226 results about "Control limits" patented technology

The invention provides an MES-based power distribution cabinet quality inspection analysis and process regulation and control method and system, and the method comprises the steps: converting a quality inspection standard into an executable rule, generating a standardized quality inspection template, evaluating the quality of a single project and a batch, monitoring the mean value and fluctuation of a production process based on a statistical process control theory, and calculating a process capability index. A dynamic threshold value formula is combined to adjust a control limit in real time, process stability is evaluated, equipment data is standardized, production batches are associated, data consistency is verified and ensured, graded early warning is performed according to the deviation degree, a machine learning model is combined to realize predictive maintenance, process parameters are dynamically corrected, historical data are analyzed through machine learning, and the process quality is improved. According to the method, high-order feature quantification influence factors are extracted, abnormal parameter combination is detected in combination with an isolated forest, a process strategy library is updated, closed-loop optimization is formed, full-process automation from data acquisition, real-time analysis and parameter adaptive adjustment to predictive maintenance is realized, and quality inspection efficiency and process stability are improved.

The invention discloses a non-stationary industrial process early fault diagnosis method based on a common trend model, and belongs to the field of industrial process monitoring and fault diagnosis. The method comprises the following steps: collecting training data of a non-stationary industrial process under a normal working condition, performing denoising by using singular spectrum analysis, and obtaining a stationary projection matrix and a stationary data matrix by using a common trend model; orthogonal conversion is carried out on the stationary data matrix, a conversion projection matrix is calculated, and an orthogonal conversion data matrix and a statistic data matrix thereof are obtained; calculating a mean value and a covariance matrix of the statistic data matrix to obtain a control limit; the method comprises the following steps: acquiring test data under a real-time working condition of an industrial process, processing to obtain statistic data, calculating a mahalanobis distance index, and comparing with a control limit to judge whether a fault occurs; and if a fault is detected, realizing fault separation according to fault reconstruction and a mapping relation. According to the method, the interference of the non-stationary characteristic on the fault diagnosis model can be reduced, and fault detection and separation in the non-stationary process are realized.

The invention discloses a process quality control method combining FMEA analysis and a multivariate control chart. The method comprises the following steps: planning and preparing PFMEA; identifying special quality indexes of the process parameter diagram; a control limit and an abnormal mode are determined through failure analysis and risk analysis in the PFMEA, so that abnormal fluctuation in the product machining process is accurately identified; performing multivariate quality index decoupling and correlation analysis; constructing a multivariate SPC control chart; performing abnormal mode identification and source tracing on the multivariate SPC control chart; pFMEA optimization is combined with an SPC control chart to adjust the sampling frequency and the control limit. According to the method, the technical problems that the product quality is difficult to control through frequent detection during large part machining, a single-variable SPC control chart is inaccurate in abnormal mode recognition, the SPC control chart cannot be formulated according to the actual production process, and the production process cannot be effectively improved by the SPC technology are solved.

The invention discloses a shield tunneling machine tunneling key parameter real-time multi-step prediction method based on ST-GCN-LSTM, and the method comprises the following steps: obtaining the operation data of a shield tunneling machine under a normal working condition, and carrying out the preprocessing of the operation data; analyzing the preprocessed operation data, and determining related variables of total thrust, cutterhead torque, cutterhead rotating speed, penetration and propelling speed average value tunneling key variables; respectively calculating the control limit of each tunneling key variable for different soil layers, and training a multi-step prediction model based on an ST-GCN and LSTM combined model; acquiring operation data of the shield tunneling machine in real time, inputting the operation data into the trained model for preprocessing, dynamically predicting tunneling key parameter values of multiple steps in the future, and judging whether predicted values continuously exceed a control limit or not; according to the method, real-time multi-step prediction of the tunneling key parameters can be carried out in combination with soil layer classification, and prediction inaccuracy caused by changes of external factors such as soil layers is reduced.

