CLIAMS:1. A system for predicting full assay of an oil consignment, said full assay includes fast measurable properties and regular measurable properties of oil consignments, said system comprising:
i. a repository adapted to store historical data of full assay including fast measurable properties and regular measurable properties of previous oil consignments;
ii. a first input and storage module adapted to accept and temporarily store values for fast measurable properties of a sample of a current oil consignment;
iii. an integrator configured to cooperate with the first input and storage module and the repository, and further configured to integrate the temporarily stored fast measurable properties with the stored historical full assay data to obtain integrated data; and
iv. a predictive model generator configured to cooperate with the integrator to receive the integrated data, and further configured to dynamically identify the key fast measurable properties that are relevant for predicting target full assay properties, and also configured to generate an ensemble of predictive models for predicting target full assay properties as a function of the dynamically identified key fast measurable properties by iteratively processing the integrated data.
2. A system for predicting full assay of an oil consignment, said full assay includes fast measurable properties and regular measurable properties of oil consignments, said system comprising:
i. a repository adapted to store historical data of full assay including fast measurable properties and regular measurable properties of previous oil consignments;
ii. a first input and storage module adapted to accept and temporarily store values for fast measurable properties of a sample of a current oil consignment;
iii. an integrator configured to cooperate with the first input and storage module and the repository, and further configured to integrate the temporarily stored fast measurable properties with the stored historical full assay data to obtain integrated data;
iv. a predictive model generator configured to cooperate with the integrator to receive the integrated data, and further configured to dynamically identify the key fast measurable properties that are relevant for predicting target full assay properties, and also configured to generate an ensemble of predictive models for predicting target full assay properties as a function of the dynamically identified key fast measurable properties by iteratively processing the integrated data; and
v. a statistical graphical model generator adapted to receive and process the ensemble of predictive models and the identified key fast measurable properties using statistical methods to generate statistical relationship graphs that provide a predicted full assay and uncertainty bounds of prediction for the current oil consignment.
3. The system as claimed in claim 1 and claim 2, wherein said system further comprises a user interface configured to receive the temporarily stored fast measurable properties of current oil consignment and the statistical relationship graphs, and further configured to publish the statistical relationship graphs providing models of predicted full assay for the current oil consignment.
4. The system as claimed in claim 1 and claim 2, wherein said system further comprises a tuner adapted to receive the full assay of previous oil consignments from the repository and the published predicted full assay models for the current oil consignment from the user interface, and derive deviation in the predicted full assay.
5. The system as claimed in claim 1 and claim 2, wherein said system further comprises a re-calibrator adapted to receive the derived deviation and update the published predicted full assay models by re-calibrating the predictive model generator and the statistical graphical model generator.
6. The system as claimed in claim 5, wherein said re-calibrator re-calibrates the predictive model generator and the statistical graphical model generator based on a pre-defined threshold value.
7. The system as claimed in claim 1 and claim 2, wherein said predictive model generator comprises:
i. a data pre-processing module adapted to examine the integrated data, and, discard, impute or predict, missing or aberrational values amongst said fast measurable properties to obtain pre-processed data;
ii. a scaling module adapted to scale the pre-processed data based on pre-determined requirement;
iii. a data summarization and regularization module adapted to manage large sets of fast measurable properties having overlapping values, and, identify, for each of the targeted full assay property, a different subset of fast measurable properties that are fundamental in predicting its values; and
iv. an ensemble modeler adapted to form an ensemble of predictive models for each of the full assay properties, and to automatically sub-select relevant machine learning techniques for corresponding full assay property.
8. The system as claimed in claim 1 and claim 2, wherein the first input and storage module is further adapted to select the fast measurable properties from the group consisting of density, specific gravity, API gravity, total acid number, total sulphur, mercaptan sulphur, viscosity, NIR Spectra.
9. The system as claimed in claim 2, wherein said statistical relationship graphs are constructed based on at least one of domain knowledge and ensemble of predictive models.
10. The system as claimed in claim 5, wherein the re-calibrator employs at least one of time triggered model tuning mechanism and event triggered method tuning mechanism.
