Smart Building Energy and Comfort Management Based on Sensor Activity Recognition and Prediction
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Title of Series | ||
| Number of Parts | 12 | |
| Author | ||
| Contributors | ||
| License | CC Attribution - ShareAlike 4.0 International: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor and the work or content is shared also in adapted form only under the conditions of this | |
| Identifiers | 10.5446/51309 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | ||
| Genre | ||
| Abstract |
|
00:00
BuildingPlastikkarteFood energyData managementPattern recognitionPredictionTelecommunicationInformation securityProcess modelingBuildingPattern recognitionPlastikkarteRevision controlFood energyData managementFocus (optics)Goodness of fitPredictability
00:15
Physical systemModule (mathematics)Pattern recognitionPredictionScheduling (computing)SimulationData managementSequenceStatisticsComplex (psychology)PlastikkarteBuildingFood energyConstraint (mathematics)Integrated development environmentComputer networkUser interfaceUser profileEvent horizonInstance (computer science)SpacetimeVector spaceData modelBit rateAveragePhase transitionEquals signVector graphicsFunction (mathematics)InformationPoint (geometry)2 (number)Presentation of a groupElement (mathematics)PlastikkartePhysical systemBuildingSimulationEndliche ModelltheoriePattern recognitionSet (mathematics)Observational studySound effectData managementModule (mathematics)ResultantComplete metric spaceSoftwareConstraint (mathematics)Interactive televisionStatisticsProcess (computing)Software developerState observerEvent horizonVector spaceMultiplication signInformationAlgorithmFunction (mathematics)Scheduling (computing)PredictabilitySequenceFood energyPerformance appraisalSpacetimeInstance (computer science)Bit rateIntegrated development environmentGame controllerData storage deviceWave packetReduction of orderPhase transitionCross-correlationAssociative propertyDifferent (Kate Ryan album)User profileTotal S.A.NumberWindowCountingNeuroinformatikTask (computing)VideoconferencingThomas BayesMusical ensemblePower (physics)VoltmeterRight angleCoalitionBitType theoryExpressionCoefficient of determinationShooting methodArithmetic meanRevision controlStress (mechanics)Level (video gaming)Focus (optics)Metropolitan area networkDisk read-and-write headLogic gateOcean currentValuation (algebra)Compass (drafting)Beat (acoustics)Fisher's exact testSlide ruleLatent heatServer (computing)Inflection pointSequelElectronic program guideTube (container)Computer animation
06:59
Virtual machineThresholding (image processing)VacuumScheduling (computing)Water vaporAlgorithmLaptopTask (computing)Food energyModule (mathematics)Pattern recognitionPredictionPerfect groupSimulationFrequencyRange (statistics)Bit rateStandard deviationDistribution (mathematics)Maxima and minimaPhysical systemCross-correlationSource codeInformationPhysical systemAlgorithmPerformance appraisalPattern recognitionModule (mathematics)Multiplication signWater vaporVirtual machineDifferent (Kate Ryan album)Table (information)Contrast (vision)Right angleFood energyScheduling (computing)PredictabilityFrequencyTask (computing)Computer simulationThresholding (image processing)Shift operatorWave packetMoment (mathematics)IterationSoftware testingCASE <Informatik>Error messageEndliche ModelltheorieBit rateResultantReduction of orderCross-correlationInformationSlide ruleGraph (mathematics)Incidence algebraMathematicsException handlingLevel (video gaming)Adaptive behaviorClassical physicsPairwise comparisonGame controllerEvent horizonMonster groupInterior (topology)Numbering schemeShared memoryPlastikkarteBeat (acoustics)Group actionGodForm (programming)Electronic visual displayCorrespondence (mathematics)Degree (graph theory)Binary fileWeightThomas BayesCopyright infringementComputer fileDisk read-and-write headPresentation of a groupSimulationSpecial unitary groupNormal (geometry)Link (knot theory)MiniDiscGoodness of fitRow (database)Reading (process)Direction (geometry)Electronic program guideHand fanSingle-precision floating-point formatFocus (optics)Perfect groupExpandierender GraphCountingPhysical lawComputer programmingFamilyLine (geometry)Interactive televisionComputer animation
13:44
TelecommunicationComputer animation
Transcript: English(auto-generated)
00:02
Good morning, I'm Francesca Marcello and today I'm presenting a research work focused on smart building energy and comfort management based on sensor activity recognition and prediction. After a brief introduction about smart buildings and some of the key points that characterized them, there will be
00:23
the explanation of the problem under investigation and the presentation of the system the TR is proposed as a solution. In particular, there are three main elements of this system that will be explained in details, which are a model for the recognition of activity usually performed by users inside their home.
