A COMPUTER IMPLEMENTED METHOD FOR DEMAND DISAGGREGATION OF A POWER CONSUMPTION CURVE

Abstract

The present invention is related to a computer implemented method for demand disaggregation of a power consumption curve. The consumption curve, being the result of the use of a plurality of appliances operating in an overlapping manner where each of the appliances may be operating several times a day, start operating at different times and under different operating conditions. The present invention is a method that allows to extract very useful information from a single curve since it identifies which appliance is on at each instant of the day and allows to determine the consumption pattern of a user. The method, according to various embodiments, makes it possible to obtain the result with a reduced and affordable computational cost.

Claims

1. A computer implemented method for demand disaggregation of a power consumption curve E(t) of a user, the method comprising: obtaining a measurement, from a measurement instrument, of the power consumption curve E(t) of the user, the user having a plurality of appliances contributing to the consumption measured by the measurement instrument, the measurement obtained for a predetermined period of time T, wherein the period of time T is discretized giving rise to a discrete power consumption curve E.sub.m(t) in m predetermined subintervals; determining a set of appliances (a) that are susceptible to cause consumption power; for each of the appliances, obtaining a signature (f) of its consumption (h.sub.a,1, h.sub.a,2, . . . , h.sub.a,n.sub.a), a vector of n.sub.a coordinates with n.sub.am, being h.sub.a,i the power consumption of the appliance (a) at time subinterval i, being the power consumption of the appliance (a) measured from the moment the appliance (a) is switched-on during subintervals 1 to n.sub.a, being n.sub.a the last subinterval in that the appliance (a) is working; for each of the appliances (a) generating a set of powerlets, w, of the form
w=(.sub.a, . . . ,.sub.a,.sup.k.sup.0.sup.)h.sub.a,1,h.sub.a,2, . . . ,h.sub.a,n.sub.a,.sub.a, . . . ,.sub.a) being a powerlet a vector comprising at each of the coordinates values of power consumption in a plurality of subintervals for the period of time T; the powerlet comprising the values of the signature (f) of the appliance (a) located from a certain coordinate k.sub.0 of the powerlet within the time period T and with a value of energy .sub.a consumed in the remaining coordinates, and wherein each powerlet is generated for the signature (f) located at a different position within said powerlet; determining a combination of powerlets among the set of powerlets w fulfilling that $E_m\left(t\right)-{.Math.}_{i=1}^n{w}_i$ is positive and is minimum, being n the number of selected powerlets w.sub.i of said combination; outputting at least one identification of the appliance/appliances of the user associated to the combination of powerlets determined in the previous step and the power consumption of said appliance in the period of time T.

2. The method according to claim 1 wherein the selected powerlets generated for the same appliance (a) are restricted to those fulfilling that are orthogonal.

3. The method according to claim 1, wherein the selected powerlets are restricted to those wherein the number of powerlets corresponding to a predetermined appliance (a) is less than a predetermined frequency number.

4. The method according to claim 1, wherein a powerlet w.sub.i is discarded among the selected powerlets if E.sub.m(t)<w.sub.i at any of its coordinates.

5. The method according to claim 1, wherein, when generating the set of powerlets, the method further comprises the following steps: determining a probability density function of the appliance (a) which establishes the probability that the appliance (a) is switched on; determining a probability threshold value; determining those time subintervals in which the probability of the appliance (a) being switched on is greater than the probability threshold value; generating a powerlet (w) for each subinterval fulfilling the previous condition, wherein the signature (f) of the powerlet is located at said subinterval.

6. The method according to claim 1, wherein for determining the combination of powerlets, the method further comprises the following steps: determining a probability density function of the appliance (a) which establishes the probability that the appliance (a) is switched on; determining a probability threshold value; determining those time subintervals in which the probability of the appliance (a) being switched on is greater than the probability threshold value; selecting a powerlet (w) for each subinterval fulfilling the previous condition, wherein the signature (f) of the powerlet is located at the subinterval.

7. The method according to claim 5, wherein the probability density function is determined by: determining a set of time instants in the time interval T wherein the appliance (a) is switched on; determining a normal distribution for each time instant with the mean value at the time instant and with a predetermined variance; providing the probability function as the mixture of Gaussian distributions being normalized.

8. The method according to claim 5, wherein the probability density function depends on contextual parameters: the period between consecutive uses of an appliance (a); or the period between the use of an appliance (a) and the use of other appliance (b); or the day of the week; or the month of the year; or the festive days; or the holidays; or the temperature; or the zip code of the house; or a combination of any one of the previous.

