AUTOMATICALLY CONFIGURING OF A LIGHTING

Abstract

The invention relates to automatically configuring of a lighting created by a lighting system, particularly to creating a lighting, which follows a person, with a networked lighting system. A basic idea of the invention is to adapt the learning based processing of activations of lamps and presence detectors received as input data to changes in a system for automatically configuring lighting with a variable adaptation rate in order to make the automatic lighting configuration more robust with regard to system changes. An embodiment of the invention relates to a system (10) for automatically configuring a lighting, wherein the system comprises—a network of lamps (12), wherein each lamp is activated by a control signal and outputs its activation,—a network of presence detectors (14), wherein each presence detector is activated by detecting an activity and outputs its activation, and—a lighting configurator (16) for processing activations of lamps and presence detectors received as input data (18) based on learning and outputting control signals (20) for lamps depending on the processed input data for configuring the lighting, wherein the processing of input data by the lighting configurator can be adapted to changes in the system with a variable adaptation rate (22).

Description

TECHNICAL FIELD

The invention relates to automatically configuring of a lighting effect created by a lighting system, particularly to creating a lighting effect which follows a person, with a networked lighting system.

BACKGROUND ART

In the non-pre-published international patent application PCT/EP2010/050836 a system for automatically configuring a lighting is described, which is able to learn to predict which lamp will be the next lamp to turn on. Based on this information a lighting control system can turn lamps on before their connected presence detector will detect activity, increasing the user comfort.

SUMMARY OF THE INVENTION

It is an object of the invention to further improve the automatically configuring of a lighting as described in the non-pre-published international patent application PCT/EP2010/050836.

The object is solved by the subject matter of the independent claims. Further embodiments are shown by the dependent claims.

A basic idea of the invention is to adapt the learning based processing of activations of lamps and presence detectors received as input data to changes in a system for automatically configuring lighting with a variable adaptation rate in order to make the automatic lighting configuration more robust with regard to system changes. Thus, a system for automatically configuring a lighting according to the invention can take into account that noise can occur, particularly when designing a system such as the one described in PCT/EP2010/050836, and there may be a fundamental trade off between how robust such a learning based system is against noise and how fast the system is able to adapt to a new pattern of input data caused by a change in the system configuration. Noise can be for example created by random deviations in the patterns of input data of the input data processing. This follows from the fact that if the configuration of the system changes it will always appear as if the system did not change but is experiencing noise on its original input data pattern. Therefore the more robust a system is against noise the slower it will be in adapting to a new system configuration or a change in usage input data patterns. An input data pattern is a certain pattern of activations of lamps and presence detectors, which are received as input data and processed for the lighting configuration.

An embodiment of the invention provides a system for automatically configuring a lighting, wherein the system comprises:

• • a network of lamps, wherein each lamp is activated by a control signal and outputs its activation,
  • a network of presence detectors, wherein each presence detector is activated by (in) directly detecting an activity and outputs its activation, and
  • a lighting configurator for processing activations of lamps and presence detectors received as input data based on learning and outputting control signals for lamps depending on the processed input data for configuring the lighting, wherein the processing of input data by the lighting configurator can be adapted to changes in the system with a variable adaptation rate.

The lighting configurator may comprise

• • a model of a random process of activations of lamps and presence detectors, wherein the model of the random process is configured
  • to learn patterns of activations of lamps and presence detectors from the received input data,
  • to predict a pattern of activations of lamps and presence detectors based on received input data, learned patterns of activations of lamps and presence detectors and a current setting of the adaptation rate, and
  • to output the predicted pattern as the control signals.

The lighting configurator may comprise a configurator to lower or increase the adaptation rate. The configurator may particularly enable a user to manually lower or increase the adaptation rate.

