CN113420799A - Sample enhancement method, model training method and system - Google Patents
Sample enhancement method, model training method and system Download PDFInfo
- Publication number
- CN113420799A CN113420799A CN202110646519.7A CN202110646519A CN113420799A CN 113420799 A CN113420799 A CN 113420799A CN 202110646519 A CN202110646519 A CN 202110646519A CN 113420799 A CN113420799 A CN 113420799A
- Authority
- CN
- China
- Prior art keywords
- data
- sample
- variables
- input data
- variable
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 86
- 238000012549 training Methods 0.000 title claims abstract description 36
- 238000004088 simulation Methods 0.000 claims abstract description 99
- 238000009826 distribution Methods 0.000 claims abstract description 55
- 230000007246 mechanism Effects 0.000 claims abstract description 43
- 238000012163 sequencing technique Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims abstract description 10
- 238000004364 calculation method Methods 0.000 claims description 77
- 238000004519 manufacturing process Methods 0.000 claims description 44
- 230000001502 supplementing effect Effects 0.000 claims description 10
- 238000004458 analytical method Methods 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000012800 visualization Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 5
- 238000010219 correlation analysis Methods 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 3
- 238000010946 mechanistic model Methods 0.000 claims 1
- 230000009286 beneficial effect Effects 0.000 description 10
- 238000003860 storage Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 239000000047 product Substances 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000007418 data mining Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000003921 oil Substances 0.000 description 2
- 238000007670 refining Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000003889 chemical engineering Methods 0.000 description 1
- 239000010779 crude oil Substances 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005293 physical law Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000010206 sensitivity analysis Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000005292 vacuum distillation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
Abstract
The invention discloses a sample enhancement method, a model training method and a system, relates to the technical field of data processing, and can improve the quantity and quality of sample data and reduce the sample generation time. The sample enhancement method comprises the following steps: dividing variables into independent variables, partial associated variables and residual associated variables; randomly generating independent variable data, calculating partial associated variable data according to the independent variable data and a specified formula, taking the two types of data as input, and calculating by adopting a simplified model to obtain residual associated variable data; and combining the three types of data to be used as input data of a strict mechanism model simulation sample. Dividing the sample input data into a plurality of sub-sample sets according to Euclidean distances, sequencing the samples in each sub-sample set, sequentially carrying out step-by-step simulation by using a strict mechanism model according to the sample sequence, obtaining sample output data, and combining the sample output data with corresponding sample input data to obtain complete sample data. And the sample distribution is visualized through the samples, and the samples in the sparse area are supplemented.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a sample enhancement method, a model training method and a model training system.
Background
With the advent of the industrial big data age, research and application of a modeling method based on data driving in the modern process industry have attracted extensive attention.
In recent years, the process industry is continuously invested in informatization, online instruments and analytical equipment are continuously used, massive production process data are accumulated, and a good data base is laid for the application of a data-driven modeling method; the rapid development of technologies such as big data and deep learning provides rich and efficient algorithms, and the performance of intelligent chips such as GPU and TPU is continuously improved, so that a technical basis is laid for the large-scale industrial application of data-driven modeling.
Although the production data is large, the data diversity is small because the operation of the device is less varied. In addition, the data distribution is not balanced, and a large amount of data is missing or the data quality is not high. In order to make the model cover the optimal operation area as much as possible and accurately reflect the real-time characteristics of the device, it is necessary to expand the data coverage and uniform data distribution density.
Aiming at the requirement, a strict mechanism model simulation method is generally adopted to expand the data. In this process, because the amount of data is large, input data required for simulation is generally generated by a method of randomly generating data. For a device with a complex flow, the method has the problems of low simulation convergence rate and long time consumption for producing enough samples, and in addition, the data distribution is not necessarily uniform.
Disclosure of Invention
The invention aims to provide a sample enhancement method, a model training method and a system suitable for oil refining and chemical engineering devices, which can improve the quantity and quality of sample data and reduce time cost.
In order to achieve the above object, a first aspect of the present invention provides a sample enhancement method, comprising:
dividing variables into independent variables, partial associated variables and residual associated variables;
acquiring a historical data distribution range corresponding to each variable according to historical data corresponding to each variable in historical production data;
randomly generating a plurality of random data in a distribution range corresponding to each independent variable, calculating intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, calculating by taking the random data and the intermediate data as input by adopting a simplified model which is trained in advance, judging whether a calculation result is converged, and acquiring residual data corresponding to residual associated variables when the calculation result is converged;
the random data, the intermediate data and the residual data are collated and combined into input data used for subsequent simulation;
clustering input data by adopting Euclidean distance to obtain a plurality of groups of sub-sample sets, wherein the sub-sample sets comprise a plurality of samples, and sequencing the samples in each group of the sub-sample sets;
performing analog calculation on samples in each sub-sample set in batches and step by step according to a strict mechanism model trained in advance to obtain output data;
summarizing and combining input data and corresponding output data of each sample to obtain a plurality of complete sample data obtained by sample enhancement;
and performing visualization analysis on the distribution range of the obtained multiple sample data, and supplementing the samples in the sparse region.
Preferably, the method for obtaining the historical data distribution range corresponding to each variable according to the historical data corresponding to each variable in the historical production data includes:
and counting the upper limit and the lower limit of the historical data corresponding to each variable in the historical production data, and taking the range of the upper limit and the lower limit as the historical data distribution range of the corresponding variable.
Preferably, the method for randomly generating a plurality of random data within the distribution range corresponding to each independent variable includes:
and randomly generating a plurality of random data based on the uniform read data distribution range corresponding to each independent variable.
Preferably, the method for calculating the intermediate data corresponding to the partial associated variables according to the random data corresponding to the independent variables comprises:
analyzing the data correlation between partial associated variables and independent variables in the historical production data by adopting a correlation analysis method based on random data corresponding to each independent variable, and obtaining pairwise mutual influence relation of the partial associated variables and the independent variables to obtain a relational expression of the partial associated variables and the independent variables;
according to the relational expression, taking the random data of the independent variable as input to calculate the intermediate data of the related variable of the corresponding part;
and merging the intermediate data of the partial associated variables belonging to the same group and the random data of the independent variables to form the input data of the same group.