The invention provides a generator excitation system fault detection method and system, and the method comprises the steps: firstly collecting the historical data of a generator excitation system, and forming a historical data set; then, constructing a stacked sparse auto-encoder network, taking the historical data set as input, and training the stacked sparse auto-encoder network in combination with a physical constraint equation of the generator excitation system; and finally, collecting real-time data of a generator excitation system, inputting the real-time data into the trained stacked sparse auto-encoder network, calculating an SPE index, and carrying out fault diagnosis through an SPE index control limit. Physical information is fused in the model training process, the detection result has high accuracy, the interpretability of the result is enhanced, and compared with a pure data driving method, the method has the advantages that dependence on large-scale fault samples is remarkably reduced, and the detection accuracy is greatly improved in an unknown scene or a scene without faults. High detection precision and robustness can still be kept, and the generalization ability and engineering applicability of the model are improved.

The invention relates to the technical field of industrial process fault detection, and discloses a KL divergence tiny fault detection method based on an improved self-attention mechanism, and the method comprises the steps: offline training: carrying out the principal component analysis modeling based on training data, obtaining a load matrix, and calculating a score matrix of the training data and a score matrix of test data; performing KL divergence test based on a sliding window strategy and the score matrix to obtain KL divergence statistics; an improved self-attention mechanism and a local outlier factor weighting method are introduced, a training weighted score matrix is obtained based on KL divergence statistics, second KL divergence calculation is carried out, attention statistics are obtained, and then final statistics of historical production data are obtained; estimating a control limit by using a kernel density estimation method; and calculating an online statistical magnitude, and comparing the online statistical magnitude with a control limit to judge whether a fault occurs in the industrial process at the current moment. The method is more sensitive to small fault sensing, and has a high fault detection rate and good robustness.

The invention discloses a regulation and control quota distribution method and system fusing subjective and objective multi-dimensional features. The method comprises the following steps: acquiring electrical load data, historical response behavior data and subjective response intention information of a user; firstly, a Ward system is used for clustering, users are clustered according to active power, then a primary clustering center is used as an initial clustering center of secondary clustering, FCM clustering is carried out, and a typical load curve of the users is described based on a secondary clustering method; load prediction is carried out based on an NARX neural network, a predicted load curve is compared with a typical load curve, and adjustable potential is calculated; constructing a DR feature data set; and the DR feature data set is fused with an entropy weight method and an analytic hierarchy process to obtain a combined weight, a fuzzy relation matrix is constructed, a user comprehensive score is quantified, and a comprehensive response potential score of the user is formed.

Aspects of the disclosure relate to computing hardware and software for dynamic control limit configuration. A computing platform may receive and format historical data. The computing platform may input the formatted data into a control limit prediction model, which may output predicted control limits by weighting, using an ensemble model, outputs from a plurality of other models to produce predicted control limits. The computing platform may adjust existing control limits based on predicted control limits to create actual control limits. The computing platform may receive real time data, and may identify, using the actual control limits and the real time data, a deviation score for the real time data. The computing platform may compare the deviation score to the actual control limits. Based on detecting that the deviation score breaches the actual control limits, the computing platform may send an indication of the breach.

The invention provides an SPC control method, system and device based on dynamic threshold optimization and a medium, and can solve the problems caused by a traditional SPC static threshold, and the method comprises the steps: collecting and preprocessing an original data stream containing key quality characteristics or key process parameters from various data sources on a production line in real time, obtaining a structured valid data sequence; intercepting current sliding window data from the structured valid data sequence by adopting a sliding time window mechanism, performing abnormal value processing, and dynamically calculating a control limit of the SPC control chart based on the processed sliding window data to obtain a first control limit; when the process parameter step change of the production line is monitored, updating the control limit of the SPC control chart to obtain a second control limit; and comparing the latest effective data output in the structured effective data sequence with the first control limit or the second control limit in real time, and triggering an alarm when judging that the production process is abnormal according to a comparison result.