11. A method for predicting full assay of an oil consignment wherein said full assay includes fast measurable properties and regular measurable properties of oil consignments, said method including the following:
• storing historical data of full assays including fast measurable properties and regular measurable properties of previous oil consignments;
• accepting and temporarily storing values for fast measurable properties of a sample of a current oil consignment;
• integrating temporarily stored fast measurable properties and obtaining integrated data of individual fast measurable properties; and
• receiving the integrated data and stored historical data and, dynamically identifying the key fast measurable properties that are relevant for predicting the target full assay properties, and generating an ensemble of predictive models for predicting the target full assay properties as a function of the dynamically identified key fast measurable properties by iteratively processing the integrated data.
12. A method for predicting full assay of an oil consignment wherein said full assay includes fast measurable properties and regular measurable properties of oil consignments, said method including the following:
• storing historical data of full assays including fast measurable properties and regular measurable properties of previous oil consignments;
• accepting and temporarily storing values for fast measurable properties of a sample of a current oil consignment; and
• integrating temporarily stored fast measurable properties and obtaining integrated data of individual fast measurable properties;
• receiving the integrated data and stored historical data and, dynamically identifying the key fast measurable properties that are relevant for predicting the target full assay properties, and generating an ensemble of predictive models for predicting the target full assay properties as a function of the dynamically identified key fast measurable properties by iteratively processing the integrated data; and
• receiving the ensemble of predictive models and processing the ensemble of predictive models using statistical methods for generating statistical relationship graphs that provide a predicted full assay for the current oil consignment and estimating uncertainty in the predicted full assay.
13. The method as claimed in claim 11 and claim 12, wherein said method further comprises the following:
• receiving and publishing the statistical relationship graphs providing models of predicted full assay for the current oil consignment;
• receiving the stored full assay of previous oil consignments and the published predicted full assay models and, deriving deviation in the predicted full assay; and
• receiving the derived deviation and updating the published full assay models by re-calibration.
14. The method as claimed in claim 13, wherein said step of re-calibration includes re-calibrating based on a pre-defined threshold value and employing at least one of time triggered model tuning mechanism and event triggered method tuning mechanism..
15. The method as claimed in claim 11 and claim 12 , wherein said step of generating ensemble of predictive models further comprises steps of:
• pre-processing the integrated data and discarding, imputing or predicting, missing or aberrational values amongst the fast measurable properties to obtain pre-processed data;
• scaling the pre-processed data based on pre-determined requirement;
• summarizing and regularizing the scaled pre-processed data to manage large sets of fast measurable properties having overlapping values and identifying for each of the targeted full assay property, a different subset of fast measurable properties that are fundamental in predicting its values; and
• forming an ensemble of predictive models for each of the full assay properties, and automatically sub-selecting relevant machine learning techniques for corresponding full assay property.
16. The method as claimed in claim 11 and claim 12, wherein the step of accepting and storing values for fast measurable properties of a sample of current oil consignment further involves selecting the properties from the group consisting of density, specific gravity, API gravity, total acid number, total sulphur, mercaptan sulphur, viscosity, NIR Spectra, NMR, FTIR.
17. The method as claimed in claim 12, wherein the step of generating statistical relationship graphs further includes utilizing at least one of domain knowledge and ensemble of predictive models. ,TagSPECI:FIELD OF DISCLOSURE
The present disclosure relates to prediction of unknown properties of the mixture in liquid state in short time period. The present invention is particularly related to the prediction of full assay of the crude and purified oil blend.
DEFINITIONS OF THE TERMS USED IN THE SPECIFICATION
The expression ‘fast measurable property value’ used hereinafter in this specification refers to values obtained after performing quick experiments on oil consignments.
The expression ‘key fast measurable properties’ used hereinafter in this specification refers to the properties of oil that are essential in determining full assay and are measured by performing quick experiments.
The expression ‘oil’ used hereinafter in this specification refers to crude petroleum oil and purified petroleum oil blend.
These definitions are in addition to those expressed in the art.