Further, the method for calculating and judging whether the calculation result is converged by using the pre-trained simplified model and taking the random data and the intermediate data as input, and acquiring the residual data corresponding to the residual associated variables during convergence comprises the following steps:
simulating input data comprising random data and intermediate data by adopting a pre-trained simplified model, if the simulation result is convergence, considering the group of input data as valid sample data, and using the simulation result as residual data corresponding to residual associated variables, and if the simulation result is non-convergence, considering the group of input data as invalid sample data, and removing the input data;
and combining the residual data corresponding to the residual associated variables belonging to the same group with the input data of the partial associated variables and the independent variables to jointly serve as the input data of the same group of the strict mechanism model.
Further, the method for obtaining the output data by sequentially carrying out batch and step-by-step simulation calculation on the samples in each sub-sample set through a strict mechanism model trained in advance comprises the following steps:
distributing the sub-sample sets to different simulation nodes, calculating input data by adopting the same pre-trained strict mechanism model of each simulation node, and calculating sample input data and judging whether an output result is converged by each simulation node according to the sample sequence in the received sample set;
if the convergence is detected, saving the output data in a first convergence sample, if the convergence is not detected, modifying and adjusting corresponding input data, inputting the modified and adjusted input data into a strict mechanism model again, and saving the converged output data in a second convergence sample;
combining the first convergence sample and the second convergence sample to obtain output data corresponding to the remaining meters;
wherein, if convergence, saving the output data after the first convergence sample further comprises: and performing energy balance verification on the sample data, if the verification result is energy balance, confirming convergence, and saving the output data in the first convergence sample, and if the verification result is energy imbalance, judging the output data to be not converged.
Further, the method for performing step-by-step simulation on a sample based on the pre-trained strict mechanism model comprises the following steps:
dividing all variables in input data into a first class, a second class and a third class, wherein each class at least comprises one variable; the simulation calculation is divided into two major steps, wherein the first major step comprises data replacement and calculation of first-class and second-class variables, and the second major step comprises data replacement and calculation of third-class variables;
adopting a working condition which is successfully simulated through a strict mechanism model in advance as a basic working condition, wherein the basic working condition comprises basic input data;
correspondingly dividing variables in each group of input data and basic input data into a first class, a second class and a third class, and taking sample data in the input data as a target value;
replacing data corresponding to the first type of variable in the basic input data with a corresponding target value in the input data;
meanwhile, data corresponding to the second type of variables in the basic input data are replaced according to rules, if the current target value does not affect convergence relative to the change direction of the basic data, the basic input data are replaced once according to the current target value, otherwise, the second type of variables need to be replaced step by step, the basic input data are gradually adjusted to the target value direction according to the preset step length, once the second type of variables are replaced, analog calculation is carried out, each replacement is carried out on the basis of the previous calculation, and the first large-step calculation is stopped until all the second type of variables reach the preset constraint condition or target value or the simulation does not converge;
and taking the data after the first large-step calculation as a basis, taking the sample data corresponding to the third type variable in the input data as a target value, taking the data corresponding to the basic input data and the third type variable as a basis, gradually adjusting and replacing according to a preset step length, inputting the data into a strict mechanism model for calculation, and stopping the second large-step simulation calculation until all the data corresponding to the third type variable reach a preset constraint condition or target value or the simulation is not converged.
Compared with the prior art, the sample enhancement method provided by the invention has the following beneficial effects:
the sample enhancement method provided by the invention comprises the steps of dividing variables into independent variables, partial associated variables and residual associated variables, counting the distribution range of historical data corresponding to each variable according to the historical data corresponding to each variable in historical production data, randomly generating a plurality of random data in the distribution range corresponding to each independent variable, calculating the intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, adopting a simplified model which is trained in advance and taking the random data and the intermediate data as input calculation and judging whether the calculation result is converged, acquiring the residual data corresponding to the residual associated variables when the calculation result is converged, finally sorting and combining the random data, the intermediate data and the residual data into input data used for subsequent simulation, and clustering the input data by adopting Euclidean distance to obtain a plurality of groups of subsample sets, the method comprises the steps that a plurality of samples are included in a sub-sample set, the samples in each sub-sample set are sequenced, simulation calculation is carried out on the samples in each sub-sample set in batches and step by step according to a strict mechanism model which is trained in advance to obtain output data, input data and corresponding output data of each sample are collected to obtain a plurality of complete sample data which are obtained through sample enhancement, visual analysis is carried out on the distribution range of the obtained plurality of sample data, and the samples in a sparse area are supplemented. The method has the advantages that the sample data can be rapidly amplified, time consumption for generating enough samples is reduced, the samples are uniformly distributed, the quantity and the quality of the sample data are improved, and the performance of model training is further improved.
A second aspect of the present invention provides a model training method, to which the sample enhancement method according to the above technical solution is applied, the model training method including:
training a device model for assisting production equipment optimization adjustment based on a plurality of sample data obtained by sample enhancement;
and optimizing and adjusting part or all operation corresponding variables in the production equipment according to the device model so as to enable the production equipment to be in the optimal working state.
Compared with the prior art, the beneficial effects of the model training method provided by the invention are the same as those of the sample enhancement method provided by the technical scheme, and are not repeated herein.