Lowering the adaptation rate may comprise one of the following:

• • changing the amount of historically input data that is used by the model of the random process in the process of learning,
  • changing how historically input data is weighted by the model of the random process in the learning process,
  • changing a parameter that determines how fast changes are made to the model of the random process that predicts the paterns of activations,
  • any action that turns the learning behavior completely on or off,
  • setting the adaptation rate such that the model of the random process is not learning at all,
  • resetting a memory of the lighting configurator such that all estimations of parameteres of the random process are removed,
  • any other action that makes the system adapt the speed with which it adapts its prediction when it receives feedback on whether its predicting model is correct or not.

The lighting configurator may be configured to detect a change in the system in case of an event from the group consisting of

• • a user input related to the adaptation rate received by the the system via a user interface,
  • the adding or removing of a presence detector and/or lamp to/from the network of presence detectors and/or lamps,
  • any changes in the network configuration including (relative) signal strengths, routing tables, physical network connections,
  • noise in the model of the random process that matches a set of changes to the system, wherein the set of changes comprises the following: a lamp or presence detector has been added; a lamp or presence detector has been removed/got broken; a lamp or presence detector has been put to another location; the network of lamps and/or presence detectors got portioned.

A further embodiment of the invention relates to a method for automatically configuring a lighting in a system of the invention and as described above, wherein the method comprises the acts of

• • processing activations of lamps and presence detectors received as input data and
  • outputting control signals for lamps depending on the processed input data for configuring the lighting, wherein the processing of input data by the lighting configurator can be adapted to changes in the system with a variable adaptation rate.

The act of processing the input data may comprise the model of the random process implemented by a function for predicting whether a lamp i will be activated in a first timestep t₁ based on all previous events e_t2 and activation of the lamp i in a second timestep t₂, wherein an event e_tx comprises any information that can be used to predict whether a particular lamp of the network of lamps will be activated in a certain time period t_x.

The function may receive as further input a learning rate parameter L_R that indicates how more recent events e_t1 should be weighted compared to older events e_t2.

The function may comprise

• • a first function f that takes as inputs events E={e_t2, 0<e_t2<e_t1} at the second timestep t₂, the lamp activation L={l_it2, 0<l_it2<l_it1} at the second timestep t₂ and the learning rate parameter L_R, and
  • a second function m that is output by the first function and takes as inputs events E={e_t2, 0<e_t2<e_t1} at the second timestep t₂ and outputs whether the lamp activation l_it1 at the first timestep t₁ is true or false.The first function f and the second function m may be combined in one function.The first function f and/or the second function m may be implemented by one of the group consisting of
  • a neural network algorithm,
  • a machine learning algorithm,
  • a statistical pattern recognition algorithm,
  • a reinforcement learning algorithm,
  • a dynamic programming algorithm,
  • an optimisation function algorithm,
  • a genetic algorithm.In the method
  • the learning rate parameter L_R may change over the time and/or
  • the actication l of a lamp i may be a value indicating a dimming level of the lamp i.

A yet further embodiment of the invention provides a computer being configured to perform a method according to the invention such as a PC (Personal Computer) and comprising an interface to a network of lamps and a network of presence detectors. The computer may execute a program with a graphical user interface, allowing a user to comfortably adjust the adaptation rate for adapting the processing of input data to changes in the system. The computer may be further configured by a program to implement the lighting configurator of the invention.

Another embodiment of the invention provides a computer program enabling a processor to carry out the method according to the invention and as specified above.

According to a further embodiment of the invention, a record carrier storing a computer program according to the invention may be provided, for example a CD-ROM, a DVD, a memory card, a diskette, internet memory device or a similar data carrier suitable to store the computer program for optical or electronic access.

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

The invention will be described in more detail hereinafter with reference to exemplary embodiments. However, the invention is not limited to these exemplary embodiments.

BRIEF DESCRIPTION OF DRAWINGS

FIG. 1 shows a system for automatically configuring a lighting according to an embodiment of the invention; and

FIG. 2 shows a block diagram of a lighting configurator according to an embodiment of the invention.

DESCRIPTION OF EMBODIMENTS

In the following, functionally similar or identical elements may have the same reference numerals.