9. The method according to claim 5, wherein the probability density depends on a user consumption patterns or the probability density function depends on statistics of user consumption behaviors.

10. The method according to claim 1, wherein the signature (f) of the appliance (a) is measured with sensors.

11. The method according to claim 1, wherein the signature (f) of the appliance (a) is obtained from statistics.

12. The method according to claim 1, wherein the signatures (f) of the appliances (a) are stored in a database.

13. The method according to claim 1, wherein the step of determining the combination of powerlets among the whole set of generated powerlets w fulfilling that $E_m\left(t\right)-{.Math.}_{i=1}^n{w}_i$ is positive and is minimum and is performed gradually, so that the method minimizes at least a first power consumption curve E.sub.1 and subsequently a remaining power consumption curve E.sub.2(t): $E_1=E_m\left(t\right)-{.Math.}_{i=1}^j{w}_{a_i}$ $E_2=E_1-{.Math.}_{i=j+1}^n{w}_{b_i}$ and wherein; 1<j<n; w.sub.a are the powerlets associated to a predetermined first set of appliances (a); w.sub.b are the powerlets associated to the set of remaining appliances (a) resulting from discarding the first set of appliances (a); and the second index i in w.sub.ai and w.sub.bi is the time index.

14. The method according to claim 13, wherein: appliances with powerlets w.sub.ai comprise a refrigerator and standby; and appliances with powerlets w.sub.bi comprise at least one of a dishwasher, a cloth dryer, a washing machine, an oven, a kitchen robot, a ceramic hob, a water heater, an electric space heater or an air-conditioning, and further appliances with powerlets w.sub.ci comprise at least one of a microwave, a television, an iron, a coffee maker, a kettle, a toaster, a stove, a lighting device, a vacuum, a hair appliance, video game console or a computer.

15. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out a method according to the steps of claim 1.

Description

DESCRIPTION OF THE DRAWINGS

[0130] These and other features and advantages of the invention will be seen more clearly from the following detailed description of a preferred embodiment provided only by way of illustrative and non-limiting example in reference to the attached drawings.

[0131] FIG. 1 This figure shows the power consumption curve of a real case wherein E.sub.1, E.sub.2, E.sub.3 and E.sub.4 are individual appliances causing abrupt changes in the function.

[0132] FIG. 2 This figure shows schematically the content of a signature stored in a vector for an specific appliance.

[0133] FIG. 3, 4 This figure shows schematically the content of a different signature used for generating two different powerlets. The two powerlets are shown in FIG. 4.

[0134] FIG. 5 This figure shows a specific example where a signature with a higher resolution is used for generating different firms already assigned to a powerlet, these signatures associated with different time instants of start-up of the appliance.

[0135] FIG. 6 This figure shows a process in which, starting from a very large set of powerlets generated for a set of signatures, criteria are progressively established that limit the powerlets selected to establish the search space to be used in the optimization process.

[0136] FIG. 7 This figure shows a specific embodiment wherein the generation of powerlets are limited to those wherein the start-up of the appliance is located on time coordinates wherein a probability density function is higher than a predetermined threshold.

DETAILED DESCRIPTION OF THE INVENTION

[0137] As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or computer program product. According to a first aspect of the invention, an example of disaggregation of the power consumption curve of a given user is described resulting in: the identification of a set of appliances responsible for the consumption that has given rise to the power consumption curve, the instant at which each of the appliances have been switched on and, the consumption signature for each of said appliances.

[0138] In this way, the inverse problem of determining which devices have contributed with their individual consumption to the power consumption curve is posed, starting from the reading of the power curve of a given user without having physical access to the devices at his disposal and without having precise information of the moments in which the user switch on each of the devices and which ones.

[0139] FIG. 1 shows a power consumption curve showing abrupt changes due to switching on or off of different appliances. The same curve may be located higher or lower, for example, depending on whether there are simultaneously devices that have a continuous consumption. This is the case of devices that instead of being switched off completely go into a standby state.

[0140] The curve depicted in FIG. 1 is usually recorded because the measurement equipment samples the power consumption signal or variable with a sampling rate sufficient to show significant changes even within a one-minute interval. This function with a high resolution can be transformed into a function with a lower resolution, with a coarse discretization, for example by imposing that the area of the second low resolution curve maintains the area of the first high resolution curve and therefore represents the same overall consumption.

[0141] The second input data used by the method is a set of power consumption signatures for a plurality of appliances. Although the word appliance is used, it is understood that this is any device connected to the network that is monitored to give rise to the power consumption curve. That is, an electric vehicle plugged in to charge the battery is also an appliance in the context of the invention.

