FR3044139A1 - METHOD FOR DETERMINING THE RESIDUAL DEMAND ADDRESSED TO A POWER GENERATION PARK - Google Patents

Abstract

Method for determining a residual demand defined by the relation DR (t, T) = Conso (t, T) - PV (t, T) - eol (t, T), where DR (t, T) is the demand residual, Conso (t, T) is a representative component of an electrical consumption, PV (t, T) is a photovoltaic component representative of a photovoltaic production and eol (t, T) is a representative wind component of a production wind turbine. The method comprises: - a determination step (S2) during which, for one or more times t and for the instants T, a forecast term and an error term of each component are determined from a process autoregressive system in which, and - a prior calibration stage (S1) during which at least prediction and error parameters of the autoregressive processes are determined and, for the forecasting and error processes, a transformation providing the prediction term from the forecasting process and a transformation providing the error term from the error process.

Description

Method for determining the residual demand addressed to an electricity generating fleet

The present invention relates to the determination of the residual demand of a power generation park.

In a known manner, the electricity generation parks are controlled to the production so that the electricity produced and made available by the park corresponds ideally at every moment exactly to the quantity of electricity drawn to the park by the various consumers who there are related.

The development of photovoltaic and wind power generation forms profoundly affects the maintenance of equilibrium of an electric generating fleet that includes such means of production. Indeed, the wind and photovoltaic productions are by construction little controllable since they depend strongly on uncontrollable external phenomena such as wind and sunshine. By satisfying part of the energy demand addressed to the parks, however, they profoundly affect the level, the structure and the variability of the demand which is addressed to the historical production means of the thermal power plants (nuclear, coal, gas,. ...), hydroelectric power plants, etc., which coexist with photovoltaic and wind farms within the same parks.

They also introduce greater uncertainty in the monitoring of the park load due to the low predictability of the associated meteorological hazards, this uncertainty being in addition to the sensitivity of the electricity consumption to the meteorological conditions.

In this context, the residual electricity demand addressed to a generating fleet, which can be defined as the demand for electricity addressed to this fleet at a given time less what is made available by photovoltaic and wind power, is made However, this is a critical aspect of fleet management given this underlying principle of balancing the park's electricity demand with the park's electricity generation.

It is therefore conceivable that it is necessary to have a tool for determining this residual demand that is reliable and promotes control of the controllable portion of electricity production within a power generation park.

Also, the invention relates to a method for determining a residual electricity demand of a power generation park serving points of consumption, said residual demand being viewed at a time t for at least one time T belonging to at a given horizon with respect to time t, the residual demand being defined by the relation DR (t, T) = Conso (t, T) - PV (t, T) - eol (t, T), where DR (t, T) is the residual demand, Conso (t, T) is a representative component of an electrical consumption associated with the consumption points and seen at time t for the said at least one instant T, PV (t, T ) is a photovoltaic component representative of a photovoltaic production within said electricity generation park and seen at time t for said at least one instant T and eol (t, T) is a wind component representative of a production wind turbine within said electricity generation park and seen at the moment t for said ego ns a moment T, each component being modeled via the relation P (t, T) = Pprev (t, T) + e (t, T), where P (t, T) is representative of the considered component, Pprev (t , T) is a predictor term of the component and e (t, T) is a prediction error term of the considered component, the method being implemented by computer means and comprising: - a determination step at during which, for one or more instants t and for the instants T, each term of each component is determined from an autoregressive process comprising a prediction process representative of the forecast term and an error process representative of the term d error, in which: o the process of the forecast term seen from a moment t for a moment T is expressed as a function of the process of the forecast term seen from a moment t-1 for the moment T, d a Tt-based prediction parameter and a forecast residual, and o the error term process seen from a moment t for a moment T is expressed as a function of the process of the error term seen from the instant t for a moment T-1, of an error parameter function of Tt and of a residual of error, - a preliminary calibration step during which, from data of realization of the components recorded for a first period on the one hand and on the other hand, of data of forecasting and data of realization of the components for a second period on the other hand, at least the prediction and error parameters of the autoregressive processes are determined and, for the forecasting and error processes, a transformation providing the forecast term from the forecasting process and a transformation providing the error term from the error process.

According to one aspect of the invention, the autoregressive process is defined by:

where /? Prev and βen are the forecast and error parameters respectively, sprev and serr are the prediction and error residuals respectively, uprev and ue are the prediction or error processes representative of the forecast term Pprev respectively of the error term e of the corresponding component.

According to one aspect of the invention, during the calibration step, for the different components of the residual demand: a climatic average is determined towards which the average of the corresponding forecast term tends and a climatic uncertainty towards which the standard deviation of the corresponding error term from the output data over the first period, - a function representative of the variation of the standard deviation of an intermediate centric forecasting process determined from the forecast term is determined, of the climatic average and climatic uncertainty, said function being dependent on Tt and being determined from the production data, and - the parameters of prediction of the autoregressive processes are determined from the forecast data, from the climate average, from the climate uncertainty and said function.

According to one aspect of the invention, the transformation of a forecasting process into a forecast term is written as:

Pprev (t, T) - Pclim (T) + d (Tt) Ociim (T) uprev (t, T),

where Pciim (T) is the climate average, d is the said function and aCiim (T) is the climatic uncertainty.

According to one aspect of the invention, during the calibration stage, for the different components of the residual demand: error data are determined from the production data and the forecast data for the second period, a function representative of the variation of the standard deviation of an intermediate error centric process determined from the error term and the climatic uncertainty, said function being dependent on Tt and being determined from the data of error and climatic uncertainty, and the error parameters of the autoregressive processes are determined from the error data, the climatic uncertainty and the said representative function of the variation of the standard deviation of the intermediate process of centered error.