FIG. 1 shows a system 10 for automatically configuring the lighting in a network of lamps 12, each being coupled to a presence or movement detector 14. The shown system 10 can be implemented as described in the non-pre-published international patent application PCT/EP2010/050836, which is incorporated herein by reference. In this system, activations of lamps 12 and presence detectors 14 are processed to configure the lighting. The system described in PCT/EP2010/050836 tries to find patterns in the activation of the presence sensors 14, UI (user interface) switches and activations of the lamps 12. An assumption underlying the system described in PCT/EP2010/050836 is that there is a random process of activations of lamps and presence detectors that generates the data and that by (implicitly) estimating the parameters of this random process the patterns can be predicted as good as possible. The random process is approximated in the system by estimating parameters of the random process and using the estimations for the approximation. Deviations from the approximation can cause an inappropriate configuration of the ligthing.

The system 10 of FIG. 1 comprises a lighting configurator 16, which is shown in FIG. 2 in detail. The lighting configurator 16 comprises a model of the random process 24. The model 24 is configured to learn patterns of activations of the lamps 12 and presence detectors 14 from the received input data 18. The learned patterns are stored in a database 26 in the lighting configurator 16. The learned patterns 26 together with the model 24 are an estimation of the parameters of the random process and form an approximation of the random process. The model 24 of the random process processes the received input data 18, i.e. activations of lamps 12 and presence detectors 14 in the system, by using the learned patterns 26 and a current setting of the adaptation rate 22. The processing predicts a certain pattern of actiavations of lamps and presence detectors, which is output as control signals 20 for the lamps 12 of the system 10.

The lighting configurator 16 can be implemented as a separate aparatus, for example a computer coupled with the network of lamps 12 and presence detectors 14, or also as a distributed system, particularly implemented in the lamps 12, which may comprise an embedded computer to perform tasks of the lighting configurator.

Communication in the system 10 can be wired and/or wireless. For example, the lamps 12 and presence detectors 14 can communicate with the lighting configurator 16 by using a wireless technology such as Bluetooth®, ZigBee™, Wi-Fi™. The lamps 12 can also be implemented to use a Powerline technology to cuumincate via a LAN (Local Area Network) with the lighting configurator 16.

The presence detectors 14 may be integrated in the lamps 12 and be part of the lamps 12 or implemented as separate devices and broadcast their activation signals. As already mentioned above, the lighting configurator 16 receives as input data 18 activations of lamps 12 and presence detectors 14 and outputs control signals 20 for the lamps 12 in order to configure the lighting of the lamps 12. The lighting configurator 16 receives as further input a variable adaptation rate 22, with which the processing of the input data 18 by the model 24 can be adapted to changes in the system 10, which may cause deviations from the approximation of the random process with the model.

The variable adaptation rate 22 can be for example input by a user via a UI switch of the system 10 or by a device coupled with the system 10, for example a smartphone 28, Tablet-PC, notebook, PDA (Personal Digital Assistant) and comprising a UI for obtaining access to functions of the system 10. FIG. 1 shows a smartphone 28 with a lighting control application (or app), which can communicate 30 with the lighting configurator 16 of the system via some kind of wireless technology such as Wi-Fi™ or a 3G mobile communication technology. With the app, a user can adjust the variable adaptation rate 22 by using the smartphone.

As mentioned before, changes in the system 10 can cause deviations from a current approximation of the random process underlying the automatic configuring of the lighting. Deviations can in principle have two causes:

The deviations are caused by noise in the random processThe random process generating the random data has changed for example (due to a system change):A lamp has been addedA lamp has been removed/got brokenA lamp has been put in another locationA sensor has been addedA sensor has been removed/got brokenA sensor has been put in another locationThe network got portioned

There is fundamentally no way to differentiate between these two events, the only thing that can be done is changing the speed with which the estimation of the random process is adapted. The adaption of the estimation of the random process can be performed with the variable adaptation rate 22. The adaptation rate is herein also called the learning rate since it influences a learning of the random process to varying input data.

