Systems and methods for emission source attribution

Patent metadata

Field	Value
Publication number	US-2023282316-A1
Application number	US-202217807685-A
Country	US
Kind code	A1
Filing date	Jun 17, 2022
Priority date	Mar 1, 2022
Publication date	Sep 7, 2023
Grant date	—

Abstract

A method for source attribution comprises receiving measurements of a chemical species at a spatially distributed sensor array for a given set of spatially positioned emission sources in a physical environment using a dispersion model. Based on the received measurements, a concentration field is mapped from the emission sources to the sensor array using a forward operator. For each emission source, a likelihood data set is evaluated at least by fitting an emission rate of the chemical species using a regression model based on the mapped concentration field and real-world, runtime measurements from the sensor array. A posterior data set is evaluated based at least on the evaluated likelihood data set and historical data for the physical environment. For each sensor of the sensor array, estimated emission rates and contribution rankings for emission sources are determined and output based on the evaluation of the posterior data set.

First claim

1 . A computer-implemented method for source attribution, comprising:
at a dispersion model, receiving measurements of a chemical species at multiple sensors of a spatially distributed sensor array in a physical environment for a given set of spatially positioned emission sources of the chemical species; based on the received measurements, mapping a concentration field of the chemical species from the set of emission sources to each of the multiple sensors using a forward operator; for each emission source, evaluating a likelihood data set that includes at least a set of sensor measurements, a variable set of source emission rates, and a parameter set for the physical environment, at least by fitting an emission rate of the chemical species using a regression model based on the mapped concentration field and real-world, runtime measurements from sensors of the spatially distributed sensor array; evaluating a posterior data set based at least on the evaluated likelihood data set and prior data for the physical environment; and for each sensor:
determining an estimated emission rate and contribution ranking for each emission source based on the evaluation of the posterior data set; and
outputting the determined estimated emission rates and contribution rankings for a predetermined number of highest ranking sources.

2 . The method of claim 1 , wherein the dispersion model is a physics-based dispersion model.

3 . The method of claim 1 , wherein receiving measurements of the chemical species at multiple sensors of the spatially distributed sensor array in the physical environment for the given set of spatially positioned emission sources of the chemical species includes generating a distribution of sensor measurements based on a prior data set representing a distribution of a background emission rate of each source and a distribution of wind velocity in the physical environment.

4 . The method of claim 1 , wherein forward operator based mapping is in the form of a non-linear, time-dependent mapping.

5 . The method of claim 1 , wherein the regression model is a Bayesian regression model.

6 . The method of claim 5 , wherein the Bayesian regression model is a non-linear Bayesian regression model.

7 . The method of claim 5 , wherein the dispersion model is refined based on feedback from the Bayesian regression model.

8 . The method of claim 1 , wherein evaluation of each posterior data set further includes:
generating a kernel density estimation of the posterior data set; sampling the kernel density estimation of the posterior data set to obtain samples; and determining a probability-weighted emission rate based on the obtained samples.

9 . The method of claim 8 , wherein the dispersion model is refined based on one or more of the obtained samples and the probability-weighted emission rate.

10 . A system for determining source attribution, comprising:
a spatially distributed sensor array; a communication subsystem; a logic machine; and a storage machine holding instructions executable by a logic machine to execute a source attribution module, the source attribution module configured to:
at a dispersion model, receive measurements of a chemical species at multiple sensors of the spatially distributed sensor array in a physical environment for a given set of spatially positioned emission sources of the chemical species;
based on the received measurements, map a concentration field of the chemical species from the set of emission sources to each of the multiple sensors using a forward operator;
for each emission source, evaluate a likelihood data set that includes at least a set of sensor measurements, a variable set of source emission rates, and a parameter set for the physical environment, at least by fitting an emission rate of the chemical species using a regression model based on the mapped concentration field and real-world, runtime measurements from sensors of the spatially distributed sensor array;
evaluate a posterior data set based at least on the evaluated likelihood data set and historical data for the physical environment; and
for each sensor:
determine an estimated emission rate and contribution ranking for each emission source based on the evaluation of the posterior data set; and
output the determined estimated emission rates and contribution rankings for a predetermined number of highest ranking sources.

11 . The system of claim 10 , wherein the dispersion model is a physics-based dispersion model.

12 . The system of claim 10 , wherein receiving measurements of the chemical species at multiple sensors of the spatially distributed sensor array in the physical environment for the given set of spatially positioned emission sources of the chemical species includes generating a distribution of sensor measurements based on a prior data set representing a distribution of a background emission rate of each source and a distribution of wind velocity in the physical environment.

13 . The system of claim 10 , wherein forward operator based mapping is in the form of a non-linear, time-dependent mapping.

14 . The system of claim 10 , wherein the regression model is a Bayesian regression model.

15 . The system of claim 14 , wherein the Bayesian regression model is a non-linear Bayesian regression model.

16 . The system of claim 14 , wherein the dispersion model is refined based on feedback from the Bayesian regression model.

17 . The system of claim 10 , wherein evaluation of each posterior data set further includes:
generating a kernel density estimation of the posterior data set; sampling the kernel density estimation of the posterior data set to obtain samples; and determining a probability-weighted emission rate based on the obtained samples.

18 . The system of claim 17 , wherein the dispersion model is refined based on one or more of the obtained samples and the probability-weighted emission rate.

19 . A computer-implemented method for source attribution, comprising:
receiving sensor measurements for multiple sensors of a spatially distributed sensor array in a physical environment; receiving emission rates and known positions for emission sources positioned in the physical environment; receiving climatology variables for the physical environment; at a process model forward simulator including at least a physics-based model, simulating dispersion of a chemical species within a physical environment based on the emission rate of each emission source, the climatology variables, and the known positions of emission source; based at least on the simulated dispersion:
at a non-linear Bayesian regression model, fitting leak-rates and previously observed measurements at each sensor;
at a deep Bayesian neural network, adding model uncertainty to the non-linear Bayesian regression model;
at a hybrid physics-machine learning model, selecting a subset of a plurality of physical solutions based on physical constraints of the physical environment; and
at an inverse model, generating an adjoint source-sensor relationship;
at a source attributor, generating leak rate quantifications for each emission source based on at least the outputs of the non-linear Bayesian regression model, the hybrid physics-machine learning model, the deep Bayesian neural network, and the inverse model; and indicating leak rate quantifications for at least each emission source associated with a known emission source position in the physical environment.

20 . The method of claim 19 , fu

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Malvar Maua Sara
• de Oliveira Nunes Leonardo
• Milletari' Mirco
• Talbert Neera Bansal
• Alaudah Yazeed Khalid
• Reynolds Jeremy Randall
• Oruganti Yagna Deepika
• Bhatia Ashish
• Badam Anirudh

Classifications

• G06N7/01 Primary

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G16C20/70 Primary

  Machine learning, data mining or chemometrics · CPC title

• G06N7/005

  Physics · mapped topic

• G06N20/20

  Ensemble learning · CPC title

• G16C20/80

  Data visualisation · CPC title