Heuristic method of automated and learning control, and building automation systems thereof

• US 20180356775A1
• Filed: 06/12/2018
• Published: 12/13/2018
• Est. Priority Date: 06/12/2017
• Status: Active Grant

• Current Assignee: PassiveLogic, Inc.
• Original Assignee: PassiveLogic, Inc.
• Inventors: Harvey, Troy Aaron
• Granted Patent: US 10,705,492 B2
• Time in Patent Office: Days
• CPC Class Codes: F24F 11/63 Electronic processing
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• CPC Class Codes: G05B 13/048 using a predictor
• CPC Class Codes: G05B 15/02 electric
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G05B 2219/2642 Domotique, domestic, home control, automation, smart house

Abstract

Disclosed is a method of physical-model based building automation using in-situ regression to optimize control systems.

Claims

1. A building system controller comprising:

a controlled system;

a physical model, comprised of a mathematical model of the controlled system;

a simulation engine;

a heuristic;

a data stream, comprised of data from said controlled system; and

a training loop;

wherein the simulation engine can simulate the behavior of the controlled system by means of evaluating the physical model; and

whereby said training loop compares the output of the simulation engine to the data stream using the heuristic, such that the physical model is regressed in a manner that the output of the simulation engine approaches the data stream.

2. The building system controller of claim 1, wherein the regressed physical model is utilized by the simulation engine to more accurately predict the future behavior of a control loop, such that an optimal control regime may be more accurately computed.

4. The building system controller of claim 1, wherein the building system controller also comprises a control loop that controls the controlled system.

5. The building system controller of claim 1, wherein the building system controller also comprises a control loop that controls the controlled system using the output of the simulation engine to predict the future behavior of the controlled system under an arbitrary control regime.

6. The building system controller of claim 2., wherein the control loop may apply the control regime to the controlled system and observe the response of the controlled system via the data stream, thereby completing the control loop.

7. The building system controller of claim 5, wherein the control loop utilizes a cost function of the computed physical model to evaluate the cost of a given control regime.

8. The building system controller of claim 7, wherein the cost may be evaluated in terms comprising one or more of, but not limited to:

energy use, primary energy use, energy monetary cost, human comfort, the safety of building or building contents, the durability of building or building contents, microorganism growth potential, system equipment durability, system equipment longevity, environmental impact, and/or energy use CO2 potential.

9. The building system controller of claim 2, wherein the optimal control regime minimizes a cost function such that the control loop may control the controlled system with least expense, according to the cost function.

10. The building system controller of claim 2, wherein the optimal control regime may be selected through comparison of one or more potential control regimes, wherein the comparison is performed by means of one or more of the following but not limited to:

differential comparison, multivariate population selection, statistical classification, clustering, feature extraction, preference ranking, and/or benchmarking.

11. The building system controller of claim 1, wherein the controlled system may comprise one or more of the following:

building automation systems, heating systems, cooling systems, ventilation systems, power management systems, renewable energy systems, irrigation systems, occupancy systems, lighting systems, environmental control systems, humidity control systems, air quality management systems, window operators, and/or shade systems.

12. The building system controller of claim 1, wherein the regression comprises one or more of the following, but not limited to:

differential comparison, multivariate population selection, statistical classification, clustering, feature extraction, preference ranking, and/or benchmarking.

13. The building system controller of claim 1, wherein the regression is performed in a manner that the output of the simulation engine approaches the data stream, wherein the extent to which the simulation engine output approaches the data stream may mean, but is not limited to:

a convergence of predicted and observed values;

a reduction in error between predicted and observed values to within an arbitrary threshold;

a reduction in uncertainty of predicted values to within an arbitrary threshold;

reaching an arbitrary threshold on number of erroneous predictions;

reaching an arbitrary threshold on number of accurate predictions;

reaching an arbitrary threshold on number of data points processed;

reaching an arbitrary threshold on computational time spent processing data.

14. The building system controller of claim 1, wherein the regression evaluation method may comprise one or more of, but is not limited to:

data slicing time slicing, time windowing, time batching, parameter slicing, parameter windowing, single-point, and/or multi-point.

15. The building system controller of claim 1, wherein the heuristic may comprise one or more of, but is not limited to:

decomposition methods, inductive methods, reduction methods, constructive methods, and/or local search methods.

16. The building system controller of claim 1, wherein the heuristic may comprise a heuristic, metaheuristic, or hyperheuristic.

17. The building system controller of claim 16, wherein the heuristic, metaheuristic, or hyperheuristic may be comprised of, but is not limited to, one or more of:

particle swarm organization, self-organizing migration algorithm, neural networks, group method of data handling, differential evolution, genetic algorithm, memetic algorithm, random forest, hill climbing algorithm, simulated annealing, monte-carlo methods, random search, fuzzy-logic, arithmetic mean, geometric mean, harmonic mean, trimean, median, mode, mid-range, quadratic mean (RMS), cubic mean, generalized mean, weighted mean, linear regression, logistic regression, polynomial regression, k-means clustering, k-nearest neighbors, decision trees, perceptron, multi-layer perceptron (neural network), kernel methods, support vector machines, ensemble methods, boosting, bagging, naï

ve Bayes, expectation maximization, Gaussian mixture models, Gaussian processes, principal component analysis, singular value decomposition, reinforcement learning, Voronoi decomposition, or social theory voting techniques and concepts, such as social welfare functions, social choice functions, single transferrable vote, Bucklin'"'"'s rule, social decision schemes, collective utility functions, and. Condorcet method and extensions such as Copeland'"'"'s rule, maximin, Dodgson'"'"'s rule, Young'"'"'s rule, and ranked pairs.

18. The building system controller of claim 1, wherein the data stream may be comprised of one or more of, but is not limited to:

a data store, a real-time data feed.

19. The building system controller of claim 1 wherein the data stream may comprise one or more of, but is not limited to:

sensor data, actuation data, system parameters, equipment state, weather data, environmental data, occupant input, and/or occupant behavior.

20. The building system controller of claim 1, wherein the training loop performs regression using the heuristic on the simulation engine output and the data stream to improve the physical model, by varying the model parameters or adapting the underlying structure and composition of the model.

3. The building system controller of claim I, wherein the regressed physical model is utilized to detect faults, by detecting divergent parameters through regression of a previously regressed model.

21. A electronic device or network of electronic devices comprising:

a coupled building system;

a data stream;

at least one memory; and

one or more processors operatively coupled to the data stream and one or more processors operatively coupled to the memories, where the processors are configured to execute program code stored in the memories to;

control the coupled building system;

receive sensor measurements obtained with the data stream, and store the data he memory;

compute a mathematical model that substantially imitates the behavior of the controlled system, generating a predicted data stream measurement;

analyze the variation between the stored data stream and the predicted data stream, using a cost function.

iterate the process of computation and analysis, while changing parameters of the mathematical model as to regress the model in a manner that the output of the simulation engine approaches the data stream;

control the coupled building system using predicted actuation using the regressed mathematical model;

22. The electronic device of claim 21, wherein the mathematical model may be computed repeatedly, in arbitrary order, to evaluate each of an arbitrary number of different physical models.

23. The electronic device of claim 21, wherein the mathematical model computation engine may comprise one or more computational entities in a network of simulation engines.

24. The electronic device of claim 21, wherein the mathematical model computation engine may be comprised of collocated computational entities as well as remote computational entities.