[0142] The function depicted in FIG. 1 is usually acquired because the measurement equipment samples the power consumption variable with a sampling rate sufficient to show significant changes even within a one-minute interval. This function with a high resolution can be transformed into a function with a lower resolution, for example by requiring that the area of the second best resolution curve maintains the area and therefore represents the same overall consumption.

[0143] The second input data used by the method is a set of power consumption signatures for a plurality of appliances. Although the word appliance is used, it is understood that this is any device connected to the network that is monitored to give rise to the power consumption curve. That is, an electric vehicle plugged in to charge the battery is also an appliance in the context of the invention.

[0144] A consumption signature is to be defined as a vector in which, for each coordinate, contains the value of the power consumption of the device in a set time interval. For example, if the signature sets a sampling period every 15 minutes and the appliance remains on for three and a half hours from the time it is switched on then the vector representing the signature will have dimension 14 which is the result of dividing the time it remains on (3.5 h) and the duration of the interval (15 min=0.25 h). In this case each coordinate has the value of the power consumption of the device along the 15 min corresponding to said coordinate.

[0145] FIG. 2 shows a schematic representation of the signature of an appliance according to an operational mode. That is, the same appliance can run for three hours if it operates in a certain operating mode (e.g. a washing machine operating in a certain washing program) or two hours in a different program (the same washing machine operating in a less demanding program). In addition to having a different duration, the operating pattern is different. For example, the most demanding program can raise the washing temperature by activating the heating elements that have a high consumption and that this consumption is at the beginning, after having filled the water tank, turning off at the moment of having reached the operating temperature.

[0146] FIG. 2 shows schematically a consumption signature for a given household appliance, a washing machine as shown in the left side. The horizontal line corresponds to the time where each coordinate is represented by a certain width, the width of the time interval in which the accumulated power consumption is measured.

[0147] The value of each vertical bar has a height proportional to the consumption of the appliance in that interval. Although the vector is represented by a union of sub-intervals, it can be stored in a computational system as a vector of length n.sub.a, where n.sub.a is the dimension of the vector, i.e. (h.sub.a1, h.sub.a2, . . . , h.sub.an).

[0148] The next stage of the method is the generation of powerlets from appliance signatures.

[0149] The powerlet is a vector, denoted as w throughout this description, showing the consumption of a given appliance when it has been operating with a consumption according to a given signature. This consumption could have been initiated at a certain instant of time. When the same appliance is started at different time instants then two different powerlets are generated. These different powerlets will show the consumption signature at different time intervals and without overlapping since the sum of one and the other powerlet would give rise to an incongruence since the same machine cannot be started when it has not yet finished operating.

[0150] The simplest way to generate powerlets is to use vectors whose coordinates represent the same time period as the signature coordinates. In this case the generation of a powerlet from a signature consists of an operation of assigning each of the coordinates of the signature to successive coordinates of the powerlet starting at the instant when the appliance is switched on.

[0151] FIGS. 3 and 4 show this assignment operation. FIG. 3 shows schematically the power consumption signature vector of a given appliance, a coffee maker.

[0152] FIG. 4 shows two possible powerlets generated from the same power consumption signature.

[0153] A first powerlet generated with a time instant when the appliance is switched on near the end of the period T covered by the powerlet and a second powerlet where the appliance is switched on at an earlier time instant.

[0154] FIG. 5 shows an example of how to combine appliance signatures generated with a temporal resolution greater than the temporal resolution of the powerlet.

[0155] The left side of the figure shows the signature of an appliance with a temporal resolution of 15 minutes. In this example the time vector has been extended even if there are zero-valued coordinates to facilitate the understanding of the process by making use of vectors having the same length. Thus, it is easier to compare each of the vectors showing the same signature shifted at different time instants. That is, with the 15-minute resolution, the same appliance will show its behavior pattern shifted in time depending on when it starts to start up.

[0156] If the powerlet does not have the 15-minute resolution, it seems a priori that it is not possible under this lower resolution to distinguish at what time the appliance starts up.

[0157] However, it can be seen that a powerlet with a resolution of one hour will show the cumulative values shown on the right. The same signature with a resolution of 15 minutes shifted every 15 minutes results in 5 distinct 1-hour signatures. Thus, by using a larger number of signatures at lower resolution, it is possible to identify not only the signature of the appliance but also the instant of time at which it was started up with a higher resolution than that allowed by the powerlet.

