Intelligent sensing analysis system and analysis method based on distributed cloud computing

Abstract

The invention provides an intelligent sensing analysis system and analysis method based on distributed cloud computing, and the system comprises an equipment control module which is used for obtaining the original measurement data of at least one instrument equipment; the network module is connected with the equipment control module and is used for transmitting the original measurement data acquired by the equipment control module; and the server module is connected with the network module, predicts and analyzes the original measurement data transmitted by the network module, and marks the original measurement data and an analysis result. The method has the advantages that the method is carried out in a full-automatic or semi-automatic mode at the server side, manual intervention is not needed, a complete closed loop of measurement data uploading, measurement data prediction, model automatic optimization and measurement result sending is formed, analysis of instrument measurement data becomes a part of an automatic process, the difficulty of obtaining accurate detection data is reduced, and the detection efficiency is improved. And manpower resource consumption is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of measuring instruments and equipment, and in particular relates to an intelligent perception analysis system and analysis method based on distributed cloud computing. Background Art

[0002] A wide variety of measuring instruments are emerging. In some professional testing institutions or corporate testing centers, a single testing laboratory may contain dozens or even hundreds of various instruments. Their proper operation and proper management require significant human and material resources. Even instruments of the same type can vary from manufacturer to manufacturer or model to model, creating numerous difficulties and confusion for users. In fact, many testing centers also face the challenge of managing multiple testing laboratories in different locations. Due to variations in instrument performance and the varying quality of personnel, obtaining accurate and reliable measurement results is a common challenge faced by all organizations. In particular, for instruments based on spectroscopy, chromatography, and mass spectrometry, the raw data often requires specialized analysis algorithms or models to achieve optimal results. Establishing, optimizing, and maintaining these algorithms and models often requires specialized personnel support.

[0003] In order to deal with such difficulties, people have proposed a variety of management methods, such as the introduction of software such as Laboratory Information Management System (LIMS), which has improved the digitalization of laboratory management. However, there are few mature solutions to the difficulties faced in instrument application. Summary of the Invention

[0004] In view of the above problems, the present invention provides an intelligent perception analysis system and analysis method based on distributed cloud computing to solve the above or other former problems existing in the prior art.

[0005] To solve the above technical problems, the present invention adopts a technical solution: an intelligent perception analysis system based on distributed cloud computing, comprising:

[0006] A device control module, configured to obtain raw measurement data of at least one instrument;

[0007] A network module is connected to the device control module and is used to transmit the original measurement data obtained by the device control module;

[0008] The server module is connected to the network module, performs predictive analysis on the original measurement data transmitted by the network module, and marks the original measurement data and analysis results.

[0009] Furthermore, the server module is a distributed cloud server, including:

[0010] A file server for storing received raw measurement data;

[0011] Model server, used to store model sets and perform quantitative or qualitative analysis on raw data;

[0012] Algorithm server, used to store algorithms;

[0013] Database server, used to store analysis results.

[0014] Furthermore, the network module includes an Internet network, an Intranet network or an Extranet network, and the network module uses an encryption protocol to transmit data.

[0015] Furthermore, the encryption protocol method includes MD5 encryption, DES encryption or RSA encryption.

[0016] Furthermore, the device control module obtains the original measurement data by reading a database or a stored data file.

[0017] Furthermore, the analysis and prediction values ​​of the server module include the T value and Q value of each original measurement data.

[0018] Furthermore, the algorithms in the algorithm server include competitive ensemble learning algorithms.

[0019] An analysis method for an intelligent perception analysis system based on distributed cloud computing uses the above-mentioned intelligent perception analysis system based on distributed cloud computing to perform predictive analysis on the original measurement data of at least one instrument and equipment. The equipment control module obtains the original measurement data of each instrument and equipment, and the original measurement data is transmitted to the server module through the network module. The server module analyzes and predicts the original data, calculates the T value and Q value of each original measurement data, and marks the original measurement data and analysis results.

