Multi-model blending
A method and a system to perform multi-model blending are described. The method includes obtaining one or more sets of predictions of historical conditions, the historical conditions corresponding with a time T that is historical in reference to current time, and the one or more sets of predictions of the historical conditions being output by one or more models. The method also includes obtaining actual historical conditions, the actual historical conditions being measured conditions at the time T, assembling a training data set including designating the two or more set of predictions of historical conditions as predictor variables and the actual historical conditions as response variables, and training a machine learning algorithm based on the training data set. The method further includes obtaining a blended model based on the machine learning algorithm.
1. A method of multi-model blending, comprising:
obtaining, with a processing device, one or more sets of predictions of historical conditions, the historical conditions corresponding with a time T that is historical in reference to current time, and the one or more sets of predictions of the historical conditions being output by one or more models;
obtaining actual historical conditions, the actual historical conditions being measured conditions at the time T;
assembling a training data set including designating the one or more set of predictions of historical conditions as predictor variables and the actual historical conditions as response variables;
training a machine learning algorithm based on the training data set; and
obtaining a blended model based on the machine learning algorithm.
2. The method according to claim 1 , wherein the assembling the training data set includes additionally designating one or more parameters as the predictor variables, the one or more parameters not being output by the one or more models.
3. The method according to claim 1 , wherein the obtaining the one or more sets of predictions of the historical conditions includes executing the one or more models based on receiving historical measurements from one or more sources.
4. The method according to claim 1 , wherein the obtaining the one or more sets of predictions of the historical conditions includes receiving the one or more sets of predictions of the historical conditions from one or more sources.
5. The method according to claim 1 , wherein the training the machine learning algorithm includes training a neural network, linear regression model, generalized additive model, random forest model, gradient boosted regression model, or supported vector machine model and using supervised learning, unsupervised learning, statistical classification, association rule learning, hierarchical clustering, partitional clustering, reinforcement learning, or deep learning.
6. The method according to claim 1 , wherein the one or more models forecast weather conditions and the blending outputs of the one or more models includes blending outputs of temperature, pressure, humidity, wind speed, global solar irradiance, direct normal solar irradiance, accumulated rain, or snow depth.
7. The method according to claim 6 , wherein the blending outputs of the one or more models includes blending outputs of one or more of a Eulerian model, a Lagrangian model, a numerical weather model, or a climatology model.
8. The method according to claim 1 , wherein the one or more models forecast pipeline corrosion and include a corrosion science model, electrochemical-passivity model, electrochemical-nucleation model, propagation model, corrosion engineering model, mechanistic model, or statistical model.
9. A system to perform multi-model blending, the system comprising:
an input interface configured to receive inputs;
a processor configured to obtain one or more sets of predictions of historical conditions based on the inputs, the one or more sets of predictions of the historical conditions being output by one or more models, to obtain actual historical conditions, the actual historical conditions being measured conditions at the time T, to assemble a training data set by designating the one or more set of predictions of historical conditions as predictor variables and the actual historical conditions as response variables, and to train a machine learning algorithm based on the training data set; and
an output interface configured to output blended output based on a blended model resulting from the machine learning algorithm.
10. The system according to claim 9 , wherein the processor assembles the training data set by additionally designating one or more parameters as the predictor variables, the one or more parameters not being output by the one or more models.
11. The system according to claim 9 , wherein the inputs are historical measurements, and the processor obtains the one or more sets of predictions of the historical conditions based on processing the historical measurements with the one or more models, or each of the inputs is a set of the one or more sets of predictions.
12. The system according to claim 9 , wherein the machine learning algorithm includes a neural network, linear regression model, generalized additive model, random forest model, gradient boosted regression model, or supported vector machine model using supervised learning, unsupervised learning, statistical classification, association rule learning, hierarchical clustering, partitional clustering, reinforcement learning, or deep learning.
13. The system according to claim 9 , wherein the one or more models forecast weather conditions and output temperature, pressure, humidity, wind speed, global solar irradiance, direct normal solar irradiance, accumulated rain, or snow depth, and the two or models include two or more of a Eulerian model, a Lagrangian model, a numerical weather model, or a climatology model.
14. The system according to claim 9 , wherein the one or more models forecast pipeline corrosion and include two or more of a corrosion science model, electrochemical-passivity model, electrochemical-nucleation model, propagation model, corrosion engineering model, mechanistic model, or statistical model.
15. A non-transitory computer program product having computer readable instructions stored therein which, when executed by a processor, cause the processor to implement a method of multi-model blending, the method comprising:
obtaining one or more sets of predictions of historical conditions, the historical conditions corresponding with a time T that is historical in reference to current time, and the one or more sets of predictions of the historical conditions being output by one or more models;
obtaining actual historical conditions, the actual historical conditions being measured conditions at the time T;
assembling a training data set including designating the one or more set of predictions of historical conditions as predictor variables and the actual historical conditions as response variables;
training a machine learning algorithm based on the training data set; and
obtaining a blended model based on the machine learning algorithm.
16. The computer program product according to claim 15 , wherein the assembling the training data set includes additionally designating one or more parameters as the predictor variables, the one or more parameters not being output by the one or more models.
17. The computer program product according to claim 15 , wherein the obtaining the one or more sets of predictions of the historical conditions includes executing the one or more models based on receiving historical measurements from one or more sources, or the obtaining the one or more sets of predictions of the historical conditions includes receiving the one or more sets of predictions of the historical conditions from one or more sources.
18. The computer program product according to claim 15 , wherein the training the machine learning algorithm includes training a neural network, linear regression model, generalized additive model, random forest model, gradient boosted regression model, or supported vector machine model and using supervised learning, unsupervised learning, statistical classification, association rule learning, hierarchical clustering, partitional clustering, reinforcement learning, or deep learning.
19. The computer program product according to claim 15 , wherein the one or more models forecast weather conditions the blending outputs of the one or more models includes blending outputs of temperature, pressure, humidity, wind speed, global solar irradiance, direct normal solar irradiance, accumulated rain, or snow depth, and the blending outputs of the two or more models includes blending outputs of two or more of a Eulerian model, a Lagrangian model, a numerical weather model, or a climatology model.
20. The computer program product according to claim 15 , wherein the one or more models forecast pipeline corrosion, and the blending outputs of the one or more models includes blending outputs of two or more of a corrosion science model, electrochemical-passivity model, electrochemical-nucleation model, propagation model, corrosion engineering model, mechanistic model, or statistical model.