The invention discloses a real-time comprehensive working condition superiority evaluation method in a cement calcination process. The method comprises the following steps: step 1, carrying out working condition category grading on a historical operation data set in a cement production process; step 2, screening process variables with high relevancy with the key indexes to train a CVAE network to obtain a pre-trained CVAE network; 3, calculating four groups of reconstruction error control limits by adopting a kernel density estimation (KDE) method; 4, constructing an SFA network, obtaining a chronic feature mapping matrix, and calculating a statistic control limit S2; 5, fusing dual-channel monitoring results to obtain a working condition comprehensive scoring index CSI and a working condition dynamic index DI; step 6, monitoring current operation data, inputting the optimality evaluation model, calculating corresponding statistics and indexes, and outputting a current monitoring state and a real-time score; according to the invention, multi-index comprehensive optimization evaluation and visual monitoring of the quality, the yield, the energy consumption and the working condition stability of the cement calcining operation working condition can be realized.

The invention relates to the technical field of fault detection, in particular to a chemical production online fault detection method and system based on big data, and the method comprises the steps: collecting operation data of production equipment, constructing real-time data matrixes with different sizes, intercepting historical data with the same size as the real-time data matrixes, and constructing a historical data matrix; and calculating the similarity between any real-time data matrix and each historical data matrix with the same size by using a sliding window. According to the method, the real-time data matrixes of different sizes are constructed to analyze the possible long and short term trends of the data, the similarity between the real-time data matrixes and the historical data matrixes is calculated by moving the sliding window, and the estimated value of the comprehensive control limit value of the real-time data is estimated through the control limit values of the similar historical data matrixes. And the fault condition of the equipment can be accurately judged according to the difference between the estimated value and the comprehensive control limit value of the real-time data.

The invention discloses a multi-working-condition process monitoring method and system based on zero-forgetting continuous dictionary learning, and the method comprises the steps: carrying out the offline modeling: firstly, carrying out the dictionary learning through employing an initial working condition data set, obtaining an initial dictionary, carrying out the decomposition, obtaining a low-rank matrix of an initial working condition, and calculating the control limit of the initial working condition through calculating a sample reconstruction error; performing incremental updating on the low-rank matrix obtained by learning the old working condition by using the new working condition data set to obtain a low-rank matrix of a new working condition, and calculating a control limit of the new working condition; on-line monitoring comprises the following steps: firstly, a weight selector is used for distributing weights for a low-rank matrix of a learned working condition according to monitoring data, and a monitoring dictionary is constructed in a self-adaptive manner; reconstructing the monitoring data by using the monitoring dictionary, solving a reconstruction error, and further judging the working condition of the monitoring data; and finally, calculating a fault detection statistic according to the reconstruction error and the control limit of the corresponding working condition, and judging whether the monitoring data is abnormal or not according to the statistic. According to the invention, accurate monitoring of a multi-working-condition process is realized.

The invention discloses a power grid state limit operation regulation and control method and system based on a topological structure. The method comprises the following steps: constructing a power grid topological structure according to nodes and tie lines of a power grid; monitoring equipment real-time state data of the power grid in real time; estimating the real-time state data of the equipment according to a state estimation algorithm to obtain an actual operation state of the power grid; constructing a section parameter model of the standardized operation mode, wherein the section parameter model comprises a power failure equipment parameter table, a control grading parameter table and a control limit parameter table; the power failure equipment parameter table and the control grading parameter table are condition parameters, and the control limit parameter table is limit parameters; and inputting the power grid topological structure, the equipment real-time state data and the actual operation state of the power grid into the section parameter model to obtain control limit parameters of the power grid section. Dynamic calculation of section control parameters is achieved through the standardized section parameter model, it is guaranteed that limit parameters are adjusted and controlled in real time in a complex power grid environment, and reliable power supply of a power grid is guaranteed.