[0020] Furthermore, when the server module performs predictive analysis on the original measurement data, the model server and the algorithm server perform quantitative analysis or qualitative analysis on the original measurement data and store the analysis results in the database server. The analysis results are pushed to the corresponding terminal through the file server or transmitted to the third-party program through the cloud data interface.

[0021] Furthermore, the original measurement data of each instrument is used as an original sample set, and the model server and the algorithm server perform quantitative analysis or stereotyping analysis on the original measurement data, including:

[0022] S1: A part of the samples are extracted from the original sample set using the bootstrap method as the training set, and the rest of the samples are used as the validation set;

[0023] S2: Call the algorithm in the algorithm server, use a training set to obtain a model, obtain multiple models, use each model to analyze the samples in the validation set, and record the evaluation of the analysis results;

[0024] S3: Automatically update the parameters used in the algorithm, repeat step S2, sort the generated models according to the evaluation results, and retain the top n models, where n is not less than 50;

[0025] S4: Traverse all algorithm parameter combinations, repeat step S3, and save the final first z models, where z is not less than 50;

[0026] S5: The saved z models are saved as a model set in the model server;

[0027] S6: The model set uses the voting method to obtain the qualitative analysis results, and the mean of the model set is calculated as the quantitative analysis result.

[0028] Furthermore, the measurement data marked with abnormal Q values ​​or abnormal T values ​​are used as increments of the sample set to re-run the competitive ensemble learning algorithm of the model and optimize the model in the algorithm server.

[0029] Due to the adoption of the above technical solution, the intelligent perception analysis system based on distributed cloud computing uses a device control module to obtain measurement data from multiple instruments and equipment. The measurement data is transmitted to the server module through the network module. The server module performs quantitative or qualitative analysis on the measurement data received from each instrument and equipment as required, calculates the T value and Q value of the measurement data, and marks the original measurement data and analysis results. The analysis method can be performed in a fully automatic or semi-automatic manner on the server side without human intervention, forming a complete closed loop of measurement data uploading, measurement data prediction, model automatic optimization, and measurement result sending. It is particularly suitable for application scenarios with multiple measuring instruments, multiple test plans, multiple types of samples, and multiple test locations, such as third-party testing, analysis and testing centers in large groups, etc., applying the computing power of the cloud server and utilizing the continuously self-correcting machine learning algorithm to replace some manual decisions, making the analysis of instrument measurement data part of the automated process, reducing the difficulty of obtaining accurate test data and reducing human resource expenditure. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 1 is a schematic structural diagram of an intelligent perception analysis system based on distributed cloud computing according to an embodiment of the present invention;

[0031] Figure 2 FIG. 1 is a schematic diagram of a guiding method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0033] Figure 1 A structural schematic diagram of an embodiment of the present invention is shown. This embodiment relates to an intelligent perception analysis system and analysis method based on distributed cloud computing, which is used to perform predictive analysis on measurement data of multiple instruments and equipment of different types and manufacturers. The equipment control module and the server module are connected through a network module. The equipment control module transmits the original measurement data of the instrument and equipment to the server module through the network module. The server module automatically completes the analysis of the measurement data and the establishment and optimization of the measurement model with an intelligent algorithm, and completes cloud storage, cloud computing and information interaction, etc., so that the analysis of the measurement data of the instrument and equipment becomes part of the automated process, thereby reducing the difficulty of obtaining accurate detection data and reducing human resources.

[0034] An intelligent perception analysis system based on distributed cloud computing, such as Figure 1 and 2 As shown, it includes a device control module, a network module and a server module. The device control module is connected to the server module through the network module.

[0035] The device control module is used to obtain raw measurement data of at least one instrument. Here, the number of instruments is multiple, and they can be instruments of the same type or different types, or instruments from different manufacturers. The device control module is not limited to a specific instrument type or manufacturer. The device control module can simultaneously obtain raw measurement data from multiple instruments.

[0036] In some feasible embodiments, preferably, the instrument is a detection instrument of the spectral type, chromatographic type, mass spectrometric type, etc. that can obtain data containing complex sample information.