According to one aspect of the invention, the transformation of an error process into an error term is written as: e (t, T) = aciim (T) c (Tt) ue (t, T) , where Ociim (T) is the climatic uncertainty and c is the representative function of the variation of the standard deviation of the intermediate error process (ve) centered.

According to one aspect of the invention, for at least one component, said function representative of the variation of the standard deviation of the intermediate error centered process depends on Tt and the forecasting process.

According to one aspect of the invention, the forecasting residues are correlated with each other, and, during the calibration step, for each component, an autocorrelation matrix of the forecasting residues is determined between the instants T.

According to one aspect of the invention, during the determination step, for each component, one or more correlated forecast residues are generated at different times t from the corresponding autocorrelation matrix, and the forecasting process is determined. corresponding from said forecast residues and the corresponding forecast parameter.

According to one aspect of the invention, the error residues of a component are independent of each other, and during the determination step, error residues are generated and the error process is determined. and the process of predicting each component from the generated error residues and the corresponding error parameter.

According to one aspect of the invention, during the calibration step, the production data for the first period associated with the wind component and the photovoltaic component are initially transformed, and the forecast and implementation data for the second phase period associated with the wind component and the photovoltaic component through a logistical transformation.

According to one aspect of the invention, during the determination step, the photovoltaic and wind components are determined by applying an inverse logistic transformation determined from said logistic transformation to the sum of the error term and the corresponding forecast.

According to one aspect of the invention, the method further comprises a step in which, from the residual demand determined during the determination step, a planning of the operation of at least one production facility is constructed. electricity of said park, and pilot said installation according to said planning.

According to one aspect of the invention, said planning comprises at least a decision to stop or start the power generation installation, or a production instruction by said installation of a given power level for a given period of time. . The invention also relates to a computer program containing instructions for implementing the method defined above when executed by a processor. The invention will be better understood on reading the detailed description which follows, given solely by way of example and with reference to the appended figures, in which: FIG. 1 is a schematic illustration of a production fleet of electricity considered in the context of the invention; - Figure 2 is a schematic illustration of a curve illustrating electricity consumption, photovoltaic generation, wind generation and residual demand associated with the park of Figure 1, - Figure 3 is a schematic illustration of component terms. residual demand; FIG. 4 illustrates a method according to the invention; FIG. 5 illustrates the progress of a stalling step of the method of FIG. 4; FIG. 6 illustrates a KPV parameter (T) determined during the process of FIG. 4; FIG. 7 illustrates the behavior of a prediction parameter determined during the process of FIG. 4; FIG. 8 illustrates the behavior of a load factor determined during the process of FIG. 4; FIG. 9 illustrates a correlation matrix determined during the process of FIG. 4; Figure 10 illustrates the variations in the standard deviation of an intermediate forecasting process as a function of the corresponding forecasting process; FIG. 11 illustrates the progress of a step of determining the method of FIG. 4; - Figure 12 illustrates an example of a forecast of the residual demand obtained; - Figure 13 illustrates the daily profiles used for the determination of the photovoltaic component; and FIG. 14 illustrates computer means for implementing the method of FIG. 4.

Figure 1 illustrates a park P of electricity production extending over an area 2. Zone 2 is for example a large area, such as for example an area covering an entire country.

Park P is configured to produce electricity and to convey this electricity to consumption points rep marked with triangles in FIG. 1.

Park P includes power generation facilities of several types. More specifically, the park P comprises: photovoltaic production facilities rated Pp, wind power facilities rated Pe, and so-called "conventional" production facilities, noted Pc.

The photovoltaic production facilities Pp comprise photovoltaic production devices configured to produce electricity by photoelectric effect.

Pe wind generation facilities include wind turbines configured to produce electricity under the influence of wind.

Conventional production facilities Pc correspond to means of electricity production not based on photovoltaic or wind power. Conventional production facilities Pc are controllable. They include nuclear facilities, coal and gas power plants, hydroelectric power stations, etc.

In addition, the park P includes a network R connecting the various production facilities to the consumption points Ç.

In the context of the invention, with reference to FIG. 2, the residual electricity demand addressed to the park P by the consumption points ((the net demand in FIG. 2) is defined as the difference between the electricity consumption generated. by the points of consumption C and the sum of the electricity productions of the photovoltaic installations Pp and the wind-powered installations Pe.

In addition, the residual demand is seen at a time t for a set of instants T belonging to a given horizon H with respect to time t.

In other words, we have the relationship; DR (t, T) = Conso (t, T) - PV (t, T) - eol (t, T), where DR (t, T) is the residual demand, Conso (t, T) is a representative component of the consumption of consumption points Ç at time T as seen from time t, PV (t, T) is a photovoltaic component representative of the photovoltaic production by photovoltaic installations Pp seen at time t for the moment T and eol (t, T) is a wind component representative of the electricity production of the wind power plants Pe within said electricity generation park and seen at the moment t for the moment T. The horizon H is for example counted in days from time t. For example, in the context of the invention, the horizon H is a few days, for example 3 days. Alternatively, the horizon H is longer and corresponds for example to 10 days.

As described in more detail below, one of the objects of the invention consists in determining, by fixing at a time t, and for at least one instant T subsequent to t, the residual demand from a forecast of this residual demand and of an error affecting this forecast, and proceeding from the decomposition of the residual demand into its Conso, PV and eol components. Advantageously, at given t, these quantities are determined by scanning the horizon H by incrementation (or decrementation) of T. In addition, t is itself varied so as to obtain information on the dynamic of the residual demand.