The invention relates to the technical field of aviation safety and data processing, in particular to a flight data anomaly detection method based on DEWMA-IForest, and the method comprises the following steps: carrying out the preprocessing of original flight data; initializing a smoothing coefficient and a sliding window size parameter; calculating an arithmetic mean value and a standard deviation by taking data in the sliding window as a statistical sample, and respectively taking the arithmetic mean value and the standard deviation as a DEWMA statistic initial value and a control limit calculation reference value; setting an initial control limit range; calculating the DEWMA statistical magnitude and the standard deviation of the data to be measured; dynamically updating the control limit range; and DEWMA fitting reconstruction data and a control limit are calculated. According to the method, the dynamic control limit is adapted to the time-varying property of the QAR data, so that the sensitivity of small-amplitude trend anomaly is improved; high-dimensional data dimension disasters are avoided through multi-algorithm fusion; and the abnormal criterion can be explained, so that a scientific basis is provided for aircraft operation monitoring and intelligent maintenance decision making.

The invention provides a non-stationary industrial process monitoring method and system. The method comprises an offline training stage: calculating a time Laplacian matrix and a space Laplacian matrix based on a historical data matrix; constructing an objective function of the stationary subspace analysis method, and adding a time constraint term of a time Laplacian matrix and a space constraint term of a space Laplacian matrix into the objective function; solving the objective function to obtain a stable projection matrix; calculating a stationary component and a monitoring index of each sample in the data matrix X in sequence; determining a control limit by using a kernel density estimation method; an online monitoring stage: based on the real-time operation data x, calculating a stationary component of the real-time operation data x and a corresponding real-time monitoring index according to the stationary projection matrix, and if the real-time monitoring index is greater than a control limit, judging that the operation of the non-stationary process has a fault; the monitoring accuracy can be improved.

The invention discloses a chemical process anomaly type identification method based on a multi-dimensional causal association index, and belongs to the technical field of chemical process monitoring and fault diagnosis. The method comprises the following steps: firstly, carrying out offline modeling: acquiring normal working condition data, constructing a monitoring model by utilizing a trend slow feature analysis algorithm, and determining a statistic control limit; during online application, statistics of real-time data are calculated through a model, and an alarm is given and a type identification process is triggered if the statistics exceed a limit. A multi-dimensional causal association index (MCAI) is constructed by analyzing whether a quality variable deviates or not and further performing comprehensive analysis on a root variable causing an anomaly from four dimensions of a position (CPI) of the root variable in a process, an attribute (CAI) of the root variable in a causal network and an influence propagation capability (CPD) of the root variable. The problem that a traditional anomaly detection method cannot distinguish process faults from new working conditions is solved, false alarm can be effectively avoided, and a decision basis is provided for accurate maintenance.

The invention relates to a current transformer error evaluation method based on current imbalance contribution analysis, and the method comprises the following specific steps: collecting CT group data in a normal error state, and constructing CT group modeling current vector data; constructing a current transformer error state evaluation model based on an overdetermined linear equation set; solving a reference weight coefficient matrix in the current transformer error state evaluation model; constructing an error fluctuation characteristic quantity; calculating a Q statistic of the error fluctuation characteristic quantity, and solving a statistic control limit Qcl of the error fluctuation characteristic quantity Ef based on empirical estimation of confidence; judging whether the operation error abnormal condition of the mutual inductor exists in the line or not; and if it is judged that the current transformer in the line at the to-be-detected moment is abnormal, positioning of the abnormal transformer is realized based on the contribution degree of each phase of current transformer to the real-time statistical magnitude. According to the invention, accurate discrimination of the metering performance of the current transformer in operation is realized, and technical support is provided for troubleshooting, state evaluation and the like of the current transformer.

