EP-4558940-B1 - SYSTEMS AND METHODS FOR GREENHOUSE GAS MITIGATION

Inventors

• BRONEVETSKY, Grigory
• PRADHAN, SALIL VIJAYKUMAR
• STIVORIC, JOHN MICHAEL
• WILLIAMS, Dominic Deshawn
• BOISSEREE, KAITLIN MARIE
• SINGAL, DHRUV
• CHONA, ASHISH JAGMOHAN

Dates

Publication Date

20260513

Application Date

20240717

Claims (15)

1. A computer-implemented method for updating a set of tasks comprising: generating (210) a set of tasks, wherein each task comprises a greenhouse gas, GHG, offset potential and one or more failure mechanisms, and wherein the set comprises a metric; determining (220), by a machine learning model and based on multiple data types from a plurality of sources, wherein at least one of the data types is based on real-world measurements, that an overall risk score of the set exceeds a first failure threshold due to a risk score of a task of the set of tasks exceeding a second threshold; selecting (230) a replacement task for the task, the selecting comprising: receiving (231), a plurality of replacement candidates, each replacement candidate comprising a candidate GHG offset potential and one or more candidate failure mechanisms; assigning (232), by the machine learning model and to each of the plurality of replacement candidates, a replacement score for the replacement candidate based on a failure correlation of the replacement candidate with respect to each other sets of the set of tasks; ranking (233) the plurality of replacement candidates based on the replacement scores; and selecting (234), based on the ranking, the replacement task generating (240) an updated set of tasks including the replacement task; and generating instructions to manage the updated set of tasks with one or more project performers.
2. The method of claim 1, wherein assigning the replacement score for the replacement candidate based on failure correlation comprises assigning the replacement score based on (i) predictive rates of failure and (ii) a predicted GHG offset potential.
3. The method of claim 1, wherein: a) ranking the plurality of replacement candidates based on the replacement scores further comprises: i) determining, for the task of the set of tasks exceeding the failure threshold, a mitigation failure value; and ii) ranking the replacement candidates based on respective potential of each candidate replacement project to repair the mitigation failure value; or b) ranking the plurality of replacement candidates further comprises ranking the replacement candidates based on real-time data collected from similar mitigation projects.
4. The method of claim 1, wherein determining, by the machine learning model and based on multiple data types from the plurality of sources, that the task of the set of tasks exceeds the failure threshold for the one or more failure mechanisms comprises: predicting, by the machine learning model and based on real-time data collected for the task based on the multiple data types from the plurality of sources, that future variations of a mitigation for the task are below a threshold mitigation.
5. The method of claim 1, further comprising: determining, by the machine learning model and for a failure scenario, an impact on mitigation outcomes for respective failure mechanisms of each of the tasks of the set; predicting, based on aggregated impacts across all the tasks of the set, a total impact of the scenario on the set of tasks; and using the total impact, determining the overall risk score.
6. The method of claim 1, further comprising training the machine learning model comprising: receiving training data, from the plurality of sources and including multiple data types, data representative of a plurality of tasks; and providing, to the machine learning model, the training data, wherein the training data representative of each task includes (i) rates of failure, (ii) GHG offset results, (iii) correlation strength to one or more other tasks.
7. The method of claim 6, wherein providing the training data comprises updating the training data using a transfer learning machine learning model, and training the machine learning model comprises using the updated training data to train the machine learning model, optionally wherein the transfer learning machine learning model is configured to: based on an image associated with one of the tasks, determine a set of simulation parameters for a simulation simulating the one of the tasks; and provide the set of simulation parameters as a portion of the training data to a machine learning model.
8. The method of claim 1, further comprising: determining, for one or more of the tasks of the set of tasks, a permanence action supportive of the task, the permanence action counteracting at least one of the one or more failure mechanisms; and generating, the updated set of tasks including the permanence action.
9. The method of claim 8, wherein: the method further comprises determining that the selected replacement task has a first failure mechanism, wherein the permanence action has a second failure mechanism different from the first failure mechanism; or the permanence action comprises generating, in a market ecosystem, an incentive supportive of one or more of the tasks, wherein the incentive reduces a probability of the at least one of the one or more failure mechanisms.
10. The method of claim 8, wherein determining the permanence action comprises: determining, for a set of two or more tasks of the set, that the permanence action counteracts the respective failure mechanisms of the set of two or more tasks; or determining that the permanence action is supportive of the task for a period of time, wherein the permanence action supportive of the task is an investment of a carbon credit, carbon offset, or a combination thereof, for a period of at least 5 years.
11. The method of claim 1, further comprising: receiving new input data including data indicating at least one of the GHG offset potential, the failure mechanism, and the risk score of an associated task is incorrect; determining updated values for the at least one of the GHG offset potential, the failure mechanism, and the risk score of the associated task that is indicated to be incorrect; and updating the set with the updated values for the at least one of the GHG offset potential, the failure mechanism, and the risk score of the associated task that is indicated to be incorrect.
12. The method of claim 11, wherein: a) the new input data includes data regarding ecological conditions related to the failure mechanisms associated with respective tasks; or b) the method further comprises: in response to updating the set with the updated values, determining that the overall risk of the set exceeds the failure threshold, and in response to determining that the overall risk of the set exceeds the failure threshold, selecting another replacement task for the set.
13. The method of claim 1, further comprising: receiving measurements indicating progress of the set of tasks, the receiving comprising: receiving, from a sensor, data indicative of an ecological condition, and using the data indicative of the ecological condition as input data, executing a simulation to provide output data, wherein comparing the measurements of the metric comprises using the output data, and comparing the measurements to the metric; and in response to the comparison between the measurements and the metric, determining to select another replacement task.
14. A system comprising one or more computers (700) and one or more storage devices (730) on which are stored instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform the method of any of the previous claims.
15. A non-transitory computer storage medium (730) encoded with instructions that, when executed by one or more computers (700), cause the one or more computers to perform the method of any of claims 1 to 13.