The learning rate can vary between no learning at all, usually depicted as 0 and only taking into account the last received information, usually depicted as 1. Choosing the right setting for the learning rate is therefore balancing between a system that will adapt very quickly whenever the system is changed but is not very robust against noise and a system that is very robust against random noise but will not learn changes to the random process very fast. In the system 10, the learning rate can be varied based on evidence that the random process has changed.

The learning rate should be lowered when a user suspects or knows that the setup of the system has changed and increased when a user is very confident that her/his estimation that the learning rate is correct. This includes setting the learning rate such that the system is not learning at all, or resetting the memory of the learning system such that all estimations of the random process are removed.

The indicators that can be used to adapt the current setting of the learning rate include:

A user action using some sort of interface where he could explicitly change:

The system to non-learning

Resetting the system to forget all current estimations of the model behind the random process

Setting the learning rate to any other speed

The adding or removing of a sensor and or lamp to the network

Changes to the communication network including: routing tables, signal strengths, physical connections

Noise in the random process that matches a limited set of changes to the system:

A lamp has been added

A lamp has been removed/got broken

A lamp has been put in another location

A sensor has been added

A sensor has been removed/got broken

A sensor has been put in another location

The network got portioned

In the following, an embodiment of a method for automatically configuring a lighting in the system 10 is described. The method may be implemented in the lighting configurator 16 and processes the input data 18 and creates the control signals 20 for the lamps for configuring the lighting. First, some important terms used for the description are defined:

Event, any information that can be used to predict whether a particular lamp will turn on in a particular time period; the letter e is used to indicate an event Lamp activation at time step t, with l_it it is indicated whether lamp I was on or off in timestep t

e_t indicates all events that were received within in timestep t, this includes all activations of other lamps.

Model of the random process, is a function that predicts whether l_i will be on in a certain timestep t₁ based on all previous events and activation e_t2 and l_it2 where 0<t₂t₁

The problem is to estimate l_t1 based on {e_t2,0<e_t2<e_t1}). In order to do this a higher order function f is created that takes as in input E={e_t2,0<e_t2<e_t1}, L={l_it2,0<l_it2<l_it1} and learning rate parameter L_R that indicates how more recent information should be weighted compared to older information, this parameter is not limited to be a single number but could also be a function, a vector, a matrix or any other form. The function f outputs another function m that takes as input {e_t2,0<e_t2<e_t1} and outputs whether l_t1 is true or false.

Further embodiments of the above described method are listed in the following:

f and m could be combined in one function where the output is given directly, it could even be the case that f and m are not explicitly separated when they are combined

L_R can be changing over time (for a list of indicators that could be used to alter L_R see earlier)

f and/or m can be implemented using a wide variety of technologies including (but not exhausting):

Neural networks

Machine learning

Statistical Pattern Recognition

Reinforcement learning

Dynamic programming

Optimisation functions

Genetic algorithms

L_i could not only have a value depicting on or off but could also indicate a dimming level between off and a certain maximum

This system could work for both one or multiple users at the same time The invention can be used in all kinds of lighting systems in which the lighting system should “think ahead” of where a person is going and turn the lights on in advance so the person can already see where she/he is going. The invention could however also be used in other situation such as in decorative or art systems in which the system somehow follows the user for example by decorative light, video or computer images projected on the wall or ceiling, or with other output modalities such as sound and water.

At least some of the functionality of the invention may be performed by hard- or software. In case of an implementation in software, a single or multiple standard microprocessors or microcontrollers may be used to process a single or multiple algorithms implementing the invention.

It should be noted that the word “comprise” does not exclude other elements or steps, and that the word “a” or “an” does not exclude a plurality. Furthermore, any reference signs in the claims shall not be construed as limiting the scope of the invention.