Subsequently, at DR (t, T) given, the objects DR (t + k, T) where k is a natural integer with t + k <T form representations of DR (t, T).

The principle of determination of the residual demand according to the invention is based on the modeling of each component of the residual demand DR, that is to say say components Conso, PV and eol, in two terms, including a forecast term Pprev and an error term e representative of the error of the forecast term and centered on the forecast term.

We thus have the following relations (1), (2) and (3): "Pconso (t, T) Pprevconso (t, T) + Cconso (t, T), (1) - Ppy (t, T) = Pprevpv (t, T) + epy (t, T), (2) Peol (fT) - Ppreveol (t, T) + eeol (t, T) (3).

Referring to Figure 3, for a given component of residual demand, it is noted that the means and standard deviations of the forecast and error terms converge to so-called climate averages and dispersions, which are values obtained from the absence of any forecast, and this with a set of communicating vessels in the respective dispersions of forecasting and error.

These values are for example obtained by a modeling including a transformation of meteorological data, in particular relating to the wind, the solar radiation and the temperature in data of wind, photovoltaic production and consumption production.

The average of the forecasts is equal to the climatic average noted Pciim (T) of the phenomenon. For a very short-term forecast, that is, when t tends to T, the dispersion, that is, the standard deviation, of the forecast term tends to the climatic uncertainty noted aciim ( T), and the dispersion of the error tends to zero. For a prediction beyond a few days, that is, when T is sufficiently far from t, the dispersion of the forecast term tends to zero (the forecast term tends to the climatic average Pciim (T)), and the dispersion of the error term tends towards climatic uncertainty

Oclim (T).

As described in more detail below, the principle of the invention is based on a transformation of the prediction and error terms into centered and reduced prediction and error processes from input data, a modeling error and prediction processes in autoregressive processes, and using the results obtained to determine each of the terms of the residual demand.

For a given component, the regressive process modeling of the forecast and error terms of the component is written in the form: oo where /? Prev and βen are prediction or error parameters, ερΚν and εβπ- are prediction and error residuals, uprev and ue are the predicted, respectively centered and reduced error processes representative of the forecast term Pprev, respectively of the error term e of the corresponding component.

The method of determining the residual demand according to the invention will now be described with reference to the figures.

With reference to FIG. 4, the method according to the invention comprises a calibration step S1 in which, from data of realization DR1 of the components recorded for

a first period Δ1, and DR2 performance data and DP2 forecast data of the components for a second period Δ2 on the other hand, at least the prediction and error parameters of the autoregressive processes are determined and, for forecast and error, a transformation providing the forecast term from the forecasting process and a transformation providing the error term from the error process.

It is noted that t and T are arbitrary with respect to these periods Δ1 and Δ2. In other words, t and T do not belong at first to these periods.

During a step S2, from the objects determined during the calibration step S1, it is determined, for one or more moments t and for at least one instant T of the horizon H considered, each term of each component . This determination is based on the following autoregressive processes: o

O

Figure 5 illustrates the details of the progress of the step S1.

During a step SI 1, one obtains the realization data DR1 and DR2, as well as the forecast data DP2.

The DR1 implementation data is a reconstruction of the electricity consumption, wind generation and photovoltaic production corresponding to the climatic conditions (temperature, wind, solar radiation) over the Δ1 period and the wind power and power consumption structures. and photovoltaic determined, for example current. This period Δ1 is advantageously a period of long duration, for example several tens of years. For example, Δ1 corresponds to a period of 50 years. These DR1 production data comprise for example a climatic series for each component, that is to say a series representative of the variation of the corresponding component as a function of the meteorological conditions. These series have for example a time step, that is to say that the step separating two values of the series is worth one hour.

With reference to relations (1), (2) and (3), these DR1 production data are representative of the quantities PCOnso, Ppv and Peoi as observed during the period Δ1.

The DR2 implementation data form a reconstruction of the history of electricity consumption, wind generation and photovoltaic production over the Δ2 period. This period Δ2 is advantageously a period of short duration compared to the period Δ1. For example, Δ2 corresponds to a period of 2 to 3 years. The implementation data DR2 comprises, for example, a climatic series for each component, that is to say a series representative of the variation of the corresponding component as a function of the meteorological conditions. These series also have, for example, also a time step, that is to say that the step separating two values of the series is worth one hour.

With reference to relations (1), (2) and (3), these DR2 production data are representative of the quantities PCOnso, Ppv and Peoi as observed during the period Δ2.

The DP2 forecast data is a forecast of electricity consumption, wind generation and photovoltaic generation over the Δ2 period. The DP2 forecast data includes a climate series for each component, i.e., a representative series of the variation of the forecast term Pprev of the corresponding component as a function of the meteorological conditions. These series also have, for example, also a time step, that is to say that the step separating two values of the series is worth one hour.

With reference to relations (1), (2) and (3), these DP2 forecast data are representative of the forecast term Pprcvconso- Pprevpv and Ppreveoi of Pconso? Ppv and Peoi as observed during the period Δ2.

The knowledge of the DR2 production data and the DP2 forecast data for the same period Δ2, that is to say data representative of the entire components (DR2) but also of their only forecasting term (DP2), makes it possible to conduct the stalling process of the step S1. Indeed, for a given moment, the difference between a production data item and a forecast data item is representative of the forecast error associated with the corresponding forecast.

The production data DR1 and DR2 are for example provided by the entity that supervises the production of the installations Pp, Pe and Pc. For example, the DP2 forecast data is also provided by this entity, in particular by the operational teams responsible for optimizing the operation of the Pp, Pe and Pc power generation facilities.

During a step S12, for the wind and photovoltaic components, a parameter Keoi (T), respectively KPV (T) of a logistic transformation is determined for application to the data DR1, DR2 and DP2 relating to the wind power generation and to photovoltaic production.