[0037] In some feasible embodiments, the device control module refers to software used to control instrument-end detection and obtain detection data, including but not limited to detection software that comes with each instrument or detection software developed by a third party, such as: Thermo Fisher Scientific's TQ software, FOSS's ISIscanNOVA software, Agilent's OpenLAB CDS, Waters' EmpowerChromatography Data Software, Shimadzu's LabSolutionsChromatographyData System, etc.

[0038] A network module is connected to the device control module and is used to transmit the original measurement data obtained by the device control module;

[0039] The server module is connected to the network module and performs predictive analysis on the original measurement data transmitted by the network module. According to actual needs, the server module performs qualitative analysis or quantitative analysis on the original measurement data, and marks the original measurement data and analysis results by calculating the T value and Q value of the original measurement data. On the one hand, the confidence level of the analysis results can be defined, and on the other hand, the sample detection data with special characteristics can be saved as model data to be built for automatic optimization of future prediction models.

[0040] Specifically, the above-mentioned device control module obtains the original measurement data by reading the database or reading the saved data file. In principle, it ensures the normal operation of the original measurement program of the instrument and equipment without changing the operating procedures established by the user. This method also reduces the difficulty of obtaining data. There is no need to know too much about the detailed operating details of the target equipment. The instrument and equipment are only responsible for correct measurement and generating measurement data. There are different clients in the device control module, and the data measured by different instruments and equipment are obtained through different clients.

[0041] The original measurement data obtained by the equipment control module includes the data measured by the instrument and the relevant sample information measured by the instrument.

[0042] The above-mentioned network module includes the Internet network, Intranet network or Extranet network, which can be selected according to actual needs. The network module adopts encryption protocol to transmit data to ensure the security of data during data transmission through the network, wherein the encryption protocol includes MD5 encryption, DES encryption or RSA encryption.

[0043] The above-mentioned server module is a distributed cloud server with higher fault tolerance and redundancy. It has extremely low latency and can provide fast response, making it convenient to expand capacity as business data increases. The distributed cloud server includes storage and computing functions, including:

[0044] The file server is used to store the received raw measurement data and transmit the prediction and analysis results to the corresponding terminal, which includes a mobile phone, computer or display terminal;

[0045] Model server, used to store model sets and perform quantitative or qualitative analysis on raw data;

[0046] The algorithm server is used to store algorithms. The algorithms in the algorithm server include a competitive ensemble learning algorithm to prevent the common over-learning problem in AI learning. The competitive ensemble learning algorithm uses bagging for averaging to achieve the minimum variance. The algorithms in the algorithm server can be continuously increased. Any algorithm suitable for the current application can be uploaded to the algorithm library after verification.

[0047] Database server, used to store analysis results.

[0048] The server module's analysis and prediction values ​​include the T and Q values ​​of the measurement data. Specifically, when the distributed cloud server analyzes and predicts the raw measurement data, the algorithm and model files used are sourced from the algorithm server and model server, respectively. The T and Q values ​​of the measurement data are simultaneously calculated in the analysis and prediction values. The T and Q values ​​of each measurement data item are used to label each prediction result. This allows for defining the confidence level of the analysis result and for saving unique sample test data as data to be built into the model for automatic optimization of future prediction models.

[0049] The parameter T represents the distance between the spectrum and the model data set, which can be calculated using Mahalanobis distance, Minkowski distance, correlation distance, etc.

[0050] The parameter Q is used to characterize the residual after simulation fitting, and can usually be calculated using the squared prediction error (SPE).

[0051] The T value represents the distance between the sample and the model data, and therefore reflects whether the sample to be tested belongs to the known model data set. If the T value is too large, it usually means that the physical and chemical indicators of the sample to be tested exceed the range of the model data set. The Q value represents the residual between the sample data and the simulated data generated by the algorithm, and therefore reflects the effectiveness of the analysis and prediction algorithm. If the Q value is too large, it usually means that the predicted value of the sample to be tested is unreliable and the algorithm model needs to be optimized and adjusted.