00:43
The second is a model for the prediction of these activities once they are known by the system. And the last one is about the scheduling of the appliances of the home according to user habits. Then there will be the presentation of the simulation that were done and the presentation of the results obtained.
01:05
And last, there are going to be the conclusions and the future works that are going to be addressed in future. Smart buildings are characterized by the presence of sensor actuators and smart devices. They give the opportunity to building energy and comfort management systems of providing support to users of various intelligent buildings.
01:31
In particular, these systems are focused on monitoring and controlling key equipment inside the buildings and understand the user behaviors with the aim of providing users with tools that support the cost effective solutions in appliance management.
01:49
To develop such kind of system, it is necessary to monitor users' habits, learning their preferences and predicting their sequences of performed activities and appliance usage during the day.
02:05
Most of the literature consider user comfort as a set of constraints on appliance usage, which are priori set based on general statistics. Furthermore, most studies that are focused on activity recognition and prediction don't consider a complete system for small building solutions.
02:29
The main idea is then to realize a system that is continuously monitoring user preference and habits about appliance usage thanks to a non-intrusive sensor network
02:40
so that it can automatically recognize the activity usually performed by users and predict future usage of appliances. In this way, the system can manage cost-saving appliance scheduling based on user preference and expected energy consumption. In this slide, an overview of the proposed system is presented and we can see the all different modules involved in the process.
03:07
First of all, sensors are used to make observations on the users and on their interaction with the surrounding environment. The Event Detection module combines the observation into events with meaningful information about activities done by the users.
03:28
The Activity Recognition module recognizes the activities thanks to the correlation between detected events and activities, while the User Profile module stores all the habits given by the recognition of the activities.
03:45
The information from User Profile and Recognition module are then used by the Activity Prediction module to predict the following activities that are expected to be carried out by the users. Next, the Appliance Scheduling module uses the information from all the other modules to find a scheduling
04:05
for controllable appliances that can guarantee the best trade-off between energy cost reduction and user comfort. More in details, the Activity Recognition approach encompasses two phases that are the training phase and the running phase.
04:26
During the training phase, the system learns the association between activities and sequences of detected events. For example, as we can see in the slide, let's suppose that we have a space of
04:42
events consisting in only three events and that we are considering two different activities, A1 and A2. Every time the activity A1 is performed, we have one different instance of that activity that depends on the sequence of detected events during the time window OA.
05:04
We then obtain a feature vector for every instance that, for each event, takes count of the time this specific event is observed with respect to the total number of events observed inside the window OA. Finally, we define a model vector for each activity, such that every feature corresponding to a
05:27
specific event is the average rate for all the observed instances associated with the same activity. Instead, during the running phase, the observed events are divided into subsequences of predefined length, equal to size w, and starting at time t.
05:51
The algorithm computes a feature vector with the rates of event occurrences. This vector is compared with the model vector obtained before from the training phase,
06:04
and it is then classified based on its probability to belong to a given activity. The main task of the prediction model is to provide a possible scenario in time t-aired in the future. The algorithm evaluates all the probabilities of the activities to be performed next in the future, thanks
06:25
to the statistics information about all the other activities carried out and recognized before the current time. The probability of each activity is then translated into the probability of one of the appliances of the house to be used.