A third aspect of the present invention provides a sample enhancement system applied to the sample enhancement method described in the above technical solution, the system including:
the dividing unit is used for dividing the variables into independent variables, partial associated variables and residual associated variables;
the distribution reading unit is used for acquiring a historical data distribution range corresponding to each variable according to historical data corresponding to each variable in historical production data;
the screening unit is used for randomly generating a plurality of random data in the distribution range corresponding to each independent variable, calculating intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, then adopting a simplified model which is trained in advance and taking the random data and the intermediate data as input calculation and judging whether the calculation result is converged, and acquiring residual data corresponding to the residual associated variables when the calculation result is converged;
the data integration unit is used for sorting and combining the random data, the intermediate data and the residual data into input data used for subsequent simulation;
the device comprises a clustering and sequencing unit, a data processing unit and a data processing unit, wherein the clustering and sequencing unit is used for clustering input data by adopting Euclidean distance to obtain a plurality of groups of sub-sample sets, the sub-sample sets comprise a plurality of samples, and the samples in each group of sub-sample sets are sequenced to group and sequence the input data;
the simulation unit is used for sequentially carrying out step-by-step simulation calculation on the samples in each sub-sample set through a strict mechanism model which is trained in advance to obtain output data of the samples;
the summarizing unit is used for summarizing the input data and the corresponding output data of each sample to obtain a plurality of complete sample data obtained by sample enhancement;
and the sample supplementing unit is used for performing visualization analysis on the distribution range of the acquired multiple sample data and supplementing the samples in the sparse region.
Compared with the prior art, the beneficial effects of the sample enhancement system provided by the invention are the same as those of the sample enhancement method provided by the technical scheme, and are not repeated herein.
A fourth aspect of the present invention provides a model training system applied to the model training method according to the above technical solution, the system including:
the training unit is used for training a device model for assisting the optimization and adjustment of the production equipment based on a plurality of sample data obtained by sample enhancement;
and the debugging unit is used for optimizing and adjusting the variables corresponding to part or all of the operations in the production equipment according to the device model so as to enable the production equipment to be in the optimal working state.
Compared with the prior art, the beneficial effects of the model training system provided by the invention are the same as those of the model training method provided by the technical scheme, and are not repeated herein.
A fifth aspect of the invention provides a computer-readable storage medium having stored thereon a computer program for performing, when executed by a processor, the above-described sample enhancement method, and/or steps of a model training method.
Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the invention are the same as those of the sample enhancement method and/or the model training method provided by the technical scheme, and are not repeated herein.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic flow chart of a sample enhancement method according to an embodiment of the present invention;
fig. 2 is a schematic flow chart of a model training method according to an embodiment of the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example one
Referring to fig. 1, the present embodiment provides a sample enhancement method, including:
dividing variables into independent variables, partial associated variables and residual associated variables; acquiring a historical data distribution range corresponding to each variable according to historical data corresponding to each variable in historical production data; randomly generating a plurality of random data in a distribution range corresponding to each independent variable, calculating intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, calculating by taking the random data and the intermediate data as input by adopting a simplified model which is trained in advance, judging whether a calculation result is converged, and acquiring residual data corresponding to residual associated variables when the calculation result is converged; the random data, the intermediate data and the residual data are collated and combined into input data used for subsequent simulation; clustering input data by adopting Euclidean distance to obtain a plurality of groups of sub-sample sets, wherein the sub-sample sets comprise a plurality of samples, and sequencing the samples in each group of the sub-sample sets; performing step-by-step simulation calculation on each sample in the sub-sample set according to the sequence through a strict mechanism model which is trained in advance to obtain output data of the sample; summarizing and combining input data and corresponding output data of each sample to obtain a plurality of complete sample data obtained by sample enhancement; and performing visualization analysis on the distribution range of the obtained multiple sample data, and supplementing the samples in the sparse region.
The sample enhancement method provided by the invention comprises the steps of dividing variables into independent variables, partial associated variables and residual associated variables, counting the distribution range of historical data corresponding to each variable according to the historical data corresponding to each variable in historical production data, randomly generating a plurality of random data in the distribution range corresponding to each independent variable, calculating the intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, adopting a simplified model which is trained in advance and taking the random data and the intermediate data as input calculation and judging whether the calculation result is converged, acquiring the residual data corresponding to the residual associated variables when the calculation result is converged, finally sorting and combining the random data, the intermediate data and the residual data into input data used for subsequent simulation, and clustering the input data by adopting Euclidean distance to obtain a plurality of groups of subsample sets, the method comprises the steps of enabling a plurality of samples to be included in a sub-sample set, sequencing the samples in each sub-sample set, sequentially carrying out step-by-step simulation calculation on the samples in each sub-sample set through a pre-trained strict mechanism model to obtain output data, summarizing sample input data and corresponding output data to obtain a plurality of complete sample data obtained through sample enhancement, carrying out visual analysis on the distribution range of the obtained sample data, and supplementing the samples in a sparse area. The method has the advantages that the sample data can be rapidly amplified, time consumption for generating enough samples is reduced, the samples are uniformly distributed, the quantity and the quality of the sample data are improved, and the performance of model training is further improved.
In specific implementation, the independent variables comprise the material property and the processing amount and the operable variables in actual production, part of the related variables are variables obtained by calculating the independent variables, and the rest of the related variables are variables obtained by calculating the independent variables and part of the related variables by the simplified model.
In the above embodiment, the method for obtaining the historical data distribution range corresponding to each variable according to the historical data corresponding to each variable in the historical production data includes:
and counting the upper limit and the lower limit of the historical data corresponding to each variable in the historical production data, and taking the range of the upper limit and the lower limit as the historical data distribution range of the corresponding variable.
As an example, historical production data is read from a database storing production data, abnormal values are removed, and then upper and lower limits of the reading data of each meter are obtained according to the reading data of each meter, and the upper and lower limits are used as the distribution range of the reading data of each meter in sample data.
In the above embodiment, the method for randomly generating a plurality of random data within the distribution range corresponding to each independent variable includes: and randomly generating a plurality of random data based on the uniform read data distribution range corresponding to each independent variable.
In the above embodiment, the method for calculating the intermediate data corresponding to the partial associated variable according to the random data corresponding to the independent variable includes:
analyzing the data correlation between partial associated variables and independent variables in the historical production data by adopting a correlation analysis method based on random data corresponding to each independent variable, and obtaining pairwise mutual influence relation of the partial associated variables and the independent variables to obtain a relational expression of the partial associated variables and the independent variables; according to the relational expression, taking the random data of the independent variable as input to calculate the intermediate data of the related variable of the corresponding part; and merging the intermediate data of the partial associated variables belonging to the same group and the random data of the independent variables to form the input data of the same group.