BACKGROUND
The value of oil is determined by its assay, i.e. properties of oil and its various distillation cuts. Oil assay is a characterization of the chemical composition of the oil and plays a pivotal role in refinery planning and scheduling. In refinery planning, oil assay is a main driver for its valuations, indicating whether or not profitable products could be obtained post-refining. In refinery scheduling, oil assay is an important driver for blending oils and setting the operating parameters. It is well-known that the oil assay of oils changes over time even though it is mined from same origin. The oil assay is typically determined using chemical laboratory tests. Many of these chemical tests are time consuming and only some tests can be done in quick time. However, these quick tests can only determine a short assay (or inspection assay), which gives a high-level overview of oil characteristics. Measuring the full assay (or comprehensive assay), which gives more accurate characterization of the oil, is time-consuming and hence is intractable to-do for every oil shipment of a refinery. Nonetheless, in order to improve the oil valuation process and decision-making in manufacturing process, it is necessary to measure the full assay and maintain an up-to-date full assay database. Since it is not tractable to measure full assay for every incoming oil shipment, usually, only inspectional or short assay is measured to track oil shipments and oil unit feedstock. As a result, quality variations beyond short assay are not captured. This often leads to sub-optimal planning and/or process parameter setting. Recent advances in analytical techniques can quickly provide partial information on oil using spectroscopy, chromatography coupled with suitable detectors (e.g. flame ionization, mass spectrometry etc.). However this information only provides overview of the actual oil properties which is a short assay. There are few computational methods which provide full assay for oil samples. But, a high error factor is observed in them that deviates predicted full assay from the actual properties of the oil.
Therefore, in order to limit the aforementioned drawbacks, there is a need for a system that predicts full assay of an oil consignment and provides measure of uncertainty in the predicted value.
OBJECTS
An object of the system of the present disclosure is to provide a system that estimates full assay for every oil consignment.
Another object of the system of the present disclosure is to provide a system that predicts full assay by utilizing results obtained from conventional experiments which characterize fast measurable properties of oil and historical full assay data of previous oil consignments.
Still another object of the system of the present disclosure is to provide a system that automatically generates relationship graphs among historical full assay properties and fast measurable properties of sample oil and quantifies full assay of sample oil based on the generated graphs.
One more object of the system of the present disclosure is to provide a system that automatically re-calibrates and adapts to changing patterns in full assay.
Yet another object of the system of the present disclosure is to provide a system for prediction of full assay of purified oil blend and blend of an oil consignment in short period.
An additional object of the system of the present disclosure is to provide a system for prediction of highly accurate full assay along with the uncertainty bound of the prediction.
Other objects and advantages of the present disclosure will be more apparent from the following description when read in conjunction with the accompanying figures, which are not intended to limit the scope of the present disclosure.
SUMMARY
The present disclosure envisages a system for predicting full assay of an oil consignment. The oil consignment includes at least one of oil consignment and blended oil consignment and the full assay includes fast measurable properties and regular measurable properties of oil consignments, wherein, regular measurable properties include properties of the oil consignments that take significantly longer time to be determined and fast measurable properties include properties of the oil consignments that can be determined in a significantly shorter period of time than the regular measurable properties.
Typically, in accordance with the present disclosure, the system comprises a repository to store historical data of full assay that includes fast measurable properties and regular measurable properties of previous oil consignments. A first input and storage module present in the system accepts and temporarily stores values for fast measurable properties of a sample of a current oil consignment in order to predict the full assay of that sample. An integrator present in the system co-operates with the first input and storage module to integrate the temporarily stored fast measurable properties to obtain integrated data of individual fast measurable properties. This integrated data is fed to a predictive model generator that dynamically identifies the key fast measurable properties that are relevant for predicting targeted full assay properties, and, generates an ensemble of predictive models for predicting targeted full assay properties as a function of the dynamically identified key fast measurable properties by iteratively processing the integrated data.
This summary is provided to introduce concepts related to prediction of full assay of an oil consignment, which is further described below in the detailed description. This summary is neither intended to identify essential features of the present disclosure nor is it intended for use in determining or limiting the scope of the present disclosure.
BRIEF DESCRIPTION OF ACCOMPANYING DRAWINGS
The system of the present disclosure will now be described with the help of the accompanying drawings, in which:
FIGURE 1 illustrates schematic of a system for predicting full assay of oil consignments.