In fact, in the context of the invention, the photovoltaic and wind components are modeled in the form of FC load factors, denoted FCpv and FCeoi respectively, defined as the ratios between the observed production and the theoretical "maximum" production in laboratory conditions for photovoltaic production, respectively wind production. In other words, the data DR1, DR2 and DP2 relating to the photovoltaic and wind productions are expressed in the form of these load factors.

The load factors are between 0 and Ke0i, respectively ΚΡγ. For the purposes of the process according to the invention, the transformed load factors x are processed by means of the logistic transformation x = ln (FC / (K-FC), denoted Logit .The variable x is assumed to be Gaussian, which facilitates its determination The inverse transformation FC = K / (l + exp (-x)) denoted invLogit, makes it possible to ensure that the determined load factors remain bounded between 0 and K. It is thus avoided to obtain negative powers or higher. to the installed power.

In the course of step S12, the determination of the Keoi (T) or Kpv (T) parameters is performed from the determination, for each time step, of the maximum theoretical theoretical maximum theoretical wind production. In practice, this maximum theoretical output is determined from the knowledge of the maximum of the production data DR1 for each instant T and for each component.

For example, the parameter Keoi (T) is determined to be the constant function equal to 1. In addition, the parameter Kpv (T) has the profile shown in Figure 6, and depends on the day of the year in question.

During a step S13, the logistic transformation whose parameter K has been determined in the preceding step S12 is applied to the data DR1.

In other words, to the data DR1 relating to the photovoltaic component of the residual demand, the logistic transformation x = ln {FCPV / (KPV - FCPV)) is applied. To the data DR1 relating to the wind component of the residual demand, one applies the logistic transformation x = ln (FCeoi / ÇKeoi - FCeol)). We note these transformed data DR1 '.

In a step S14, from the transformed data DR1 'obtained at the end of the step S13 for the wind and photovoltaic components and the data DR1 for the consumption, for each component of the residual demand, the average is determined. climatic Pciim (T) and climatic uncertainty Gciim (T) corresponding.

These averages and uncertainties are functions of T, at each time step of the period Δ1 corresponding to a value of the climatic average and a value of the climatic uncertainty.

The respective climatic averages of the residual demand components are noted Pciimconso (T), Pciimeoi (T) and PciimPv (T) · The climatic uncertainties of the components of residual demand are noted Gciimc0nso (T), Gciimeoi (T) and GciimPV ( T).

For an example of determination, for example, Δ2 extends between July 1, 2009 and June 30, 2012. A daily step is applied to t. A time step is applied to Tt. The horizon H is taken equal to 14 days. Tt varies from 12 hours to 371 hours. At each date t, for a given component, for example the wind component, the climatic prediction is given by:

Pclimeol (T) = Ppreveol (t, T = t + 371).

In addition, it has the corresponding load factor FCeoi (T) realized, that is to say, for the period Δ1. This period Δ1 comprises for example 49 years (indexed by an integer a ranging from 1 to 49), and contains, for each year, 8760 time steps. For the determination of the climatic uncertainty at each T date, the standard deviation of the load factor carried out for each calendar date is determined for each sample of 49 values. The climatic uncertainty at time T is thus for example determined as the standard deviation of the charge factors transformed at T for a ranging from 1 to 49.

In a step S15, the logistic transformations determined during the step S12 are applied to the production data DR2 and the forecast data DP2.

In other words, to the data DR2 and DP2 relating to the photovoltaic component, the logistic transformation x = ln (FCPV / (KPV - FCPV) is applied to the data DR2 and DP2 relating to the wind component of the residual demand, applies the logistic transformation x = ln {FCeoi / (Keoi - FCeoi)). We note these transformed data DR2 'and DP2'.

In a step S16, the forecast data relating to the three components of residual demand is centered on and reduced by the uncertainties and climatic averages determined during step S15.

By center, we mean to transform the corresponding magnitude so that its average is zero. By reducing, we mean transforming the magnitude so that its standard deviation is equal to 1.

For the photovoltaic and wind turbine components, DP2 'transformed data obtained from the DP2 data is processed via the logistic transformation. For the consumption component, we proceed from DP2 data representative of the term consumption forecast.

In other words, from the uncertainties and climatic averages determined in step S14 and DP2 prediction data or DP2 transformed prediction data, variables VprevconsoCfT), prcvcoi (f L) and VprevPvCf are determined. T) respectively defined by the relations:

Vprevconso (t, T) - (Pprevconso (t, T) - Pclimconso (T)) / üclimconso (T),

Vpreveol (t, T) = (Ppreveol (t, T) -Pclimeol (T)) / Gciimeol (T), and VprevPV (t, T) = (Ppevpv (t, T) -PclimPv (T)) / GclimPv ( T).

The variables vprevConso (t, T), vpreveoi (t, T) and vprevpv (t, T) are intermediate prediction processes that are centered.

In order to reduce the variable vprev, for each component, we determine a function d representative of the standard deviation of the corresponding variable vprev as a function of Tt. This standard deviation is assumed to vary from 1 when t = T to 0 when T becomes large before t. We note these onset functions? dgoi ot dpv ·

For this determination, one proceeds by determining the value of the standard deviation as a function of Tt for several different values of Tt, and one constructs the corresponding function d by interpolation of the obtained values. For example, we use a linear or polynomial interpolation fitting best to the observed data, the function d (Tt) being a monotonic function decreasing in Tt, varying between 1 for Tt = 0 to 0 for T much greater than t, of Several weeks.

Advantageously, the function is bound to a given value for values of Tt beyond a predetermined value. For example, the function is limited to 0.1 for Tt greater than 300 hours.