In specific implementation, all variables in the sample data are divided into several groups of independent variables, partial associated variables, residual associated variables and residual variables needing to be calculated through a strict mechanism model, wherein data of the independent variables, the partial associated variables and the residual associated variables are used as input of the strict mechanism model, data of the residual variables are used as output of the strict mechanism model, and input data and output data of the same group form a data sample together. Illustratively, when sample enhancement is carried out on the oil refining atmospheric and vacuum distillation unit, the screened independent variables comprise crude oil processing amount and property, feeding property of each tower, pressure of each part of the tower, stripping steam consumption, middle section taking proportion, product property and the like, and related variables of a heat exchange network, such as splitter flow dividing rate, heat exchange effective use area and the like. The coverage of the independent variables is determined by the device capabilities and historical data.
The method comprises the steps of mining historical production data, analyzing meter reading data corresponding to partial associated variables and meter reading data corresponding to independent variables in sample data by adopting a correlation analysis method to obtain linear correlation coefficients or monotonous trends of the meter reading data and the meter reading data, and determining the independent variables influencing the partial associated variables according to the magnitude of the correlation coefficients. And determining the linear expression form or the non-linear expression form of the partial correlation variable and the independent variable by judging whether the partial correlation variable and the independent variable are linearly related or non-linearly related. And fitting corresponding coefficients in the expression according to the historical data so as to obtain a relational expression between the partial associated variables and the independent variables. And then randomly generating data corresponding to the independent variables in batch in a read data distribution range aiming at the independent variables, and then calculating the data corresponding to the independent variables by using a relational expression to obtain data corresponding to part of associated variables. If the data of the partial associated variables exceed the range, the corresponding independent variables need to be adjusted, so that the adjusted data of the partial associated variables exceeding the range and the adjusted data of the corresponding independent variables meet the range requirement at the same time. And finally, combining the data of the partial associated variables and the data of the independent variables belonging to the same group together to be used as input data of the same group, and realizing the first-level expansion of the sample data.
And then screening the input data by using the simplified model which is trained in advance, and eliminating the input data group which does not meet the simplified model. Or screening the input data by using the trained simplified model, eliminating the input data group which does not conform to the simplified model, calculating sample data of the remaining associated variable instruments while keeping the input data group which conforms to the simplified model, and merging the sample data of the remaining associated variable instruments belonging to the same group with part of the associated variable instruments and the input data of the independent variables to be used as the input data of the same group. Comparing the input data and the corresponding output data which follow the simplified model with the historical data distribution range, and if the input data and the corresponding output data do not exceed the historical data range, combining the input data and the corresponding output data to be used as one type of sample data of the simplified model; if the historical data range is exceeded, optimizing and adjusting the input data to obtain new input data, calculating by using the simplified model again to obtain corresponding output data, and combining the part of input data and the corresponding output data to be used as second-class sample data of the simplified model. And merging the first-class sample data and the second-class sample data of the simplified model to realize the second-level expansion of the sample data.
In the above embodiment, the method for calculating and judging whether the calculation result converges by using the pre-trained simplified model and using the random data and the intermediate data as input, and acquiring the remaining data corresponding to the remaining associated variables during convergence includes:
simulating input data comprising random data and intermediate data by adopting a pre-trained simplified model, if the simulation result is convergence, considering the group of input data as valid sample data, and using the simulation result as residual data corresponding to residual associated variables, and if the simulation result is non-convergence, considering the group of input data as invalid sample data, and removing the input data; and merging the residual data of the residual associated variables belonging to the same group with the input data of the partial associated variables and the independent variables to jointly serve as the input data of the same group of the strict mechanism model.
During specific implementation, a plurality of random data are generated in the historical data distribution range corresponding to each independent variable; and calculating intermediate data corresponding to the corresponding partial associated variables according to the given partial associated variable-independent variable relation and the random data corresponding to the independent variables, and combining the random data of the independent variables and the intermediate data of the partial associated variables to form a plurality of groups of input data of the simplified model. Then inputting each group of input data into the simplified model for simulation, if the simulation result is convergence, considering that the group of input data conforms to the basic physical law followed by the device, and taking the simulated output result as the output data generated by the simplified model, namely the residual data corresponding to the residual associated variables; if the simulation result is not convergent, the input sample data does not conform to the basic physical rule followed by the device, and the simulation does not generate an output result. Comparing the input data and the corresponding output data which follow the simplified model with the historical data distribution range, and if the input data and the corresponding output data do not exceed the historical data range, combining the input data and the corresponding output data to be used as partial sample data of the simplified model; if the historical data range is exceeded, optimizing and adjusting the input data to obtain new input data, calculating by using the simplified model again to obtain corresponding output data, combining the part of the input data and the corresponding output data to be used as the sample data of the other part of the simplified model, combining the two parts of the data to be used as the sample data of the simplified model, and then extracting the input data required by strict mechanism model simulation from the sample data of the simplified model.
Next, clustering and grouping a plurality of input data, sorting sample data in each sub-sample set by adopting a method for solving the problem of the traveling salesman for each group of sub-sample sets, planning a simulation sequence of the samples, striving to complete simulation calculation of all the samples in the shortest time according to the sequence, and constructing the sub-sample sets as follows: and selecting part of variables as reference variables of clustering, wherein the reference variables comprise variables of raw material properties, product properties, middle section load, device characteristics such as tower plate efficiency and the like, clustering all input data by using Euclidean distance to obtain a plurality of groups of sub-sample sets, and each sub-sample set comprises a plurality of samples. And then, sequencing the samples in each sub-sample set by adopting a method for solving the problem of the traveling salesman, planning the simulation sequence of the samples, and striving to finish the simulation of all the samples in the shortest time according to the sequence.