[0052] An analysis method for an intelligent perception analysis system based on distributed cloud computing uses the above-mentioned intelligent perception analysis system based on distributed cloud computing to perform predictive analysis on the original measurement data of at least one instrument and equipment. The equipment control module obtains the original data measured by each instrument and equipment, and the original data is transmitted to the server module through the network module. The server module analyzes and predicts the original data and marks each prediction result.

[0053] When the server module performs predictive analysis on the original data, the model server and algorithm server perform quantitative analysis or qualitative analysis on the original data, and store the analysis and prediction results in the database server. The analysis and prediction results are pushed to the corresponding terminal through the file server or transmitted to the third-party program through the cloud data interface.

[0054] The device control module can simultaneously obtain measurement data from multiple instruments and equipment, and use the multiple raw data measured by each instrument and equipment as the original sample set. The model server and algorithm server perform quantitative analysis or stereotyping analysis on the raw data, including:

[0055] S1: Use the bootstrap method to extract a part of the samples from the original sample set as the training set, and obtain p training sets. The remaining samples are used as the validation set. Specifically, in each round, m training samples are extracted from the original sample set using the bootstrap method. A total of p rounds of extraction are performed to obtain p training sets. The p training sets are independent of each other, such as Figure 2 As shown;

[0056] S2: Call the algorithm in the algorithm server, use a training set to obtain a model, obtain p models, use each model to analyze the samples in the validation set, and record the evaluation of the analysis results. For each training set, call the algorithm in the algorithm server for calculation to obtain a model. After calculation of p training sets, p models are obtained. Specifically, the algorithm server stores a variety of algorithms. The algorithms contained in the algorithm server can be divided into two categories according to the application tasks performed. One is the regression algorithm for quantitative analysis, and the other is the classification algorithm for qualitative analysis.

[0057] Among them, the regression algorithms used for quantitative analysis include partial least squares regression (PLS) and its derived algorithms, support vector regression (SVR), random forest regression algorithm (Random Forest Regression), extreme learning machine (ELM), etc. The algorithms derived from partial least squares regression include kernel PLS (KPLS), local PLS (LPLS), sparse PLS (SPLS), etc.

[0058] Classification algorithms used for qualitative analysis include SIMCA (Soft Independent Modeling of Class Analogy), K-nearest neighbors (KNN) and decision tree algorithms.

[0059] The various algorithms in the algorithm server are pre-configured, and one or more algorithms can be selected to analyze the uploaded data based on the specific application example.

[0060] All pre-configured algorithms on the server must be validated using a validation set to evaluate their accuracy. Common evaluation methods for quantitative analysis include mean absolute error (MAE), standard deviation (SD), correlation statistics, and F-tests. Common evaluation methods for qualitative analysis include accuracy, precision, and recall.

[0061] S3: Automatically update the parameters used in the algorithm, repeat step S2, sort the generated models according to the evaluation results, and retain the top n models, where n is not less than 50; according to the different quantitative / qualitative algorithms set in the algorithm server, the parameters in the model can be continuously optimized and the parameters used in the algorithm can be updated. For example, the selection of data range and preprocessing method in the partial least squares regression (PLS) algorithm; the selection of kernel function (Kernel) type and kernel function coefficient, and the setting of penalty parameters in the support vector machine regression (SVR) algorithm; the setting of the number of hidden layer nodes, random projection dimension, learning rate, and activation function in the extreme learning machine (ELM) algorithm; the selection of data range, preprocessing method, and number of components in the SIMCA algorithm; the selection of the number of neighbors, weights, and distance calculation in the KNN algorithm; and the impurity calculation method, random pattern parameters, maximum tree depth, minimum number of branch samples, and category weight coefficient in the decision tree algorithm.

[0062] To better optimize parameters, the grid optimization method, that is, the exhaustive search method, is usually used to find the optimal parameters by traversing all possible combinations of hyperparameters. Algorithm servers are usually equipped with high computing power. Using this method for optimization can ensure that the optimal parameter combination for each algorithm is found.