FIGURE 2 illustrates the steps are followed for accurate prediction of the values of full assay properties from the fast measurable properties.

DETAILED DESCRIPTION OF THE ACCOMPANYING DRAWINGS
The system of the present disclosure will now be described with reference to the embodiment shown in the accompanying drawing. The embodiment does not limit the scope and ambit of the disclosure. The description relates purely to the examples and preferred embodiments of the disclosed system and its suggested applications.
The system herein and the various features and advantageous details thereof are explained with reference to the non-limiting embodiments in the following description. Descriptions of well-known parameters and processing techniques are omitted so as to not unnecessarily obscure the embodiment herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiment herein may be practiced and to further enable those of skill in the art to practice the embodiment herein. Accordingly, the examples should not be construed as limiting the scope of the embodiment herein.
In accordance with the present disclosure, the proposed system predicts full assay of an oil consignment by exploiting latent information in fast measurable properties from disparate experiments on oil. Main components of the system of the present disclosure are a predictive model generator, a statistical graphical model generator and a re-calibrator.
Referring to the accompanying drawings, FIGURE 1 illustrates a system 100 for predicting full assay of oil consignments. The oil consignments include at least one of oil consignment and blended oil consignment. The system 100 uses as input, fast measurable properties from disparate experiments (e.g., NIR Spectra, NMR, FTIR, Short Assay) and historical data on full assay for a variety of oil consignments. The fast measurable properties are accepted as input through a first input and storage module 104 that allows a user to input various the fast measurable property values of a sample of a new oil consignment and temporarily stores these values. The historical data of full assay of previous oil consignments and fast measurable properties is stored in a repository 102. Using this historical data of previous oil consignments and the fast measurable properties of a new oil consignment, the system 100 builds and publishes predictive models that are subsequently used, through a user interface 116, for predicting full assay of new oil consignments.
The system 100 comprises an integrator 106 that integrates data from different experiments available from the first input and storage module 104 and the historical data on full assay available from the repository 102 to obtain integrated data. The integrated data is then passed to a predictive model generator 108 that iteratively processes the received integrated data and develops an ensemble of predictive models for each of the full assay property. Using a variety of machine learning techniques, the predictive model generator 108 first builds predictive models for full assay by using only the values of fast measurable properties as input. The values of full assay properties with accurate predictive models are then added to the values of fast measurable properties and the machine learning techniques are again used to build more accurate models of the previous inaccurately predicted values of full assay properties. This iterative process is repeated till sufficiently accurate predictive models are obtained for all properties of the full assay.
Referring to FIGURE 1 and FIGURE 2, FIGURE 2 illustrates the steps carried out by the predictive model generator 108 for accurate prediction of the values of full assay properties from the fast measurable properties. The predictive model generator 108 processes the fast measurable properties and full assay in the historical data to obtain ensemble of predictive models. These fast measurable properties typically consist of numerous variables including absorptions at various frequencies of NIR spectra, short assay properties like API, Sulphur, and TAN. Some of these variables may not be required for predicting the values of the targeted full assay property, i.e. majority of the collected variables may be irrelevant for the target full assay property and may represent noise. It is important to differentiate automatically between the variables that matter and those that do not contribute to the prediction accuracy of the target full assay property. In one embodiment, the system 100 automatically checks if the experiments conducted to acquire the values of fast measurable properties are necessary in prediction of the targeted full assay property.