In the above embodiment, the method for performing batch and step-by-step simulation calculation on the samples in each group of sub-sample sets in sequence to obtain the output data through the strict mechanism model trained in advance includes:
distributing the sub-sample sets to different simulation nodes, calculating input data by adopting the same pre-trained strict mechanism model of each simulation node, and calculating sample input data and judging whether an output result is converged by each simulation node according to the sample sequence in the received sample set; if the convergence is detected, saving the output data in a first convergence sample, if the convergence is not detected, modifying and adjusting corresponding input data, inputting the modified and adjusted input data into a strict mechanism model again, and saving the converged output data in a second convergence sample; combining the first convergence sample and the second convergence sample to obtain output data corresponding to the remaining meters; wherein, if convergence, saving the output data after the first convergence sample further comprises: and performing energy balance verification on the sample data, if the verification result is energy balance, confirming convergence, and saving the output data in the first convergence sample, and if the verification result is energy imbalance, judging the output data to be not converged.
In the above embodiment, the method for performing step-by-step simulation on one of the samples based on the pre-trained strict mechanism model includes:
dividing all variables in input data into a first class, a second class and a third class, wherein each class at least comprises one variable; the simulation calculation is divided into two major steps, wherein the first major step comprises data replacement and calculation of first-class and second-class variables, and the second major step comprises data replacement and calculation of third-class variables;
adopting a working condition which is successfully simulated through a strict mechanism model in advance as a basic working condition, wherein the basic working condition comprises basic input data;
correspondingly dividing variables in each group of input data and basic input data into a first class, a second class and a third class, and taking sample data in the input data as a target value;
replacing data corresponding to the first type of variable in the basic input data with a corresponding target value in the input data;
meanwhile, data corresponding to the second type of variables in the basic input data are replaced according to rules, if the current target value does not affect convergence relative to the change direction of the basic data, the basic input data are replaced once according to the current target value, otherwise, the second type of variables need to be replaced step by step, the basic input data are gradually adjusted to the target value direction according to the preset step length, once the second type of variables are replaced, analog calculation is carried out, each replacement is carried out on the basis of the previous calculation, and the first large-step calculation is stopped until all the second type of variables reach the preset constraint condition or target value or the simulation does not converge;
and taking the data after the first large-step calculation as a basis, taking the sample data corresponding to the third type variable in the input data as a target value, taking the data corresponding to the basic input data and the third type variable as a basis, gradually adjusting and replacing according to a preset step length, inputting the data into a strict mechanism model for calculation, and stopping the second large-step simulation calculation until all the data corresponding to the third type variable reach a preset constraint condition or target value or the simulation is not converged.
During specific implementation, the batch input data is sent to a pre-configured distribution simulation platform, the distribution simulation platform distributes the batch input data to each simulation node to be calculated by adopting a strict mechanism model, and each simulation node obtains corresponding output data by utilizing the strict mechanism model to calculate. The distributed simulation platform adopts a parallel distributed queue frame, hundreds of high-performance computers can be used as simulation nodes for parallel computation only by interacting with a main console, data are uploaded to a database through the main console, then simulation tasks are sent to a message middleware to form a task queue, the hundreds of simulation nodes monitor the task queue at the same time, the simulation nodes receive task messages in parallel when tasks come, input data are read from the database for simulation computation, output data are written into the database finally, and after all the tasks are completed, a user extracts the output data corresponding to the input data one by one from the database through the main console.
And simulating the samples in each sub-sample set at the same simulation node according to a planned sequence, wherein the first sample of each sub-sample set is simulated by adopting the same pre-simulated basic working condition as basic data, and the simulation of the next sample is started from the second sample by adopting the data which is simulated successfully by the previous sample as the basic data. This simulation method saves time for the simulation of each subsample set, and thus saves time for the simulation of the entire sample.
In each simulation node, in order to ensure the convergence rate of the calculation of the strict mechanism model, each input data is processed by adopting a step-by-step method. Firstly, through sensitivity analysis and model mechanism analysis of each input variable, the influence of the input variable on model convergence in the change range is obtained, and the input variables are divided into the following classes according to the influence: the first is the feedstock properties and variables that do not affect model convergence; the second type is a variable whose convergence is not affected by a change of a known variable in a certain direction; the third category is a variable that cannot be judged to have an influence on convergence. Before simulation, related information of different variables needs to be set: providing target values of all input variables, and taking the input sample data formed in the previous step as the target values of the input variables; setting the direction of the second type variable which does not influence convergence, providing an adjusted step length or a step length formula, and setting a condition for stopping calculation; setting the adjustment step length or step length formula of the third type variable and giving the condition of stopping calculation. And on the basis of the working condition, adjusting the value of the input variable step by step to a target value of the sample or to reach a constraint condition or to simulate non-convergence. And (3) starting simulation, firstly inputting target values of variables which do not influence convergence into basic working conditions for calculation to obtain calculation results, inputting second type variable data on the basis of the calculation results, inputting the target values into a simulation file if the target values of the second type variables and the change direction of the last calculation result do not influence convergence, or sequentially increasing or decreasing the target values on the basis of the last simulation result according to step length, and calculating once every time until all the second type variables reach the target values or given constraint conditions or simulation does not converge. On the basis, the third-class variables are simulated, the variables are adjusted according to the step length on the basis of the last simulation result, and the calculation is carried out once every adjustment until all the third-class variables reach target values or given constraint conditions or the simulation is not converged. And if the variable inputs the target value, the simulation result is not converged, namely the simulation calculation has no solution, and the input data and the output data of the previous step are stored in the database to be used as converged sample data. While outputting the non-converged target value as a non-converged sample.
And acquiring sample data from the database, wherein the samples with output data in the database are classified as convergence samples, and the samples without the output data are unconverged samples. And judging whether the convergent sample really converges or not through energy balance calculation, wherein the convergent sample is determined if the energy balance is true, and the unconverged sample is determined if the energy balance is false. And visualizing and counting the output variable data of the convergence sample, if the output variable is within the required range, keeping the output variable in the convergence sample to obtain a first convergence sample, and if the output variable is beyond the required range, combining the first convergence sample with the non-convergence sample.