[0158] The right side of the figure shows the four signatures with a resolution of 1 hour already assigned in the powerlet at the same initial time instant t=8. The signature will therefore be a vector of dimension two taking the values of the coordinates shown at t=8 and t=9.

[0159] The power consumption curve E(t) is the result of adding up the consumption of the actual appliances that have been activated at certain points in time.

[0160] The powerlets are vectors that serve as a basis to explain the consumption of the consumption curve E(t) in such a way that, in the ideal case, the powerlets correspond in a real way to the consumptions of each appliance both because the signature from which they have been generated is the signature of the appliance and because the location of the signature in the powerlet corresponds to the exact instant of time the appliance has been switched on.

[0161] However, in practice this ideal case does not happen, since the signature may be an approximation of the real signature, the resolution of the signature is low or, the instant of time used during the generation of the powerlet is not exactly the same.

[0162] Therefore, the minimization process will identify those powerlets among the set of generated powerlets that provides a better approximation of the measured power consumption curve.

[0163] For each signature, it would be possible to generate as many powerlets as coordinates are available in the powerlet for representing a different instant for switching on the appliance. However, the number of combinations would be so high that the result may require too much computational cost or even be unaffordable.

[0164] This situation is schematically represented in FIG. 6 (a) wherein a specific signature is shown at many locations of the coordinates of a powerlet filling all the gaps. The total number of powerlets generated with a single signature will be very high and all of them would be part of the search space in the minimization process making the process highly computationally intensive.

[0165] Therefore, according to several embodiments of the invention, the generation process of powerlets limits the number of generated powerlets to those that are most likely to be involved in a combination that minimizes the objective function. In this embodiment and any further embodiment, the restriction of reduction of powerlets may be applied during the generation process causing a reduction of the generated set of powerlets or, in a subsequent selection process.

[0166] Step (b) shows a first restriction based on feasibility. According to this criterion, only those powerlets that correspond to viable situations are maintained. For example, an orthogonality condition between two powerlets for the same appliance is equivalent to requiring that an appliance cannot be started until a previous process has been completed for the same appliance. This is a non-overlapping condition. It is possible to include additional conditions such as not combining an air conditioner and a heating appliance because it is highly unlikely that both will be switched on simultaneously with opposite functions.

[0167] This same figure shows a stage (c) in which a probability density function is first established so that only those powerlets that verify that they are located in time coordinates where the probability density function is higher than a certain threshold prevail.

[0168] Therefore, according to other examples of realization, although the number of powerlets is high, they reduce the computational cost in the process of selecting the powerlets to be combined by discarding powerlets even if they have been generated.

[0169] According to the example of step (c), it is shown how the number of powerlets to be generated is reduced is based on the use of probability density functions that represent the probability that at a given instant of time a given appliance will be switched on.

[0170] The information that allows determining which time instants are more likely that a certain appliance will be switched on can come from statistics obtained based on a user profile, from previous measurements or even because the user wanted to share information requested through a mobile application.

[0171] If, for example, for a given household appliance it is known that there are three-time instants in which it is likely to be switched on, the method makes use of three normal or Gaussian probability density functions whose mean is located at those time instants and, with a predetermined variance.

[0172] The sum of the probability density functions is a new function that, once normalized, becomes the probability density function associated with the appliance in question.

[0173] As it is shown in FIG. 7, the method also requires a threshold value such that all those sub-intervals of time in which the probability density function is greater than this threshold are sub-intervals that will give rise to a new powerlet in which the start of the signature of the appliance is located. Likewise, all those intervals below the threshold value will be sub-intervals in which the start of the appliance signature is not located, thus reducing the number of powerlets generated per appliance.

[0174] The significant reduction in computational cost and memory consumption in the computational system responsible for executing the method lies in the fact that the number of powerlets generated is restricted to only those where the signature starts at powerlet coordinates where the probability density function exceeds the threshold value.

[0175] Raising the threshold value therefore reduces the total number of powerlets generated and, lowering the threshold value increases the number of powerlets as shown in FIG. 7 where the cut-off of the threshold value occurs at values closer to the base of each of the bells and therefore resulting in wider valid intervals for placing the signatures.

[0176] According to an embodiment, the probability density function is the one closest in time. A further embodiment extend the information to those available in the historical record where those probability density functions that are closer in time, or those that meet an additional criterion that makes the pattern of behavior closer, are more important. For example, that it is the same day of the week.

[0177] In either case, the result is a restricted search space that results in a much less computationally and memory costly minimization process.