| CPC G06N 3/08 (2013.01) [G06N 3/044 (2023.01); G06N 3/045 (2023.01)] | 20 Claims |
|
1. A computer system for generating computerized predictions from first, second, and third data model architectures given one or more estimated environmental scenarios, the system comprising:
one or more processors operating with computer memory and non-transitory computer readable media, the one or more processors configured to:
receive a historical weather time-series data set and a historical climate time-series data set representative of environmental conditions present during one or more corresponding durations of time;
receive one or more target variable time-series historical data sets;
receive a predicted weather time-series data set and a predicted climate time-series data set representative of the one or more estimated environmental scenarios;
instantiate the first data model architecture configured to generate a first forecast future-state target variable for predicted environmental conditions obtained from the predicted weather time-series data set and the predicted climate time-series data set;
generate an intermediate de-trended target variable historical data set using a plurality of analogous waveform transformation features extracted from the one or more target variable time-series historical data sets, wherein the plurality of analogous waveform transformation features include amplitudes and periods of a plurality of cyclical variations at different frequencies in the one or more target variable time-series historical data sets indicated by timestamps;
instantiate the second data model architecture configured to utilize the intermediate de-trended target variable historical data set to determine a quantified weather sensitivity metric using at least a first or a second order co-efficient of variation using the historical weather time-series data set;
instantiate the third data model architecture configured to utilize the intermediate de-trended target variable historical data set to determine a quantified climate sensitivity metric using at least a first or a second order co-efficient of variation using the historical climate time-series data set;
wherein each of the first, second, and third data model architectures comprises at least one long short term memory model architecture, at least one dense neural network architecture and at least one transformer connected in at least one of a series and parallel arrangement such that the at least one long short term memory model architecture retains information from an input data set prior to passing the information from the input data set to the at least one dense neural network architecture, and the at least one transformer is configured to merge geospatial and temporal information with feature data from the input data set to model an impact of the environmental conditions on a specific feature;
generate an output data set representative of the computerized predictions for the one or more estimated environmental scenarios to determine a target variable given the predicted environmental conditions, encapsulated alongside the quantified weather sensitivity metric, and the quantified climate sensitivity metric, and
render a dynamic visual dashboard for downstream analytics using the output data set, wherein the first and the second order co-efficients of variation using the historical weather time-series data set and the historical climate time-series data set are used to prioritize a plurality of variables to be adjustable in the dynamic visual dashboard, wherein the dynamic visual dashboard includes a plurality of interactive interface controls that are utilized to modify the plurality of variables and simulate values of the plurality of variables predicted by each of the first, second, and third data model architectures, wherein the plurality of interactive interface controls are instantiated based on the first and the second order co-efficients of variation using the historical weather time-series data set and the historical climate time-series data set.
|