Once the function d is obtained, we reduce the corresponding variable vprev by dividing it by the corresponding function d. In other words, we build the centered and reduced variables uprevconso, upreveoi and uprevpv from the relations:

Uprevconso (t, T) - Vprevconso (t, T) / dConso (Tt),

Upreveol (t, T) = Vpre \ eol (t, T) / deol (Tt), and - UprevPv (t, T) = VprevPv (t, T) / dpv (Tt).

In what follows, these variables are modeled by an autoregressive process (of order 1) expressed by the general relation (A):

This relationship (A) therefore declines for each component of residual demand. So we have the following relationships.

(Al)

During a step S18, for each component, the corresponding prediction parameter Pprev is determined from the corresponding relation (A1), (A2), (A3).

To do this, in a first realization, noting

we have the following relation because the process is autoregressive: a = E [uprev (t, T) uprev (tl, T)] - E [uprev (t, T)] E [uPrev (tl, T)] where E [ X] designates the expected value of the variable [X], which is estimated by the empirical average of the variable X on the sample made up of DR2 and DP2.

Since the uprev process is centrally constructed, we obtain the relation a = E [uprev (t, T) uprev (tl, T)]. From the knowledge of the terms uprev obtained following step S17, one thus determines a then Pprev for each interval Tt.

More precisely, ppicv is determined as a polynomial function of Tt approaching the results obtained.

Alternatively, in a second embodiment, pprev is determined from the analysis of the standard deviation of the forecast changes from one moment to the next, for example from one day to the next, and it is deduced therefrom at. This realization assumes that Pprev is equivalent to the standard deviation of uPrev (t, T) - uPrev (tl, T), which supposes that a is close to 1 and that Pprev is thus close to 0. In the same way as previously Pprev is determined as a polynomial function approaching the results obtained.

An example of the result obtained for the wind component is illustrated in Figure 7. Ppreveoii corresponds to the results provided by the first realization from which the polynomial function is built, and ppreveoi2 the results provided by the second embodiment, ppreveoimodei corresponds to the polynomial function selected .

Note that the two embodiments provide different results for large quantities Tt (for example greater than 100 hours), but similar for small quantities Tt.

It is also recalled that for a daily step chosen for t, PP, CV corresponds to an evolution of the forecast in 24 hours.

In a step S19, for each component, a correlation matrix of the prediction residuals sprev between them is determined.

In the context of the invention, the evolutions of the forecasts for different deadlines T are not independent. Figure 8 illustrates the predictions of the 24-hour charge factor FCeoi of the same day for different times t, in this case for the previous t of 3 days, 2 days and 1 day respectively.

It can be seen from this figure that the forecast curve moves in the same direction for the different times t, even if this evolution can be different for different time slots. This indicates that the eprev prediction residuals are not independent for different times T.

During this step S19, for each component, this matrix, denoted by Rconso (T, T), Reoi (T, T) and Rpy (T, T), is determined as a function of the component under consideration, from the prediction parameter βρκγ previously determined and DP2 forecast data for the consumption component or DP2 'transformed forecast data for the wind and photovoltaic components via the following relation:

This relationship makes it possible to calculate the sprev residues (t, T) found on the data sample DP2 for the consumption component or DP2 'for the wind and photovoltaic components. The matrices RCOnso (T, T), Reoi (T, T) and RPV (T, T) are the empirical correlation matrices estimated between the residues corresponding to the different instants T, for each of the components.

Advantageously, this correlation matrix is determined between the residues of the three components, making it possible to account for the dependencies between the forecast changes in consumption, wind generation and photovoltaic production.

Note that the βρΐΒν prediction parameters are representative of the link between the forecasts made at t and t + 1, the correlation matrix being representative of the link between the forecasts made at the same time t for the different instants T of the horizon H considered. .

Figure 9 provides an illustration of the matrix Reoi obtained for the wind component, for a horizon of 10 days. Correlations are strong around the diagonal, the dark band widening with the horizon. The correlation between forecasts made for two instants T close to a few hours is therefore strong, but becomes weak for two more distant instants T. For a horizon of 5 to 6 days, we see appear on the correlation matrix dark squares representative of the granularity of the day in the forecast: after 5 to 6 days, wind production forecasts are made "at the "day", the forecasts being identical for all the hours of the same day.

During a step SI 10, for each component of the residual demand, the error term econso, eeoi and epv is determined from the difference between the realization data DR2 and the forecast data DP2 of the component under consideration. The difference between these construction data and these forecast data constitutes error data, denoted DE2.

In other words, these error terms are determined from the relations (1) to (3), PCOnso, Ppv, Peoi corresponding to the production data DR2 (consumption component) or DR2 '(photovoltaic and wind turbine components), the terms Pprevconso , Pprevpv, and Ppreveoi corresponding to the forecast data DP2 (consumption component) or DP2 '(photovoltaic and wind components).

During a step SI 11, each error term determined during the step SI is reduced by dividing it by the climatic uncertainty Gciimconso (T), aciimeoi (T), aciimpv (T) corresponding.

In other words, the variables vec0nso, veeoi, vepv are determined from the relationships:

Veconso (t, T) = Gconso (t, T) / ^ climconso (T),

Veeol (t, T) - eeol (t, T) / Oclimeol (T), and vepv (t, T) = epy (t, T) / aciimPv (T).

The variables veConso, veeoi, vepv are intermediate error processes that are centered.

Advantageously, during this step SI 11, it is previously verified that the average of each error term is close to zero. If this were not the case, then we would be in the presence of a forecasting bias calling for an evolution of the forecasting models in order to correct it.