Claims

1. A system for automatically configuring a lighting, wherein the system comprises a network of lamps, wherein each lamp is activated by a control signal and outputs its activation,a network of presence detectors, wherein each presence detector is activated by detecting an activity and outputs its activation, anda lighting configurator comprising a model of a random process of activations of lamps and presence detectors, wherein the model of the random process is configured to learn patterns of activations of lamps and presence detectors received as input data, to predict a pattern of activations of lamps and presence detectors based on received input data, the learned patterns of activations of lamps and presence detectors and a current setting of an adaptation rate, and to output the predicted pattern as control signals for lamps, wherein the predication of the pattern of activations of lamps and presence detectors is adaptable to changes in the system by variation of the adaptation rate.

2. (canceled)

3. The system of claim 1, wherein the lighting configurator is configured to alter the adaptation rate.

4. The system of claim 3, wherein the lowering of the adaptation rate comprises one of the following: changing the amount of historically input data that is used by the model of the random process in the process of learning,changing how historically input data is weighted by the model of the random process in the learning process,changing a parameter that determines how fast changes are made to the model of the random process that predicts the paterns of activations,any action that turns the learning behavior completely on or off,setting the adaptation rate such that the model of the random process is not learning at all,resetting a memory of the lighting configurator such that all estimations of parameters of the random process are removed,any other action that makes the system adapt the speed with which it adapts its prediction when it receives feedback on whether its predicting model is correct or not.

5. The system of claim 1, wherein the lighting configurator is configured to detect a change in the system in case of an event from the group consisting of a user input related to the adaptation rate received by the the system via a user interface,the adding or removing of a presence detector and/or a lamp to/from the network of presence detectors and/or lamps,any changes in the network configuration including signal strengths, routing tables, physical network connections,noise in the model of the random process that matches a set of changes to the system, wherein the set of changes comprises the following: a lamp or presence detector has been added; a lamp or presence detector has been removed/got broken; a lamp or presence detector has been put to another location; the network of lamps and/or presence detectors got portioned.

6. A method for automatically configuring a lighting in a system a network of lamps wherein each lamp is activated by a control signal and outputs its activation, and a network of presence detectors, wherein each presence detector is activated by detecting an activity and outputs its activation, wherein the method comprises the acts of learning patterns of activations of lamps and presence detectors received as input data.predicting a pattern of activations of lamps and presence detectors based on received input data, the learned patterns of activations of lamps and presence detectors and a current setting of an adaptations rate, andoutputting the predicted pattern as control signals for lamps, wherein the prediction of the pattern of activations of lamps and presence detectors is adaptable to changes in the system by variation of the adaptation rate.

7. The method of claim 6, wherein the act of processing the input data comprises the model of the random process implemented by a function for predicting whether a lamp i will be activated in a first timestep t1 based on all previous events et2 and activation of the lamp i in a second timestept t2, wherein an event etx comprises any information for predicting whether a particular lamp of the network of lamps will be activated in a certain time period tx.

8. The method of claim 7, wherein the function receives as further input a learning rate parameter LR that indicates how more recent events et1 should be weighted compared to older events et2.

9. The method of claim 8, wherein the function comprises a first function f that takes as inputs events E={et2, 0<et2<et1} at the second timestep t2, the lamp activation L={lit2, 0<lit2<lit1} at the second timestep t2 and the learning rate parameter LR, anda second function m that is output by the first function and takes as inputs events E={et2, 0<et2<et1} at the second timestep t2 and outputs whether the lamp activation lit1 at the first timestep t1 is true or false.

10. The method of claim 9, wherein the first function f and the second function m are combined in one function.

11. The method of claim 9, wherein the the first function f and/or the second function m are implemented by one of the group consisting of a neural network algorithm,a machine learning algorithm,a statistical pattern recognition algorithm,a reinforcement learning algorithm,a dynamic programming algorithm,an optimisation function algorithm,a genetic algorithm.

12. The method of claim 8, wherein the learning rate parameter LR changes over the time and/orthe actication l of a lamp i is a value indicating a dimming level of the lamp i.

13-15. (canceled)