And transforming the non-convergence sample, finding input data corresponding to the output data to be adjusted through data mining, regressing a relational expression of the output data and the input data, calculating the value of the input data according to the relational expression and the range of the output data, and putting the adjusted input data into a distribution simulation platform to calculate and extract the convergence sample to obtain a second convergence sample.
Combining the first convergence sample and the second convergence sample, analyzing the distribution of the sample data, performing distribution check on the enhanced sample data, stopping sample enhancement if the coverage is uniformly distributed, or performing a sample supplement on a sparse region of the sample data, and generating a sample of the sparse region by using a sample enhancement method.
For example, a headAnalyzing historical distribution of variables such as feeding amount and property, steam stripping steam introduction amount, product property and the like in sample data, generating random sample points by adopting a Latin hypercube algorithm on multidimensional variables, and performing strict mechanism model simulation calculation on the sample points by adopting a calculation demand list of a distributed simulation platform after performing basic filtration of a simplified model and calculating correlation variables such as heat extraction amount of a middle section. After the simulation calculation returns a result, whether convergence is needed to be further analyzed through energy balance, if the energy balance is considered to be the convergence, otherwise the convergence is not needed. And for the unconverged sample, automatically adjusting the designated variable according to the error information. For the converged sample, it is necessary to judge whether the tray temperature range, the product quality range, and the like are within the required coverage range. If the input data is not within the range, the input data is obtained according to the response relation between the input data and the output data obtained through data mining and the variation range of the output data. And putting the sample data into the distribution simulation platform again to obtain the modified sample data. In this case, the tower model including the strict mechanism model has more than 280 input data, more than 3700 normal pressure output data, more than 2500 reduced pressure output data, more than 800 heat exchange network input data, more than 3500 output data, and 10000 instrument readings in total for a complete sample data, and the enhanced order of magnitude is 109. The convergence rate of the model is about 90%, and the running time of 100 computers can be completed in about 7 days.
In summary, in the embodiment, a strict mechanism model is used to simulate a steady-state working condition obtained based on historical production data, part of variable input data is generated in batches at random, another part of variable input data is obtained through a relational expression between variables and a device simplified model, the strict mechanism model is used to perform step-by-step calculation after the input data is calculated in batches to obtain batch output data, a large amount of sample data is generated quickly, sample expansion is further achieved, and samples in a sparse area are supplemented at the same time, so that the amplified sample data is distributed uniformly. Therefore, a large number of samples can be generated quickly, for example, thousands of steady-state working conditions obtained based on historical production data can be expanded to hundreds of thousands of steady-state working conditions, the problems that simulation results obtained in the sample generation process are not easy to converge and the sample obtaining rate is low are solved, and the sample convergence rate after enhancement can reach 90%. In addition, the embodiment provides an operation resource architecture for large-scale sample generation operation, so that the existing computational power can be fully utilized, the generation time of the sample is reduced, the operation efficiency is improved, and the acquisition and management of the operation result are facilitated. For example, approximately 20000 samples are available for 24 hours for one hundred computers.
Example two
Referring to fig. 2, the present embodiment provides a model training method, including:
training a device model for assisting production equipment optimization adjustment based on a plurality of sample data obtained by sample enhancement;
and optimizing and adjusting part or all operation corresponding variables in the production equipment according to the device model so as to enable the production equipment to be in the optimal working state.
In specific implementation, the AI model is trained based on the enhanced sample data so as to optimize the operation variables in the AI model and ensure that the production equipment is in the optimal operation state.
Compared with the prior art, the beneficial effects of the model training method provided by the embodiment of the invention are the same as those of the sample enhancement method provided by the embodiment, and are not repeated herein.
EXAMPLE III
The present embodiment provides a sample enhancement system, including:
the dividing unit is used for dividing the variables into independent variables, partial associated variables and residual associated variables;
the distribution reading unit is used for acquiring a historical data distribution range corresponding to each variable according to historical data corresponding to each variable in historical production data;
the screening unit is used for randomly generating a plurality of random data in the distribution range corresponding to each independent variable, calculating intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, then adopting a simplified model which is trained in advance and taking the random data and the intermediate data as input calculation and judging whether the calculation result is converged, and acquiring residual data corresponding to the residual associated variables when the calculation result is converged;
the data integration unit is used for sorting and combining the random data, the intermediate data and the residual data into input data used for subsequent simulation;
the device comprises a clustering and sequencing unit, a data processing unit and a data processing unit, wherein the clustering and sequencing unit is used for clustering input data by adopting Euclidean distance to obtain a plurality of groups of sub-sample sets, each group of sub-sample set comprises a plurality of samples, and the samples in each group of sub-sample sets are sequenced;
the simulation unit is used for sequentially carrying out step-by-step simulation calculation on the samples in each sub-sample set through a strict mechanism model which is trained in advance to obtain output data of the samples;
the summarizing unit is used for summarizing and combining the input data and the corresponding output data of each sample to obtain a plurality of complete sample data obtained by sample enhancement;
and the sample supplementing unit is used for performing visualization analysis on the distribution range of the acquired multiple sample data and supplementing the samples in the sparse region.
Compared with the prior art, the beneficial effects of the sample enhancement system provided by the embodiment of the invention are the same as those of the sample enhancement method provided by the embodiment, and are not repeated herein.
Example four
The present embodiment provides a model training system, including:
the training unit is used for training a device model for assisting the optimization and adjustment of the production equipment based on a plurality of sample data obtained by sample enhancement;
and the debugging unit is used for optimizing and adjusting the variables corresponding to part or all of the operations in the production equipment according to the device model so as to enable the production equipment to be in the optimal working state.
Compared with the prior art, the beneficial effects of the model training system provided by the embodiment of the invention are the same as those of the model training method provided by the embodiment, and are not repeated herein.
EXAMPLE five
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the above-described sample enhancement method, and/or steps of a model training method.
Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by this embodiment are the same as the beneficial effects of the sample enhancement method and/or the model training method provided by the above technical solutions, and are not described herein again.
