the key features after dimensionality reduction optimization; and extracting time sequence change characteristics of the roof fall risk latent variable, and making a decision according to the time sequence change characteristics to obtain a roof risk stability grade and a support strategy.

The invention discloses a label correction crowdsourcing result aggregation method and system based on deep clustering. Firstly, task features are extracted through a pre-training model to serve as input data of the model; secondly, introducing a worker loss item, defining a comprehensive loss function of the model in combination with reconstruction loss and KL divergence in the variational depth embedding model, and training the model through a gradient descent method; secondly, in the training process, clustering the hidden variables by using a Gaussian mixture model to obtain clustering clusters, and mapping a specific class label for each clustering cluster in combination with task labeling information of a worker; and finally, the contour coefficient of each task is calculated, the task with low clustering accuracy is identified, and the precision of the clustering result is further improved through readjustment and correction of worker labels. According to the method, tasks with similar features are clustered, and tasks with low clustering quality are corrected by using worker tags, so that the interference of tag noise on truth value inference is effectively reduced, and the performance of a result convergence model is improved.

The invention provides a time series data anomaly detection method, device and equipment, and belongs to the field of data processing. The method comprises the following steps: mapping target time sequence data to a potential space through a preset encoder to obtain an initial potential variable; performing fuzzy reasoning on the initial potential variable to obtain a target potential variable and a membership degree of the initial potential variable in a target category, wherein the target category indicates a normal data category or an abnormal data category; obtaining generation data corresponding to the target potential variable according to the target potential variable and a preset generator; obtaining a judgment score according to the generated data and a preset discriminator, wherein the preset discriminator is used for judging whether the generated data is real data or not; and obtaining a target score according to the target time sequence data, the generation data, the membership degree and the judgment score, wherein the target score is used for indicating whether the target time sequence data is abnormal or not. The accuracy of abnormal data judgment and the generalization ability of the model are improved.

The invention discloses a bridge bearing capacity dynamic evaluation method based on multi-source monitoring data driving, and aims to solve the problem that the bearing evaluation precision is difficult to guarantee under a sparse sensing coverage condition. According to the method, through cross-modal attention alignment guided by event anchor points, mask codes influencing line and propagation time delay constraints and geometric position coding, Green function kernels and time-varying boundary flexibility latent variables are introduced into a physical constraint neural operator model, symplectic integral propulsion is adopted, and block shape-preserving interval estimation and order-preserving calibration weighted according to working conditions are adopted. According to the method, the full-bridge space-time response and the bearing utilization coefficient are calculated, the risk level and the load limiting suggestion are output, and the technical effects of reliability improvement, physical consistency and long-time energy stability evaluation under the complex working conditions and the low signal-to-noise ratio are achieved.

The invention discloses a hydroelectric generating set fault diagnosis data enhancement method based on a diffusion model and generative adversarial training, and belongs to the field of industrial equipment fault diagnosis. According to the method, the data enhancement model combining the diffusion model and the generative adversarial network is constructed, a multi-modal non-Gaussian distribution mechanism and a latent variable control generation process are introduced, the limitation of a single Gaussian hypothesis of a traditional diffusion model is broken through, and richer and more real sample generation is realized. And a condition discriminator and a time sequence modeling mechanism are introduced, so that the training of the model under different noise levels is more stable, and the authenticity judgment capability of the discriminator on the generated sample is enhanced. The problems that an existing hydroelectric generating set fault diagnosis system is insufficient in sample, low in generated sample quality and unreal in sample distribution are solved, and the diversity and quality of data are remarkably improved.