The invention provides a strip steel hot continuous rolling industrial process monitoring method and device, and relates to the technical field of ferrous metallurgy industrial process monitoring. The method comprises the following steps: an off-line modeling stage: collecting multi-dimensional sensor data in the strip steel hot continuous rolling industrial process for standardized cleaning; carrying out data enhancement on the cleaned data by adopting DDIM; random shielding is carried out by using a feature focusing cooperation mechanism based on random shielding; constructing an unsupervised process monitoring model, and training the constructed unsupervised process monitoring model by using the data after the shielding operation; processing data reconstructed by the trained unsupervised process monitoring model by adopting a kernel density estimation method, and determining a control limit; and an on-line monitoring stage: judging whether the current sample is normal or not by using the control limit to realize abnormal working condition detection. By adopting the method, the monitoring sensitivity of the strip steel hot continuous rolling industrial process can be improved.

The invention discloses an industrial process monitoring method based on a lightweight deep principal component analysis-auto-encoder model, and the method comprises the steps: collecting the related process variable data of an industrial process on line, carrying out the preprocessing of the collected data, inputting the preprocessed data into a trained lightweight deep principal component analysis-auto-encoder model, and carrying out the recognition of the related process variable data. The T2 statistical magnitude and the SPE statistical magnitude are obtained and compared with corresponding control limits, and a monitoring result is output; the lightweight deep principal component analysis-auto-encoder model uses a PCA dimension reduction module for dimension reduction to obtain the data score, inputs the score into the auto-encoder, and outputs the reconstruction data. According to the method, the calculation efficiency of the model is improved, compared with a traditional deep network model, the calculation speed is obviously improved, compared with a single multivariate statistical model, the strong feature extraction capacity of the deep hierarchical cascade model structure is incomparable to that of the single multivariate statistical model, and the problem of complex industrial process monitoring is effectively solved.

The invention discloses a steel bar weight negative deviation measurement control system and method, electronic equipment and a storage medium. The system comprises a bale count re-checking sub-module, a negative deviation calculation module and a negative deviation control module. The bale number re-checking sub-module obtains the bale steel bar number and sends the bale steel bar number to the production and manufacturing system and the weighing system; the negative deviation calculation module calculates a count difference value according to the number of the bundled steel bars and the rated number of the production and manufacturing system, calculates a negative deviation measurement value according to the theoretical weight and the actual weight, and compares the size relation between the negative deviation measurement value and a negative deviation control limit value; and the negative deviation control module sends a control signal to the weighing system according to the magnitude relationship among the count difference value, the negative deviation measurement value and the negative deviation control limit value. According to the technical scheme, the accuracy of data in the production process is ensured, errors and cost of manual counting are reduced, and the production efficiency and the product quality are improved.

The invention relates to the technical field of flow measurement, in particular to a gas flow process monitoring and abnormity diagnosis method based on EWMA. The method comprises the following steps: when a gas flow process is in a stable state, collecting sample data and calculating a mean value mu0 and a standard deviation sigma; setting a smoothing parameter lambda and a control limit parameter L of the EWMA control chart; acquiring traffic data in real time, and constructing an EWMA statistic Zi; calculating dynamic upper and lower control limits according to the mean value, the standard deviation, lambda and L; and judging whether the statistic Zi exceeds a control limit or not, and if so, judging that the statistic Zi is abnormal and giving out an early warning. According to the method, the EWMA control chart is used for carrying out weighted averaging on historical data, the detection sensitivity of tiny and slow trend drift is enhanced, the defects that a traditional Shewhart control chart and a fixed threshold value method are prone to missing report and false report in gas flow monitoring are overcome, early-stage and accurate early warning of gas flow abnormity is achieved, and the gas flow monitoring accuracy is improved. The intelligent level and reliability of the flow metering system are improved, and hardware cost does not need to be increased.