[0063] S4: Traverse all algorithm parameter combinations, repeat step S3, and save the final first z models, where z is not less than 50;

[0064] S5: The saved z models are saved as a model set in the model server;

[0065] S6: The model set uses the voting method to obtain the qualitative analysis results, and the mean of the model set is calculated as the quantitative analysis result.

[0066] When the model needs to be optimized, the measurement data stored in the file server, especially the measurement data marked with abnormal Q values ​​or abnormal T values, can be used as an increment of the sample set to re-run the competitive ensemble learning algorithm of the model to obtain better prediction results.

[0067] If you need to control the size of the sample set and prevent computational redundancy, you can apply a screening algorithm to the new sample set before optimizing the model. Common screening algorithms include filtering, wrapping, and embedding. The screening algorithms used are stored in the algorithm server.

[0068] In some feasible embodiments, preferably, the screening algorithm is a correlation coefficient method or a partial least squares method.

[0069] In order to better describe the analysis method of the intelligent perception analysis system based on distributed cloud computing, the processing flow of a set of near-infrared spectral data is used as an example for detailed explanation.

[0070] A processing plant of a feed group has two near-infrared instruments, A and B, from different manufacturers and with different principles. The platform software of instrument A is OPUS, and the platform software of instrument B is ISIscan NOVA. Users measure samples by operating the platform software.

[0071] To more conveniently obtain instrument test data, real-time spectral data is obtained from OPUS and ISIscan NOVA through equipment control module software developed by a third party. Each spectral data includes at least X-axis data and Y-axis data as well as the sample name in the factory's unified format.

[0072] After obtaining the spectral data, the equipment control module software encrypts the data and automatically uploads it to the group's internal server group through the internal network. The server group includes a file server, model server, algorithm server and data server.

[0073] After the uploaded spectral data enters the file server, the test sample is determined to be corn by identifying the format and content of the sample name. Then, the corresponding pre-stored corn quantitative model is called from the model server to predict the content of each component of the sample.

[0074] Before model prediction, spectral data must be preprocessed by the algorithm server. Spectral data measured by instrument A undergoes interpolation and piecewise direct standardization (PDS) before prediction. Spectral data measured by instrument B undergoes X-axis transformation, smoothing, interpolation, and PDS standardization before prediction. Spectral data from different instruments undergoes different preprocessing processes to obtain uniformly formatted and standardized data for model prediction.

[0075] In the model prediction, not only are the predicted values ​​for each component of each spectral data set calculated, but T and Q values ​​are also calculated. All predicted values, T and Q values ​​are stored on the data server. If the Q value is within the allowable error range, the prediction data is highly reliable, and the prediction results are transmitted to the group's business platform via a cloud data interface.

[0076] When a quantitative model needs to be established or optimized, data that meets the requirements will be screened from the T values ​​and Q values ​​stored in the data server. Usually, spectral data with larger T values ​​exceeding the threshold are screened and added to the model as new representative data.

[0077] All newly added spectral data and their component values ​​are pushed to the algorithm server, which has preset algorithms suitable for near-infrared quantitative modeling, such as PLS, ELM, SVM, etc. The subsequent process is as follows:

[0078] S1: Assume that there are 1000 data points used to build a quantitative model. Use a random classification algorithm to divide the data into two parts: a training set and a validation set. Generally, the training set accounts for 80% and the validation set accounts for 20%. By calling the random classification algorithm multiple times, a total of 100 different training and validation sets are generated.

[0079] S2: Each set of training data is used to generate a model using the preset PLS, ELM, and SVM modeling processes. This means there are 100 PLS models, 100 ELM models, and 100 SVM models. The validation data corresponding to each training set is substituted into the corresponding model, and the prediction results are statistically analyzed, such as the root mean square error (RMSEP), coefficient of determination (R2), and mean absolute error (MAE).