It will be understood by those skilled in the art that all or part of the steps in the method for implementing the invention may be implemented by hardware instructions related to a program, the program may be stored in a computer-readable storage medium, and when executed, the program includes the steps of the method of the embodiment, and the storage medium may be: ROM/RAM, magnetic disks, optical disks, memory cards, and the like.
The above description is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present invention, and the changes or substitutions should be covered within the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (10)
1. A method of sample enhancement, comprising:
dividing variables into independent variables, partial associated variables and residual associated variables;
acquiring a historical data distribution range corresponding to each variable according to historical data corresponding to each variable in historical production data;
randomly generating a plurality of random data in a distribution range corresponding to each independent variable, calculating intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, calculating by taking the random data and the intermediate data as input by adopting a simplified model which is trained in advance, judging whether a calculation result is converged, and acquiring residual data corresponding to residual associated variables when the calculation result is converged;
the random data, the intermediate data and the residual data are collated and combined into input data used for subsequent simulation;
clustering input data by adopting Euclidean distance to obtain a plurality of groups of sub-sample sets, wherein the sub-sample sets comprise a plurality of samples, and sequencing the samples in each group of the sub-sample sets;
performing step-by-step simulation calculation on each sample in the sub-sample set according to the sequence through a strict mechanism model which is trained in advance to obtain output data of the sample;
summarizing and combining input data and corresponding output data of each sample to obtain a plurality of complete sample data obtained by sample enhancement;
and performing visualization analysis on the distribution range of the obtained multiple sample data, and supplementing the samples in the sparse region.
2. The method of claim 1, wherein the step of obtaining the historical data distribution range corresponding to each variable according to the historical data corresponding to each variable in the historical production data comprises:
and counting the upper limit and the lower limit of the historical data corresponding to each variable in the historical production data, and taking the range of the upper limit and the lower limit as the historical data distribution range of the corresponding variable.
3. The method of claim 1, wherein randomly generating a plurality of random data within the distribution range corresponding to each independent variable comprises:
and randomly generating a plurality of random data based on the uniform read data distribution range corresponding to each independent variable.
4. The method of claim 3, wherein the step of calculating the intermediate data corresponding to the partially associated variable from the random data corresponding to the independent variable comprises:
analyzing the data correlation between partial associated variables and independent variables in the historical production data by adopting a correlation analysis method based on random data corresponding to each independent variable, and obtaining pairwise mutual influence relation of the partial associated variables and the independent variables to obtain a relational expression of the partial associated variables and the independent variables;
according to the relational expression, taking the random data of the independent variable as input to calculate the intermediate data of the related variable of the corresponding part;
and merging the intermediate data of the partial associated variables belonging to the same group and the random data of the independent variables to form the input data of the same group.
5. The method of claim 4, wherein the method of calculating and judging whether the calculation result converges by using the pre-trained simplified model with random data and intermediate data as input, and obtaining the residual data corresponding to the residual associated variables during convergence comprises:
simulating input data comprising random data and intermediate data by adopting a pre-trained simplified model, if the simulation result is convergence, considering the group of input data as valid sample data, and using the simulation result as residual data corresponding to residual associated variables, and if the simulation result is non-convergence, considering the group of input data as invalid sample data, and removing the input data;
and combining the residual data corresponding to the residual associated variables belonging to the same group with the input data of the partial associated variables and the independent variables to jointly serve as the input data of the same group of the strict mechanism model.
6. The method of claim 5, wherein the step-by-step simulation calculation of the samples in each subset sequentially through a pre-trained rigorous mechanism model to obtain the output data comprises:
distributing the sub-sample sets to different simulation nodes, calculating input data by adopting the same pre-trained strict mechanism model of each simulation node, and calculating sample input data and judging whether an output result is converged by each simulation node according to the sample sequence in the received sample set;
if the convergence is detected, saving the output data in a first convergence sample, if the convergence is not detected, modifying and adjusting corresponding input data, inputting the modified and adjusted input data into a strict mechanism model again, and saving the converged output data in a second convergence sample;
combining the first convergence sample and the second convergence sample to obtain output data corresponding to the remaining meters;
wherein, if convergence, saving the output data after the first convergence sample further comprises: and performing energy balance verification on the sample data, if the verification result is energy balance, confirming convergence, and saving the output data in the first convergence sample, and if the verification result is energy imbalance, judging the output data to be not converged.
7. The method of claim 6, wherein the step-by-step simulation of a sample based on a pre-trained rigorous mechanistic model comprises:
dividing all variables in input data into a first class, a second class and a third class, wherein each class at least comprises one variable; the simulation calculation is divided into two major steps, wherein the first major step comprises data replacement and calculation of first-class and second-class variables, and the second major step comprises data replacement and calculation of third-class variables;
adopting a working condition which is successfully simulated through a strict mechanism model in advance as a basic working condition, wherein the basic working condition comprises basic input data;
correspondingly dividing variables in each group of input data and basic input data into a first class, a second class and a third class, and taking sample data in the input data as a target value;
replacing data corresponding to the first type of variable in the basic input data with a corresponding target value in the input data;
meanwhile, data corresponding to the second type of variables in the basic input data are replaced according to rules, if the current target value does not affect convergence relative to the change direction of the basic data, the basic input data are replaced once according to the current target value, otherwise, the second type of variables need to be replaced step by step, the basic input data are gradually adjusted to the target value direction according to the preset step length, once the second type of variables are replaced, analog calculation is carried out, each replacement is carried out on the basis of the previous calculation, and the first large-step calculation is stopped until all the second type of variables reach the preset constraint condition or target value or the simulation does not converge;
and taking the data after the first large-step calculation as a basis, taking the sample data corresponding to the third type variable in the input data as a target value, taking the data corresponding to the basic input data and the third type variable as a basis, gradually adjusting and replacing according to a preset step length, inputting the data into a strict mechanism model for calculation, and stopping the second large-step simulation calculation until all the data corresponding to the third type variable reach a preset constraint condition or target value or the simulation is not converged.