The variable ve is centered, but not reduced.

During a step SI 12, for each component, a function c representative of the variation of the standard deviation of the variable ve as a function of Tt is determined. In some embodiments, for at least one component, for example the wind component, this function c also depends on the corresponding uprev forecast variable. In other words, in this case, the forecast error also depends on the level of the forecast.

During this step, the evolution of the standard deviation of each variable ve is determined as a function of Tt. If the function also depends on uprev, this evolution is also determined by different classes of the associated uprev variable. The class of the uprev variable is a sub-sample of the observed data corresponding to an uprev variable included in a given interval, for example between 0 and 1, or for example between 1 and 2, etc. This evolution is determined from variables determined during step SI 11.

Figure 10 illustrates this evolution for the eeoi variable. There is a growth in the standard deviation of the forecast error eeoi as a function of Tt, the latter reaching 1 for a horizon of about 300 hours. In addition, verification of the communicating silt effect between the forecasting process and the error process is obtained. In other words, the function c evolves according to Tt inversely to the function d: for T close to t (small Tt), the function c is close to 0; for T much greater than t (Tt large) the function c is close to 1.

There is also a dependence between the standard deviation of the error and the forecast itself as illustrated by the class differences of the forecast variable. For Tt of the order of 3 days, this dependence remains ordered on the different curves, the standard deviation of Veeoi being all the more important as upreveoi is weak.

During this step, we try to retain a function c accounting for this double dependence on Tt and upreveoi, giving priority if necessary to the behavior of c for relatively low values of Tt. For example, we use a linear or polynomial interpolation fitting best to the observed data, the function c (Tt) being a monotonic increasing function in Tt, varying between 0 for Tt = 0 to 1 for T much greater than t, of Several weeks.

Once the function cconso, ceoi and cpv obtained for each component, the corresponding variable ue is reduced by dividing it by the corresponding function c. In other words, we construct the centered and reduced variables uec0nso, ueeoi and uepv from the relations:

Ueconso (t, T) - Veonso (t, T) / C conso (Tt), - Uee0l (t, T) = Veeol (t, T) / Ceol (Tt), and - Uepv (t, T) = Vepv (t, T) / Cpv (Tt).

These variables are therefore centered and reduced processes.

As previously stated, these processes are modeled as an autoregressive process of the form:

Unlike the uprev forecasting process, the error process does not establish a link between Terror at t and that at t-1. Indeed, between the two instants, one carries out a repr'vision, and it is possible not to preserve this possible coherence in t. Even if this would be desired, the available data do not allow it since the error measurement corresponds to the difference between what is achieved and what is expected during the Δ2 period. Now, between two instants t, only the forecast has changed. On the other hand, it is important to account for the behavior of Terror between two instants T, since it is this behavior which makes it possible to establish scenarios of error for a given forecast.

During a step SI 13, the error parameter β0ΙΤ is determined for each component of the residual demand. We note these parameters Perrconso, Perreoi and βεπρν ·

To do this, advantageously, it is considered that the error parameter perr corresponds to the standard deviation of the magnitude ue (t, T) - ue (t, Tl).

We then construct β0ΙΤ as a polynomial approximation of the result obtained, as a function of Tt. This approximation is between 0 and 1, without the condition of monotony. At the end of this step, the calibration step SI is complete, and all the objects necessary to determine the residual demand for different times t of a period different from the periods Δ1 and Δ2 are available.

Figure 11 illustrates the progress of the determination step S2.

During a step S21 of the determination step S2, for each component, eprev prediction residues are generated from the correlation matrix Rconso, Reoi, and corresponding Rpv for the different instants T of the horizon. considered. This step is for example carried out using conventional numerical methods of computer generation of random variables.

These residues are standard normal residues.

During a step S22, for each component, the forecasting process is determined and reduced uprev from the residuals, the prediction parameters βρΐΒν and the corresponding relation, by iterating on the various instants t:

In a step S23, from the centered and reduced process thus determined corresponding uprev, the non-centered and unrestrained process Pprev (t, T) is determined, that is, the forecast term of the corresponding component. , from the corresponding relation:

It is recalled that Pcum (T), d (Tt) and aciim (T) are obtained during the calibration stage SI for each component.

In a step S24, for each component of the residual demand, ee error residues are generated. Advantageously, these residues are generated as normal residues.

and independent. In other words, it is assumed that the errors of the different components are independent. However, alternatively, this hypothesis is not retained. In this context, a correlation matrix of the error residues is then jointly determined for the 3 components. This determination is for example conducted during the step SI.

During a step S25, for each component, the process centered and reduced ue is determined from these residues and the corresponding relation, by iterating on the instants T:

During a step S26, for each component, the error process e (t, T), that is to say the error term of the corresponding component, is determined from the centered and reduced process and of the corresponding relationship:

Advantageously, the steps S24 to S26 relating to the error process are repeated for each forecast Pprev (t, T), so that several error scenarios are obtained for each forecast.

During a step S27, each component of the residual demand is determined from the sum of the prediction term and the corresponding error term obtained via the preceding steps.

For the photovoltaic and wind components, the inverse logistic transformation invLogit is applied to this sum to go from the load factor to the component as such.

During a step S28, the residual demand is determined from the different components via the relation indicated above:

During each step S21 to S28, the determined objects are made by varying the time T so as to cover the horizon H considered. In addition, advantageously, the steps S21 to S28 are repeated, each repetition of the succession of these steps being carried out after incrementing the time t considered.

Figure 12 illustrates an example of a result obtained. In this Figure 12, the strong lines correspond to the prediction obtained, and the fine lines correspond to the error scenarios obtained. Note that in this Figure, each box has a width corresponding to a duration of 24 hours.