[0080] Statistical analysis algorithms are often used to quantify the distribution or trend of a set of data, thereby evaluating the quality of data models.

[0081] For example, the Root Mean Square Error of Prediction (RMSEP) is a commonly used metric that characterizes the difference between the predicted value and the actual value. Its core idea is to measure the error in the prediction result and provide a concise numerical value to help us understand the accuracy of the model. Its calculation formula is as follows:

[0082]

[0083] Among them, y i is the true value, is the model prediction value, and n is the number of data.

[0084] Coefficient of determination (R 2 ) characterizes the degree of fit of the data model to the data, that is, the proportion of variation in the dependent variable (target variable) that the model can explain. Its value range is usually between 0 and 1. Its calculation formula is as follows:

[0085]

[0086] Among them, y i is the true value, is the model prediction value, is the mean of the true values.

[0087] Mean Absolute Error (MAE) represents the average absolute error between the model's predicted value and the true value. Its calculation formula is as follows:

[0088]

[0089] where y i is the true value, is the model prediction value, and n is the number of data.

[0090] Sorted by model statistical indicators, only the top 100 best models are retained.

[0091] S3: According to the three algorithms preset in S2, PLS, ELM and SVM, update the parameters in the model, such as: spectral preprocessing method, wavelength screening coefficient, etc. in the PLS model; the number of hidden layer neurons, input weight coefficient, output weight coefficient, etc. in the ELM model; penalty coefficient, kernel function type, polynomial order, etc. in the SVM model.

[0092] Taking the PLS modeling algorithm as an example, commonly used spectral preprocessing methods include mean centering, standardization, vector normalization, subtracting a straight line, eliminating constant offsets, smoothing, derivatives, standard normal variable transformation, detrending algorithm, multivariate scatter correction, Fourier transform, wavelet transform, etc. These preprocessing methods can also be used in combination. Wavelength screening methods are used to simplify the model and eliminate irrelevant or nonlinear variables, thereby obtaining a correction model with stronger predictive ability and more robustness. Commonly used wavelength screening methods include correlation coefficient method, variance analysis method, stepwise regression method, interval partial least squares method, uninformative variable elimination method, Monte Carlo uninformative variable elimination method, genetic algorithm, moving window partial least squares regression, competitive adaptive reweighted sampling method, etc. Usually, wavelength screening methods do not need to be mixed.

[0093] Using different spectral preprocessing methods and wavelength screening methods can form complex parameter combinations. Grid search is usually the most suitable method. It is a commonly used model hyperparameter optimization method that searches for the best model parameter combination by traversing a given parameter grid.

[0094] In use, the grid method is usually used to update the parameters. Each time the parameters are updated, the process returns to step S2 to rebuild the model and calculate the statistical results of the model verification set. Finally, the generated models are re-ranked according to the evaluation results.

[0095] S4: After traversing all parameter combinations using the grid method, the statistical results of the final model validation set are ranked, and the top 50 models are taken as the final established / optimized model set.

[0096] S5: The new model set is stored in the model server, replacing the original model set.

[0097] S6: Each time a sample spectrum is predicted, the 50 prediction results of the model set are averaged and output as the final prediction result.

[0098] Due to the adoption of the above technical solution, the intelligent perception analysis system based on distributed cloud computing uses a device control module to obtain measurement data from multiple instruments and equipment. The measurement data is transmitted to the server module through the network module. The server module performs quantitative or qualitative analysis on the measurement data received from each instrument and equipment as required, calculates the T value and Q value of the measurement data, and marks the original measurement data and analysis results. The analysis method can be performed in a fully automatic or semi-automatic manner on the server side without human intervention, forming a complete closed loop of measurement data uploading, measurement data prediction, model automatic optimization, and measurement result sending. It is particularly suitable for application scenarios with multiple measuring instruments, multiple test plans, multiple types of samples, and multiple test locations, such as third-party testing, analysis and testing centers in large groups, etc., applying the computing power of the cloud server and utilizing the continuously self-correcting machine learning algorithm to replace some manual decisions, making the analysis of instrument measurement data part of the automated process, reducing the difficulty of obtaining accurate test data and reducing human resource expenditure.