8. A method of model training, comprising:
training a device model for assisting production equipment optimization adjustment based on a plurality of sample data obtained by sample enhancement;
and optimizing and adjusting variables corresponding to part or all of operations in the production equipment according to the device model so as to enable the production equipment to be in the optimal working state.
9. A sample enhancement system, comprising:
the dividing unit is used for dividing the variables into independent variables, partial associated variables and residual associated variables;
the distribution reading unit is used for acquiring a historical data distribution range corresponding to each variable according to historical data corresponding to each variable in historical production data;
the screening unit is used for randomly generating a plurality of random data in the distribution range corresponding to each independent variable, calculating intermediate data corresponding to partial associated variables according to the random data corresponding to the independent variables, then adopting a simplified model which is trained in advance and taking the random data and the intermediate data as input calculation and judging whether the calculation result is converged, and acquiring residual data corresponding to the residual associated variables when the calculation result is converged;
the data integration unit is used for sorting and combining the random data, the intermediate data and the residual data into input data used for subsequent simulation;
the device comprises a clustering and sequencing unit, a data processing unit and a data processing unit, wherein the clustering and sequencing unit is used for clustering input data by adopting Euclidean distance to obtain a plurality of groups of sub-sample sets, each group of sub-sample set comprises a plurality of samples, and the samples in each group of sub-sample sets are sequenced;
the simulation unit is used for sequentially carrying out step-by-step simulation calculation on the samples in each sub-sample set through a strict mechanism model which is trained in advance to obtain output data of the samples;
the summarizing unit is used for summarizing and combining the input data and the corresponding output data of each sample to obtain a plurality of complete sample data obtained by sample enhancement;
and the sample supplementing unit is used for performing visualization analysis on the distribution range of the acquired multiple sample data and supplementing the samples in the sparse region.
10. A model training system, comprising:
the training unit is used for training a device model for assisting the optimization and adjustment of the production equipment based on a plurality of sample data obtained by sample enhancement;
and the debugging unit is used for optimizing and adjusting the variables corresponding to part or all of the operations in the production equipment according to the device model so as to enable the production equipment to be in the optimal working state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110646519.7A CN113420799A (en) | 2021-06-10 | 2021-06-10 | Sample enhancement method, model training method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110646519.7A CN113420799A (en) | 2021-06-10 | 2021-06-10 | Sample enhancement method, model training method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113420799A true CN113420799A (en) | 2021-09-21 |
Family
ID=77788344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110646519.7A Pending CN113420799A (en) | 2021-06-10 | 2021-06-10 | Sample enhancement method, model training method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113420799A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109814513A (en) * | 2019-03-20 | 2019-05-28 | 杭州辛孚能源科技有限公司 | A kind of catalytic cracking unit optimization method based on data model |
| CN111027733A (en) * | 2018-10-10 | 2020-04-17 | 中国石油化工股份有限公司 | Petrochemical device product yield optimization method based on big data technology |
-
2021
- 2021-06-10 CN CN202110646519.7A patent/CN113420799A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027733A (en) * | 2018-10-10 | 2020-04-17 | 中国石油化工股份有限公司 | Petrochemical device product yield optimization method based on big data technology |
| CN109814513A (en) * | 2019-03-20 | 2019-05-28 | 杭州辛孚能源科技有限公司 | A kind of catalytic cracking unit optimization method based on data model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107169628B (en) | Power distribution network reliability assessment method based on big data mutual information attribute reduction | |
| CN104798043B (en) | A kind of data processing method and computer system | |
| CN109992921B (en) | On-line soft measurement method and system for thermal efficiency of boiler of coal-fired power plant | |
| JP2020500420A (en) | Yield prediction system and method for machine learning based semiconductor manufacturing | |
| CN109214708B (en) | Electric power system risk assessment method based on cross entropy theory optimization support vector machine | |
| CN104572449A (en) | Automatic test method based on case library | |
| CN107360026B (en) | Distributed message middleware performance prediction and modeling method | |
| JP2019537079A (en) | How to build stochastic models for large-scale renewable energy data | |
| CN106980906B (en) | Spark-based Ftrl voltage prediction method | |
| CN109981749A (en) | A kind of cloud workflow task running time prediction method promoted based on limit gradient | |
| CN112307677A (en) | Power grid oscillation mode evaluation and safety active early warning method based on deep learning | |
| CN111324635A (en) | Industrial big data cloud platform data processing method and system | |
| CN104732067A (en) | Industrial process modeling forecasting method oriented at flow object | |
| Su | Analysis of optimisation method for online education data mining based on big data assessment technology | |
| Tornede et al. | Coevolution of remaining useful lifetime estimation pipelines for automated predictive maintenance | |
| CN115185804A (en) | Server performance prediction method, system, terminal and storage medium | |
| Li et al. | Aero-engine exhaust gas temperature prediction based on LightGBM optimized by improved bat algorithm | |
| CN110837952A (en) | Game theory-based power grid new technology equipment selection method and system | |
| CN109116300B (en) | Extreme learning positioning method based on insufficient fingerprint information | |
| CN114548494A (en) | Visual cost data prediction intelligent analysis system | |
| CN104573331B (en) | A kind of k nearest neighbor data predication method based on MapReduce | |
| CN113032367A (en) | Dynamic load scene-oriented cross-layer configuration parameter collaborative tuning method and system for big data system | |
| CN109886316B (en) | Transformer state parameter combination prediction method based on cloud system similarity weight distribution | |
| CN110927478B (en) | Method and system for determining state of transformer equipment of power system | |
| CN103136440A (en) | Method and device of data processing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20211208 Address after: 6 / F, block AB, ocean Guanghua international, No. 10, Jintong West Road, Chaoyang District, Beijing 100020 Applicant after: Beijing Yineng Gaoke Technology Co.,Ltd. Applicant after: BEIJING SCIENCO TECHNOLOGY CO.,LTD. Address before: 6 / F, block AB, ocean Guanghua international, No. 10, Jintong West Road, Chaoyang District, Beijing 100020 Applicant before: Beijing Yineng Gaoke Technology Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210921 |
|
| RJ01 | Rejection of invention patent application after publication |