Advantageously, the processing of the photovoltaic component has certain peculiarities.

In particular, it is considered that each day presents a maximum value of photovoltaic production. The hourly load factor is obtained by multiplying this daily maximum value by a representative profile of the day, varying between 0 at night hours and 1 at the maximum hour of production corresponding to the meridional time. Also, we retain a single moment of forecast noted Tj for each day, which corresponds to the moment of this maximum value.

Figure 13 shows daily profiles for the PV load factor for two specific days of the year. These curves give the shape of the daily production of photovoltaic, relative to the maximum of the day. If the peak within a day is likely to vary, the shape of each profile is nevertheless very stable because it depends essentially on the hours of sunrise and sunset. The daily profiles are for example constructed from the DR1 observations, dividing for each day of the year the load factor curve observed by the maximum of the day.

Therefore, within a given day, we can have Tj-t <0 (for what is done at times after the instant Tj). In this case, the values of the parameters corresponding to Tj-t = 0 are retained.

Moreover, for the determination of the component, it is considered that the inverse invLogit transformation applied to the sum of the forecast term and the error term of this PV component provides the maximum value of the corresponding day.

In other words, we have the relation PVMax (t, Tj) = invLogit (Pprevpv (t, Tj) + e (t, Tj)), with i nvLogitK (x) = K (Tj) / (1 + exp (-x)).

The K parameter is tabulated daily. This parameter K appears in Figure 6.

We then move to the forecast hourly load curve of the component by applying the daily profile noted profilj (Th) to this maximum value via the relation; PV (t, Th) = PVMax (t, Tj) * profile (T h).

The profile of each day is also tabulated.

Advantageously, the treatment of the consumption component also has specificities.

In particular, the treatment of this component Conso takes into account the dependence of electricity consumption vis-à-vis the season of the year.

In other words, for at least a portion of the steps of the IF calibration step and / or the determination step, the processing of the consumption component is based on the consideration of this seasonal change in consumption.

In particular, during the calibration step S1, a correlation matrix R of the forecast residues for each season (spring PR, summer ET, autumn AU, winter HI) of the year is determined.

Moreover, the prediction parameter βρΐΒν (Tt) constructed is a function presenting a given form for each season.

The process according to the invention has many advantages.

In particular, it allows the elaboration of a fine forecast of the residual demand, this forecast also providing precise information on the possible evolution of the error which could be tainted the forecast obtained.

Also, advantageously, the residual demand thus determined is used as input data for controlling at least one power generation installation, in particular a so-called conventional Pc installation.

In other words, advantageously, during a step S3 (FIG. 4), a planning of the operation of an electricity generation installation is constructed according to the residual demand obtained at the conclusion of step S2. and piloting said installation according to the operation thus planned. In other words, it controls its operation to correspond to the constructed planning, itself generated from the residual demand determined as described above.

This planning advantageously comprises at least a decision to stop or start a controllable production plant, or a production setpoint of a given power level for a given duration. Advantageously, this planning includes both a decision to stop or start a controllable production plant and a production setpoint of a given power level for a given duration.

Advantageously, for controlling the one or more installations according to the determined residual demand, at least one control signal is generated for a control equipment of the one or more installations considered, this signal being representative of the constructed planning. The processing of this or these signals by the control equipment results in the implementation of the planning by the installation in question.

For example, for this purpose, the planning includes one or more instructions whose processing by a processor results in the generation of the control signal (s).

Thus, the operation of the installation is conducted according to the residual demand obtained.

Furthermore, the method according to the invention reflects the possible seasonality of the phenomena impacting the electricity generating fleet. Moreover, because of its variability at different times T, the level of uncertainty and the growth of information in the forecasts are reflected: forecasts for the day are more accurate than those for the next day, and forecasts for Large wind or photovoltaic productions are less accurate than for weak productions.

Moreover, the method results in a coherence of the forecasts at different times T, these being not independent of each other, in particular by virtue of the inertia of the meteorological phenomena at the origin of the production of electricity and consumption: temperature, radiation, wind, cloudiness, etc.

In addition, for a DR (t, T) prediction, its DR (t + 1, T) representation for the same instant T respects the statistical characteristics of the forecasting evolutions observed in reality.

In addition, forecasts of the three components of residual demand can be correlated with each other.

The method according to the invention is advantageously implemented by computer means illustrated in FIG. 14.

These means M comprise a processing unit UT comprising a processor CPU and a memory MEM.

The memory MEM contains a program or set of programs PRG including instructions whose execution by the CPU processor results in the implementation of the method according to the invention.

The computer means may further comprise an AFF display device and an SAI input device.

Claims (14)