The steps followed for accurate prediction of the values of full assay properties from the fast measurable properties are illustrated in FIGURE 2. In step 202, historical data associated with full assay is stored, which includes fast measurable properties and regular measurable properties of previous oil consignment. In step 204, fast measurable properties associated with current sample of oil consignment are accepted and stored temporarily. These temporarily stored fast measurable properties are integrated to obtain integrated data of individual fast measurable properties as denoted in step 206. In step 208, the predictive model generator 108 receives integrated data from the integrator 106 and also receives the stored historical data to dynamically identify the key fast measurable properties. An ensemble of predictive models for predicting the target full assay properties is then generated in step 208 as a function of the dynamically identified key fast measurable properties by iteratively processing the integrated data. For each of the full assay properties as target, a data pre-processing module (not shown in the figure), pre-processes the received integrated data by removing outliers from the fast measurable properties and/or the target full assay property. Statistical techniques and domain knowledge based outlier detection methods are typically employed to detect the samples which are irrelevant. In case of missing data in fast measurable properties or full assay property, various imputations are performed to obtain the missing values. In one embodiment, the data pre-processing module deletes a complete sample (i.e., both fast measurable properties and the target full assay property) of missing values. In another embodiment, the missing values can be imputed based on sufficient statistics (for example, mean, median etc.). The data pre-processing module (not shown in the figure) predicts the missing value based on inter and intra correlation amongst the attributes of fast measurable properties and full assay property. The pre-processed data is then scaled by a scaling module (not shown in the figure) as per the requirements of learning techniques used. In one embodiment, a mean centered data is used for scaling and in another embodiment scaling of the data is based on a standard normal variation. Usually, handling large set of fast measurable properties is a challenge in predicting full assay as a non-informative property value with respect to one full assay property could be an informative property value with respect to another full assay property. Also, multi-collinearity of the fast measurable properties can occur for properties within an experiment or across experiments.
The system 100 of the present disclosure resolves the issue of handling large set of fast measurable properties by handling overlapping information at multiple stages. In one embodiment, the method invokes data summarization techniques such as principal component analysis and partial least squares regression analysis for handling the same. In another embodiment, the system 100 invokes regularization techniques such as LASSO, RIDGE, RELAXO, Elastic net procedures and the like. The predictive model generator 108 also includes a data summarization and regularization module (not shown in the figure) to combine regularization and data summarization techniques for effective identification of the key fast measurable properties for the target full assay property. These key properties are dynamically identified by detecting whether a set of experiments is necessary to predict the target full assay property. The key fast measurable properties for the target full assay property prediction are deduced at step 208 and taken to the next step of model building. On determination of the key fast measurable properties, predictive models are built and validated. In one embodiment, the system 100 uses, leave-one-out cross validation technique for building efficient models. This step also provides information on those set of oils for which predicting the target full assay property is difficult within the specified accuracy limits using the fast measurable properties. Additionally, modeling of the full assay properties with the key fast measurable properties uses at least one of linear and non-linear learning techniques including principal component regression, partial least squares, ridge regression, elastic-net, artificial neural networks, support vector regression, random forests and the like to obtain predictive models. All of the learning techniques may not be applicable in modeling a target full assay property. Therefore, the predictive model generator 108 includes a step of down-selecting the best possible learning techniques to obtain best possible set of predictive models for that target full assay property. In one embodiment, the system 100 of the present disclosure uses automated rule engine to down-select the best possible learning techniques. The rule engine takes into account the variance and bias of the estimator (obtained from learning techniques), before deciding to use the estimator for predicting the value of a full assay property. In one embodiment, the rule engine chooses only those estimators whose bias is within a predefined threshold and whose sample variance (based on leave-one-out cross validation) is within a threshold. In another embodiment, the rule engine may choose those methods which satisfy the quantile statistics obtained using boxplot.
Ideally, the predicted full assay property value must be unbiased and have minimum variance. To achieve this, the predictive model generator 108 combines multiple linear and nonlinear learning techniques to obtain an ensemble of predictive models at step 208. This ensemble of predictive models is unbiased and has variance lesser than individual methods and hence efficient in predicting oil and blended oil properties. This step solves the prediction accuracy issues and also handles limitations in availability of data samples. The predictive models from the ensemble of predictive models are then checked for prediction quality at step 210 in order to confirm if the quality of predicted models is adequate for validation data. In this step 210, the ensemble of predictive models is processed to provide predicted full assay and to estimate uncertainty in the predicted full assay. If the prediction quality is inadequate, those full assay properties that are predicted with adequate accuracy are added to the set of fast measurable properties and all the steps of data pre-processing are repeated. This iterative process ensures efficiency in prediction of values of full assay properties.