The invention discloses a cloud edge collaborative intelligent autonomous monitoring method for a dynamic industrial process. The method comprises the steps that a cloud end establishes a dictionary model and trains and updates the model; the edge device obtains the latest dictionary, classifier and control limit from the cloud end, and then carries out abnormity monitoring and working condition identification on the current industrial process; wherein the cloud process comprises the following steps: establishing a dictionary model and training to obtain a dictionary, a classifier and a coding matrix; screening the historical samples based on the multi-azimuth hardness values of the historical samples, and forming a balanced data set with the new working condition samples to update the dictionary and the classifier; and taking the reconstructed data set and the reconstructed label as complete working condition knowledge, compressing and simplifying the updated over-complete dictionary and classifier, and taking the compressed and simplified dictionary and classifier as a dictionary and a classifier which are finally deployed to an edge layer. According to the invention, reliable and efficient monitoring of dynamic industrial occasions can be realized.

The invention relates to a system fault diagnosis method. The method comprises the following steps: constructing an anomaly diagnosis model based on principal component analysis; based on the abnormality diagnosis model, calculating a statistic control limit by using historical sample data in a normal operation state of the system; monitoring real-time sample data of system operation by using the abnormality diagnosis model and identifying an abnormal state of the system; obtaining abnormal sample data corresponding to the abnormal state of the system, and updating the statistic control limit based on the abnormal sample data; and according to the updated statistic control limit, re-determining whether the system is in an abnormal state by using the abnormality diagnosis model, and if the system is in the abnormal state, analyzing the abnormal sample data by using a fault identification model to obtain a fault classification result. The method can quickly and accurately judge the health state of the system and perform fault classification.

The invention discloses a real-time calculation method for a power grid operation section control limit, relates to the technical field of power grid operation control, and aims to solve the problem of unstable power grid data transmission efficiency. According to the method, the transmission channel of which the residual capacity is greater than the transmission capacity is selected as the transmission channel of the established model to be transmitted to the cache space, so that data can be transmitted more efficiently, the data transmission efficiency and the data transmission stability are effectively improved, and the transmission speed of the established model can be further improved by selecting the strongest transmission channel signal; the problem of channel blockage during transmission is reduced, and by performing safety index analysis on actual power grid section limit data and standard section control limit model data, on the premise of ensuring safety, the transmission capacity of a power transmission section can be exerted to the maximum extent, the utilization efficiency of a power grid is improved, and the power transmission efficiency is improved. And the data which are not in the safety range is alarmed, so that the operation safety of the power grid is further improved.

The invention provides a quality control system based on a multivariate exponential weighted moving average control chart, and belongs to the technical field of quality management and statistical process control. The system comprises a quality data acquisition module, a data preprocessing module, a statistical process control module, an anomaly judgment module and a quality evaluation and improvement module, and is used for monitoring and controlling the production process of a product with a plurality of quality characteristics in real time. The method comprises the following steps of: acquiring quality data of a plurality of quality characteristics in a production process by a system, preprocessing the quality data, and constructing a multivariate quality characteristic data set; estimating a mean vector and a covariance matrix in a controlled state based on the historical quality data, performing weighted updating on the multivariate quality data by adopting a multivariate exponential weighted moving average method, calculating a corresponding MEWMA statistic and generating an MEWMA control chart; and comparing the MEWMA statistical magnitude with a preset control limit to judge whether the production process is in an out-of-control state or not, and outputting early warning information and a corresponding quality evaluation result and improvement suggestion when abnormality is detected. The method can comprehensively analyze related information among multivariate quality characteristics, improves the sensitivity and accuracy of anomaly detection in the production process, and effectively improves the quality control level of the production process in the manufacturing industry.

The invention discloses a heterogeneous low-quality industrial time series data anomaly detection method, which comprises the steps of collecting heterogeneous time series data, performing continuous process variable augmentation, and further constructing a training set; in a heterogeneous feature integration pre-training stage, a heterogeneous representation learning network is constructed, and pre-training is performed by using a pre-training sample set in a training set; in a fusion detection adaptive fine tuning stage, a heterogeneous fusion detection network is constructed, and a fine tuning sample set in a training set is used for training; and obtaining a reconstruction value of a continuous process variable by using the finally trained heterogeneous fusion detection network, and completing anomaly detection according to an anomaly detection control limit. According to the method, the time sequence dependency relationship between heterogeneous variables can be accurately modeled in a data low-quality scene with variable isomerism and measurement point missing, the accuracy of anomaly detection is effectively improved, and powerful support is provided for intelligent operation and maintenance and optimization decision making of an industrial production system.