1. A method for determining a residual demand (DR (t, T)) in electricity of a park (P) for generating electricity serving points of consumption (Ç), said residual demand being seen for a moment t for at least one instant T belonging to a given horizon with respect to time t, the residual demand being defined by the relation DR (t, T) = Conso (t, T) - PV (t, T) - eol (t, T), where DR (t, T) is the residual demand, Conso (t, T) is a component representative of an electrical consumption associated with consumption points (Ç) and seen at time t for the said at least one moment T, PV (t, T) is a photovoltaic component representative of a photovoltaic production within said electricity generation park and seen at time t for said at least one instant T and eol (t, T) is a wind component representative of a wind generation within said electricity generating (P) park and seen at stant t for said at least one instant T, each component being modeled via the relation P (t, T) = Pprev (t, T) + e (t, T), where P (t, T) is representative of the component considered, Pprev (t, T) is a predictor term of the component and e (t, T) is a prediction error term of the component, the method being implemented by computer means and comprising: a step determination method (S2) during which, for one or more times t and for the instants T, each term of each component is determined from an autoregressive process comprising a prediction process representative of the forecast term and a process of error representative of the error term, in which: the process (uprev (t, T)) of the forecast term seen from instant t for a moment T is expressed as a function of the process (uprev (tl, T Of the forecast term seen from a moment t-1 for the moment T, of a parameter of prediction (βρΓεν) functio n of Tt and a prediction residual (ερΓ6ν), and o the process (ue (t, T)) of the error term seen from a moment t for a moment T is expressed as a function of the process (eu (t, T-1)) of the error term seen from the instant t for a time T-1, of an error parameter (β6ΓΤ) function of Tt and of an error residue (eerr) , a prior calibration step (SI) during which, from production data (DR1) of the components recorded for a first period (Δ1) on the one hand and on the other hand, forecast data (DP2) and of data of realization (DR2) of the components for a second period (Δ2) on the other hand, one determines at least the parameters of forecast (βρΓ6ν (Τ-t)) and of error (βεΓΓ (Τ-t)) autoregressive processes and, for the forecasting and error processes, a transformation providing the forecast term (Pprev (t, T)) from the forecasting process (uPrev (t, T)) and a transformation urnissant the error term (e (t, T)) from the error process (ue (t, T)), the method further comprising a step in which, from the residual demand determined during in the determination step, a planning of the operation of at least one power plant of said park (P) is constructed, and said installation is piloted according to said planning. The method of claim 1, wherein the autoregressive process is defined by: where / Sprev and /? err are the forecast and error parameters respectively, ερΓβν and εατ are the forecast and error residuals respectively, uprev and ue are the prediction or error processes representative of the forecast term P, ^, respectively of the error term e of the corresponding component. 3. Method according to claim 1 or 2, wherein, during the calibration step (SI), for the various components of the residual demand: - a climatic average (Pciim (T)) is determined towards which the average of the corresponding forecast term (Ppiev (t, T)) and a climatic uncertainty (ocUm (T)) towards which the standard deviation of the error term (e (t, T)) corresponding from the data of the error realization (DR1) over the first period, - we determine a function (dconSo, deoi, dpV) representative of the variation of the standard deviation of an intermediate prediction process (v ^ v) centered determined from the term of prediction (Pprev (t, T)), of the climatic average (Pcum (T)) and of the climatic uncertainty (Ociim (T)), said function being dependent on Tt and being determined from the production data, and - prediction parameters for autoregressive processes are determined from the forecast data, the mean climatic Pciim (T), climate uncertainty Odim (T) and said function (dconso, deoi, dPV). The method of claim 3, wherein the transformation of a forecasting process into a forecast term is written as: Pprev (t, T) = Pdim (T) + d (Tt) Oclim (T) uprev (t, T), where Pciim (T) is the climate average, d is the said function and aciim (T) is the climatic uncertainty. 5. Method according to claim 3 or 4, wherein, during the calibration step, for the various components of the residual demand: error data are determined from the production data (DR2) and data from prediction (DP2) for the second period, we determine a function (cconso, ceoi, cPV) representative of the variation of the standard deviation of an intermediate error process (ve) centered determined from the error term (e (t, T)) and climatic uncertainty, said function depending on Tt (cconso, Ceoi, cPV) and being determined from the error data and the climatic uncertainty, and the parameters of error (perr (Tt)) of autoregressive processes from error data, climatic uncertainty and said function representative of the variation of the standard deviation of the intermediate error process (ve) centered. The method of claim 5, wherein the transformation of an error process into an error term is written as: where ocum (T) is the climatic uncertainty and c is the representative function of the variation of the standard deviation of the intermediate error process (ve) centered. The method according to claim 5 or 6, wherein, for at least one component, said function representative of the variation of the standard deviation of the intermediate error process (Veeoiou vepv) centered depends on Tt and the forecasting process. (upreVeoi (t, T), uprevpV (t, T)). 8. Method according to any one of the preceding claims, in which the forecasting residues (eprev (t, T)) are correlated with each other, and in which, during the calibration step (SI), for each component, it is determined an autocorrelation matrix (Rconso (T, T), Reol (T, T), RPV (T, T)) of the forecast residues (sprev (t, T)) between the instants T. 9. The method according to claim 8, wherein, during the determination step (S2), for each component, one or more forecast residues (eprev (t, T)) are generated which are correlated at different times t from the corresponding autocorrelation matrix, and determines the corresponding forecasting process (uprev (t, T)) from said forecast residues and the corresponding forecast parameter (Pprev (T-t)). The method according to any one of the preceding claims, wherein the error residues (8err (t, T)) of a component are independent of one another, and wherein, in the determination step (S2), error residues are generated and the error process (Ue (t, T)) and the forecasting process (uprev (t, T)) of each component are determined from the error residuals generated and the corresponding error parameter (Pprev (Tt), Petr (Tt)). 11. Method according to any one of the preceding claims, in which, during the calibration stage (SI), the production data for the first period associated with the wind component and the photovoltaic component are initially transformed, and the forecast and realization data for the second period associated with the wind component and the photovoltaic component through a logistic transformation. 12. The method according to claim 11, wherein, during the determining step (S2), the photovoltaic and wind components are determined by applying an inverse logistic transformation determined from said logistic transformation to the sum of the term d. error and the corresponding forecast term (Pprev (t, T) + e (t, T)). 13. A method according to any one of the preceding claims, wherein said planning comprises at least a decision to stop or start the power generation plant, or a production directive by said installation of a level of electricity. given power for a given duration. 14. Computer program comprising instructions for implementing the method according to one of the preceding claims, when the program (PRG) is executed by a processor (CPU).