The predictive models are then received by a statistical graphical model generator 110. The predictive models for full assay properties include information about the key fast measurable properties that are essential for predicting the full assay property values. These key fast measurable properties could be the original fast measurable from the quick experiments or could be the full assay properties with accurate predictive models in the first rounds of iteration in the predictive model generator 108. The statistical graphical model generator 110 extracts relationships between fast measurable properties and full assay. It also captures conditional dependence structure between fast measurable properties and full assay properties, and within full assay properties. These dependencies are then encoded by generating statistical relationship graphs. The statistical relationship graphs, along with the ensemble of predictive models, are then used for estimating the uncertainty in predicted values of full assay properties.
In one embodiment, the system 100 uses Bayesian Network (BN) as a graphical model to capture intra-dependencies and inter-dependencies within fast measurable properties and full assay properties. To enhance relationship graph between the properties, the statistical graphical model generator 110 generates graphs based on at least one of domain knowledge and the ensemble of predictive models from predictive model generator 108. For each full assay property and the fast measurable properties the uncertainty in property value is characterized by appropriate probability distributions. The parameters of the probability distribution are estimated from domain knowledge, ensemble of predictive models, and historical data. The probability distribution uncovers and handles uncertainty in the prediction of the values of full assay properties. The statistical relationship graphs and the characterized uncertainty representation are used for predicting the values of full assay for the sample of a new oil consignment. In step 212 the system 100 publishes the statistical relationship graphs providing models of predicted full assay for the new oil consignment by using the user interface 116 that receives the temporarily stored fast measurable properties from the first input and storage module 104 and the statistical relationship graphs from the statistical graphical model generator 110.
For the new oil consignment, the system 100 of the present disclosure predicts full assay despite unavailability of the full set of fast measurable properties. In one embodiment, the system 100 uses Gaussian distribution with mean given by the domain knowledge or by the ensemble of predictive models and the variance estimated from historical data. In another embodiment, Markov Chain Monte Carlo (MCMC) methods including Gibbs Sampling are used to predict values of full assay properties and their probability distributions when data from only a few fast measurable properties is available.
In step 214, the stored full assay of previous oil consignment and the published predicted full assay model are received and deviation in the predicted full assay is derives.
In step 216, the published full assay models are updated using re-calibration. The predictive model generator 108 and the statistical graphical model generator 110 may need to be re-calibrated as and when new data related to a new oil consignment is available. For re-calibration, the new dataset must contain both fast measurable properties and the corresponding values of full assay properties. The system 100 of the present disclosure uses time triggered model tuning mechanism or event triggered model tuning mechanism for re-tuning. These mechanisms can be user configurable. The system 100 re-calibrates the generated models to incorporate newly available data on historical full assays and their corresponding fast measurable properties. This re-calibration is possible with the help of a tuner 112 and a re-calibrator 114. During re-tuning, the system 100 automatically adapts to new patterns in the full assay by adding new features to the ensemble of predictive model and by re-calibrating the parameters of the ensemble of predictive model. To achieve this the tuner 112 receives from the repository 102 the values of properties of full assay of previous oil consignments and the published predicted full assay model from the user interface 116, and derives deviation values in the predicted full assay. The re-calibrator 114 receives the derived deviation and updates the published predicted full assay models by re-calibrating the predictive model generator 108 and the statistical graphical model generator 110 based on a pre-defined threshold on the deviates or based on a user input.
TECHNICAL ADVANCEMENTS
A system and method for lightweight authentication on datagram transport in accordance with the present disclosure described herein above has several technical advancements including but not limited to the realization of:
• a system that estimates full assay for every oil consignment;
• a system that predicts full assay by utilizing results obtained from conventional experiments which characterize fast measurable properties of oil and historical full assay data of previous oil consignments;
• a system that automatically selects the key properties relevant for prediction of a full assay property;
• a system that automatically generates relationship graphs among historical full assay properties and fast measurable properties of sample oil and quantifies full assay of sample oil based on the generated graphs;
• a system that automatically re-calibrates and adapts to changing patterns in full assay;
• a system for prediction of full assay of purified oil blend and blend of an oil consignment in short period; and
• a system for prediction of highly accurate full assay along with the uncertainty bounds of the prediction.
The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and/or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the spirit and scope of the embodiments as described herein.