The invention discloses an industrial process fault detection method and device based on a time convolution auto-encoder, and belongs to the field of industrial process detection.The method comprises the steps that training data and test data are preprocessed, and training time sequence data and test time sequence data are obtained; inputting the training time sequence data into an industrial process fault detection model for training to obtain a target model, wherein the industrial process fault detection model at least comprises two layers of TCN and CAE; inputting the training time sequence data into a target model to calculate a control limit; inputting the test time sequence data into the target model to calculate an SPE statistical magnitude; and judging whether the to-be-detected working condition data is abnormal or not according to the control limit and the SPE statistical magnitude. According to the method, TCN is introduced on the basis of CAE, a novel time convolution auto-encoder is designed, an industrial process fault detection model is created, a convolution receptive field can be gradually expanded through at least two stacked time convolution networks to change the time scale of time sequence feature extraction, and time sequence modeling is more accurately achieved.

The invention belongs to the field of fault diagnosis, and particularly relates to a fault detection method based on sparse representation, and the method comprises the steps: a preprocessing step: collecting normal working condition data samples under a preset working condition, and constructing a fault dictionary matrix; a control limit determination step of determining a sparse representation reconstruction error control limit and a distance control limit interval based on the fault dictionary matrix; in the field monitoring step, working field data are collected, and sparse representation reconstruction error statistics and distance statistics are calculated; and a fault detection step: judging whether the system has a fault by comparing the relationship between the statistical magnitude and the control limit. According to the method, accurate detection can be carried out under the condition that data does not obey normal distribution, the method has high detection sensitivity to tiny faults, meanwhile, the method is suitable for industrial process data of Gaussian and non-Gaussian distribution, and the accuracy and reliability of fault detection are effectively improved.

The invention discloses an industrial process fault monitoring method, which comprises the following steps of: acquiring real-time multivariable time sequence data of an industrial process, and performing standardized preprocessing; extracting the preprocessed data through a multi-scale feature extraction module to generate multi-scale potential feature representation; processing the multi-scale potential feature representation through a double-branch slow feature analysis module, and extracting a slow feature representation with a stable time sequence; based on the slow feature representation and the reconstruction data, constructing a T2 statistic and an SPE statistic; comparing with a preset control limit according to the T2 statistic and the SPE statistic; and judging whether the industrial process fails or not according to the comparison result. According to the method, the multi-scale features and the time sequence stable features of the industrial process data are effectively extracted, and the fault detection performance is remarkably improved.

The invention discloses a wind turbine generator running state monitoring method based on multi-source heterogeneous data, and the method comprises the steps: collecting SCADA multi-source heterogeneous data, removing abnormal samples, and carrying out the standardization; converting the collected non-normal data into approximate multivariate normal distribution based on a space rank; an elastic network penalty likelihood function is constructed, and mean vector estimation representing SCADA sparse fault signals is solved; calculating a basic monitoring statistic based on mean vector estimation and performing robust optimization to obtain a robust monitoring statistic and an upper control limit; real-time robust monitoring statistics are calculated, and if the real-time robust monitoring statistics exceed the upper control limit, a fault alarm is given out; and after a fault alarm is triggered, reversely deducing mean vector estimation according to the optimal penalty coefficient, and identifying and outputting fault component information. According to the method, the problems of complex SCADA data distribution, variable grouping correlation interference, difficulty in sparse fault identification, insufficient controlled samples and high multi-unit monitoring cost in the prior art are solved.