[0099] The embodiments of the present invention are described in detail above, but the contents described are only preferred embodiments of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An intelligent perception analysis system based on distributed cloud computing, characterized by: include: A device control module, configured to obtain raw measurement data of at least one instrument; a network module, connected to the device control module, and configured to transmit the original measurement data acquired by the device control module; The server module is connected to the network module, performs prediction analysis on the original measurement data transmitted by the network module, and marks the original measurement data and analysis results.

2. The intelligent perception analysis system based on distributed cloud computing according to claim 1 is characterized in that: The server module is a distributed cloud server, including: A file server for storing received raw measurement data; Model server, used to store model sets and perform quantitative or qualitative analysis on raw data; Algorithm server, used to store algorithms; Database server, used to store analysis results.

3. The intelligent perception analysis system based on distributed cloud computing according to claim 1 or 2, characterized in that: The network module includes an Internet network, an Intranet network or an Extranet network, and the network module adopts an encryption protocol to transmit data.

4. The intelligent perception analysis system based on distributed cloud computing according to claim 3 is characterized in that: The encryption protocol includes MD5 encryption, DES encryption or RSA encryption.

5. The intelligent perception analysis system based on distributed cloud computing according to claim 1, 2 or 4, characterized in that: The device control module obtains the original measurement data by reading a database or a stored data file.

6. The intelligent perception analysis system based on distributed cloud computing according to claim 5, characterized in that: The analysis prediction value of the server module includes the T value and Q value of each of the original measurement data.

7. The intelligent perception analysis system based on distributed cloud computing according to claim 2, characterized in that: The algorithm in the algorithm server includes a competitive ensemble learning algorithm.

8. An analysis method for an intelligent perception analysis system based on distributed cloud computing, characterized by: The distributed cloud computing-based intelligent perception analysis system according to any one of claims 1 to 7 is used to perform predictive analysis on the raw measurement data of at least one instrument and equipment, wherein the equipment control module obtains the raw measurement data of each instrument and equipment, and the raw measurement data is transmitted to the server module through the network module. The server module analyzes and predicts the raw data, calculates the T value and Q value of each raw measurement data, and marks the raw measurement data and analysis results.

9. The analysis method of the intelligent perception analysis system based on distributed cloud computing according to claim 8, characterized in that: When the server module performs predictive analysis on the original measurement data, the model server and the algorithm server perform quantitative analysis or qualitative analysis on the original measurement data and store the analysis results in the database server. The analysis results are pushed to the corresponding terminal through the file server or transmitted to a third-party program through the cloud data interface.

10. The analysis method of the intelligent perception analysis system based on distributed cloud computing according to claim 9, characterized in that: The original measurement data of each instrument and device is used as an original sample set, and the model server and the algorithm server perform quantitative analysis or stereotyping analysis on the original measurement data, including: S1: A part of the samples are extracted from the original sample set using the bootstrap method as the training set, and the rest of the samples are used as the validation set; S2: Call the algorithm in the algorithm server, use a training set to obtain a model, obtain multiple models, use each model to analyze the samples in the validation set, and record the evaluation of the analysis results; S3: Automatically update the parameters used in the algorithm, repeat step S2, sort the generated models according to the evaluation results, and retain the top n models, where n is not less than 50; S4: Traverse all algorithm parameter combinations, repeat step S3, and save the final first z models, where z is not less than 50; S5: The saved z models are saved as a model set in the model server; S6: The model set adopts a voting method to obtain a qualitative analysis result, and the mean value in the model set is calculated as a quantitative analysis result.

11. The analysis method of the intelligent perception analysis system based on distributed cloud computing according to claim 10, characterized in that: The measurement data marked with abnormal Q values ​​or abnormal T values ​​are used as increments of the sample set to re-perform the competitive ensemble learning algorithm of the model, and the model in the algorithm server is optimized.

Images