06:44
This output is then used by the next appliance scheduling module to make a current scheduling of the appliances and to make evaluations about energy consumption.
07:01
The last module is then the appliance scheduling module. This algorithm dynamically shifts tasks of the controlled appliances to time when it is more convenient, like of peak times. It evaluates the scheduling finding the best trade-off between the overall energy cost and the annoyance experienced by the user due to this shift.
07:27
In the table are shown different activities that can be performed by the user, taken from a real-world dataset, and the corresponded appliances that are used during these activities.
07:40
Only three of the appliances among those present in the house are controlled for the scheduling and they are the water heater, the washing machine and the dishwasher. Four different activities are linked to these appliances and the scheduling algorithm performs the scheduling only when one or
08:01
more of these activities have a value of probability to be performed that is higher than a certain threshold. All simulations were done using the same dataset, considering six weeks of training and one week of testing.
08:24
The activity recognition algorithm has an overall accuracy of 82.3%, while the prediction algorithm gives an accuracy of 67%. For the evaluation of the scheduling algorithm, three different scenarios have been considered.
08:43
The first one is the classic situation where appliances are normally used by the resident and the scheduling is never programmed. The second one is based on a perfect knowledge of the time in which the user wants to use some of the appliances in the house. This case can be the one in which the user instructs the system about the moment they want the
09:07
appliances to start, but it has the disadvantage of requiring lots of iterations between the user and the system. The last one, instead, bases its scheduling evaluations on the probability of using any
09:27
of the appliances at time t, calculated as explained before for the prediction module. We've called this scenario the scheduling based on probability. In the table on the left are listed two different tariffs for energy pricing, with one that is more convenient during the off-peak times.
09:48
While in the table on the right it is possible to see how the two systems with the scheduling affect energy consumption that can be decreased by at least 39.2%, in contrast to the case without scheduling.
10:08
The graph in this slide shows the energy saving over the week, differentiated by the three appliances that we are taking into account, and considering about the scheduling based on probability and the scheduling based on the knowledge of the preferred time for the user.
10:27
It is evident how most of the savings come from a wiser use of the water heater, thanks to a scheduling only at the appropriate time, while there is a lower incidence from the washing machine and the dishwasher.
10:42
This can be explained by the fact that the preferred times for the use of these appliances are already evaluated as the best compromise between energy consumption and user comfort. Especially because in most cases they are used in times very distant from the periods of non-peak hours.
11:09
We also evaluate the annoyance caused to the users by the scheduling of the appliances. Values of annoyance go from 1 to 5, where 1 indicates that there is not any discomfort for the user in the
11:30
change of the time in which the appliance was turned on, while 5 indicates the highest level of annoyance for the user. We have found that in most cases the level of annoyance was always very low. There is only one
11:46
exception in the case where the scheduling is based on probability, where there is a higher level of annoyance in the use of the water heater due to the fact that this appliance is linked to two different activities.
12:08
One of these is the activity of washing the dishes that is recognized and predicted with some problems because it is often confused with another activity that the user performs.
12:23
So there are some errors in the prediction of this activity that led to a less accurate scheduling and to a rise in the level of annoyance. In conclusion, results obtained with the proposed system show that the appliance's scheduling can guarantee
12:40
energy saving and reduce consumption in comparison with a classic use of energy and appliances. The prediction model permitted quite accurate scheduling, even for several hours ahead in the future, basing only on probabilities. Furthermore, it was possible to guarantee that the annoyance rate was never too high to respect user comfort.
13:11
FutureWorks will focus on testing the adaptability of the system to defend real-case scenarios and on improving the prediction algorithm.
13:20
Also, it will be evaluated how the presence of renewable energy sources could affect appliance's scheduling and improving energy savings. Finally, the system could be expanded considering information about user's health and finding a correlation between right or wrong habits and user psychophysical health.