Description

Claim of Priority This application claims priority under 35 USC §119(e) to U.S. Provisional Patent Application Serial No. 63/514,040 filed on July 17, 2023. TECHNICAL FIELD This specification relates to systems and methods for selecting a set of tasks and predicting outcomes of those tasks for greenhouse gas mitigation. BACKGROUND Greenhouse gas (GHG) emissions from various economic activities are contributing to climate change in the form of higher temperatures and ocean acidification, among other effects. For example, CO2 emissions are definite and last for ~500 years in the atmosphere. Systems that encourage reduction in GHG emissions and sequestering GHGs to undo past emissions and the associated effects are under development. Changes in operations that reduce emissions or remove existing emissions from the atmosphere are called "mitigations." Conventional carbon mitigation projects are rarely a perfect match for the liability incurred by GHG emissions. This is because unlike emissions, which are definite and long-duration, mitigation projects may fail to mitigate or fail to maintain mitigation permanence for various reasons, such as forest fire, poor project management, and lack of additionality. US 2002/138318 A1 discloses an integrated project management and risk management system. A risk processor has access to a project data store containing a plurality of inter-related project actions used for project management purposes, and a risk data store which contains a plurality of inter-related project activities and associated risk indicators used for related risk management purposes. The risk processor is operable to generate and write to the risk data store changes to the project activities and risk indicators reflecting changes in the project actions, to generate or receive mitigating activities identified to reduce or prevent a risk or the consequences of a risk associated with a project activity and to write corresponding project action changes to the project data store US2012290104A1 and US2022391921A1 are further prior art. SUMMARY The matter for protection is defined by the claims. This specification describes systems and methods for selecting tasks, e.g., actions to be taken, as part of a set of tasks for greenhouse gas mitigation. The sets of tasks are dynamically updated depending on real-time data. For example, changes over time in the availability of project opportunities, societal demand for carbon mitigation, and market value of GHG offset credits, e.g., carbon offset credits, can affect task selection. The systems and methods can generate unique solutions for multiple sets such that no two sets of tasks necessarily have the same set of multi-year solutions and design high quality sets of tasks for distinct, future scenarios (e.g., climate change, economics, and fires rates). The system can include one or more machine-learning (ML) models trained and reinforced on a large dynamic data set including a diversity of data sources and data types. The data set can be generated and updated in real-time to maintain a dynamic predictive outcome of the set, e.g., to maximize a portfolio's carbon mitigation, reflecting past events and possible futures. The methods and system being dynamic refers, in part, to how updating the set of tasks is proactive, e.g., updating the set of tasks based on future predictions. The trained ML model can be used to learn high-quality task selection to achieve long-term mitigation goals that are resilient to scenarios where mitigation storage fails (permanence failures). The trained ML model can be used to generate non-intuitive predictions including failure/risk correlations of GHG mitigation tasks which can refine a set by, for example, (i) ranking tasks based on predicted rates of failure, (ii) predicting GHG offset potential, (iii) correlation of risk in current tasks included in a dynamic set, and (iv) balancing the time when a task starts against its efficiency (earlier starts mitigate more emissions but by delaying by a few years may enable the use of new technologies, e.g., more reliable and/or cheaper technologies). The subject matter described in this specification as implemented in particular embodiments realizes one or more of the following technical advantages. The system includes a dynamic predictive feedback loop using data obtained from multiple resources (e.g., having different sources and different formats) to maintain and update an outcome prediction of a set of tasks. The large data set generated from the multiple different sources and having different formats can be used to train a ML model to generate predictions related to, for example, diversification between tasks in a set, individual and group task risk/failure mechanisms, and retroactive long-term GHG mitigation effectiveness. A ML model, can be trained and reinforced on this dynamic, real-time data set to generate non-intuitive predictions related to failure mechanisms of the tasks in a set,