Lighting load classification and dimmer configuration based thereon

Patent metadata

Field	Value
Publication number	US-2025031292-A1
Application number	US-202418904538-A
Country	US
Kind code	A1
Filing date	Oct 2, 2024
Priority date	Jun 18, 2019
Publication date	Jan 23, 2025
Grant date	—

Abstract

Approaches are provided for lighting load classification using machine learning models, and selection of parameters for dimmer operation based on the classification. Approaches provide requests and/or use of updated models, building datasets for model construction, model construction, evaluation, and selection, application of machine learning models and handling unsuccessful classification attempts or classification into a class corresponding to an unknown lighting load type, and selection of operating parameters based on the foregoing.

First claim

What is claimed is:

1 . A dimmer for controlling conduction of a supply of power to a lighting load, the dimmer including:
a line input terminal and a load output terminal, the line input terminal configured to be electrically coupled to the supply of power, and the load output terminal configured to be electrically coupled to the lighting load; a switching circuit electrically coupled in series between the line input terminal and the load output terminal, the switching circuit configured to be selectively controlled between an ON state and an OFF state; a memory; and a processing circuit in communication with the memory, wherein the dimmer is configured to perform a method including:
based on conducting the supply of power to the lighting load, obtaining electrical data representing properties of electrical current through the lighting load over a duration of time;
communicating with a remote entity for application of an updated machine learning model to classify the lighting load based on the obtained electrical data;
receiving, from the remote entity, a response to the communicating, the response being dependent on compute resources of the dimmer; and
based on the response from the remote entity, configuring the dimmer with one or more dimmer operating parameters that control operation of the dimmer.

2 . The dimmer of claim 1 , wherein the response includes:
the updated machine learning model, based on the compute resources of the dimmer being sufficient to perform classification of the lighting load within a selected amount of time by applying the updated machine learning model; or an indication of the one or more dimmer operating parameters or a lighting load class into which the lighting load is classified.

3 . The dimmer of claim 2 , wherein the response includes the updated machine learning model, and wherein the method further includes applying the updated machine learning model using the obtained electrical data to classify the lighting load into a lighting load class.

4 . The dimmer of claim 3 , wherein the dimmer stores an initial machine learning model configured for classifying lighting loads into a plurality of different lighting load classes based on properties of electrical current through the lighting loads, the plurality of different lighting load classes corresponding to different lighting load types, and wherein the method further includes applying the initial machine learning model using the obtained electrical data, wherein the applying the initial machine learning model fails to classify the lighting load into any lighting load class of the plurality of different lighting load classes with at least a threshold confidence level, and wherein the communicating with the remote entity is performed in response to the failure to classify the lighting load into any lighting load class of the plurality of different lighting load classes with at least the threshold confidence level.

5 . The dimmer of claim 4 , wherein the method further includes providing an alert to a user indicating the failure to classify the lighting load into any lighting load class of the plurality of different lighting load classes with at least the threshold confidence level, the alert including at least one of an audio alert, a visual alert, or haptic feedback alert.

6 . The dimmer of claim 1 , wherein the one or more dimmer operating parameters are selected, based on the application of the updated machine learning model, to configure the dimmer to avoid flicker of the lighting load, and wherein the one or more dimmer operating parameters include at least one of: a selection between forward and reverse dimming modes, a selection of firing and ending conduction angles for each half cycle of the supply of power, a selection of a maximum conduction angle, or a selection of one or more controls for one or more dimmer startup options.

7 . A method including:
building a dataset of electrical data for machine learning model training, the building the dataset including:
sampling and storing, to the dataset, and for each different lighting load of a collection of different lighting loads, electrical data representing properties of electrical current through the lighting load, wherein for each different lighting load, the sampling is taken over a plurality of periods of an electrical waveform having a phase and the sampled electrical data is sampled at various angles of the phase across the plurality of periods, wherein the collection of different lighting loads include a plurality of different light-emitting diode (LED) lighting loads and at least one compact fluorescent lamp lighting load; and
augmenting the dataset, the augmenting including:
iteratively applying a phase shift to values of the sampled electrical data to produce additional values at corresponding shifted angles of the phase; and
adding the produced additional values to the dataset;

wherein the dataset includes time-domain data and frequency-domain data; and
constructing a machine learning model to use in classifying target lighting loads into a plurality of different lighting load classes based on properties of electrical current through the target lighting loads, the plurality of different lighting load classes corresponding to different lighting load types of the collection of different lighting loads, and wherein the constructing includes:
building and training, using the dataset, including the time-domain data and the frequency-domain data of the dataset, a plurality of different neural networks, the plurality of different neural networks having varying model parameters and being built and trained using varying artificial intelligence platforms;
evaluating performance of the different neural networks based on performance requirements that include accuracy; and
selecting a neural network of the built and trained plurality of different neural networks as the machine learning model to use, wherein the selected neural network includes a multilayer perceptron having at least one hidden layer.

8 . The method of claim 7 , wherein a sampling frequency used for a lighting load of the collection of different lighting loads is selected to be at least an order of magnitude greater than a Nyquist rate for the sampling.

9 . The method of claim 7 , wherein the augmenting by iteratively applying the phase shift accounts for varying start and stop times of sampling gathering to build the dataset.

10 . The method of claim 7 , wherein a degree of the phase shift is based on a number of samples per period of the electrical waveform.

11 . The method of claim 10 , wherein the phase shift is 1.8 degrees.

12 . The method of claim 7 , wherein the plurality of different LED lighting loads include LED lighting loads having differing converter circuits, the differing converter circuits including buck converter, buck-boost converter, tapped-buck converter, and flyback converter circuits, and wherein the plurality of different lighting load classes into which the machine learning model is configured to classify target lighting loads includes a respective at least one class for each of the differing converter circuits.

13 . The method of claim 7 , wherein the constructing the machine learning model further includes associating flicker characteristics to classes of the plurality of different lighting load classes into which the machine learning model is configured to classify target lighting loads, wherein the associating trains the machine learning model to map current waveforms to flicker properties for identifying likelihood that a given dimming strategy for use with a target lighting load will

Assignees

• Leviton Manufacturing Co

Inventors

• Jahan A M Sarwar
• Gu Wen
• Huang Tianjian

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• H05B47/1965

  using handheld communication devices · CPC title

• H05B47/175 Primary

  by remote control · CPC title