PREDICTION OF ELECTRICAL POWER FLOWING THROUGH AN ELECTRICAL TRANSFORMATION STATION

Description

Domaine technique

[0001] The invention relates to the field of processing electrical power data transmitted at an electrical transformer station. One field of application of the invention is that of the forward planning of distribution networks, the aim of which is to anticipate constraints on the distribution network and, where appropriate, to implement proactive resolution levers. Consumption and production forecasts (active P and reactive Q) constitute essential input data for forward planning. Technique antérieure

[0002] A substation is an electrical structure located at the junction of high- and medium-voltage power lines. To transmit electrical energy throughout France, the voltage must first be reduced to adapt to the different needs of consumers; this is where a substation comes in, transforming high-voltage electricity into medium-voltage. It therefore plays a vital role in the overall electrical system. The substation includes transformers, monitoring, protection, and remote control equipment (for example, for tariff changes), energy metering equipment, and automatic load shedding systems to contribute to the safety of the electrical system.

[0003] In electricity, active energy refers to energy that is fully transformed into heat, light, or motive power. Reactive energy complements active energy in the operation of equipment such as transformers and motors by magnetizing the magnetic parts. All electrical networks distribute these two types of energy. However, the circulation of reactive power on networks leads, due to a higher current demand, to heating of the power cables, additional losses, significant voltage drops, overloads at the transformer level, and requires oversizing of the installations.

[0004] Forward planning of distribution networks is a process of optimizing network operation by anticipating constraints (in voltage and current) and levers to implement to remove them or reduce their impact. Forward planning is based on power forecasts (active only today, the reactive being deducted from the active forecast by applying a constant power factor) which pass through the source substations and on the active production forecasts of producers connected in medium and low voltage.

[0005] Constraints have an impact on the premature aging of the equipment constituting the distribution networks and certain deleterious effects for users that can go as far as cutting off their power supply. A power forecast integrating the reactive power allows for a better estimation of constraints. Furthermore, controlling reactive power flows in the forecasting process will allow us to consider the implementation of a new lever: the reactive lever. The paper "Forecasting Active and Reactive Power at

[0006] J.N. Figalgo's "Substations' Transformers" (IEEE, June 26, 2003) provides an overview of the prior art. Problème technique

[0007] Today, reactive power consumed / produced is modeled in forecasting tools using constant power factors. However, the behavior of reactive power flows has undergone significant changes, linked, among other things, to the evolution of uses connected to the network, the emergence of renewable energies, and the burial of power lines. The constant power factor assumption is therefore no longer appropriate.

[0008] Finally, Forecast Management requires power forecasts at local grids for which, as the swarming effect diminishes, the power curves are more difficult to predict. Indeed, forecasting at the local grid level must take into account a problem of greater spatio-temporal variability than at an aggregate level. Another source of variation at the source substations comes from load transfers during network reconfiguration and / or the activation of varmetric regulation automatons for reactive power regulation via capacitor banks. The load transfer situation implies the implementation of a network operating scheme known as "outside the normal operating scheme", a normal scheme corresponding to a network topology with minimal losses.

[0009] The challenge is therefore to be able to estimate / predict the flow of reactive power at a local grid point in the distribution network, excluding events such as outages linked to the activation of capacitor banks or load transfer, in order to be able to feed into forecast management calculations. Another challenge is the simultaneous processing of curves from thousands of source stations, in a high-performance and rapid manner. Exposé de l'invention

[0010] To this end, the invention relates to a method for processing electrical power data passing through an electrical transformer station of an electricity distribution network, comprising the following steps: a. Obtain active power and reactive power data measured at the said substation over an observation period; b. Identify, from the power data measured over the observation period, power data outside the normal operating pattern during the said observation period, by monitoring the variation over time of the reactive power and / or active power and by detecting jumps in the said variation; c. Identify at least one type of situation causing the identified power data outside the normal operating pattern; d. Process the power data outside the normal operating pattern Dhsne according to the type of situation identified to obtain corrected power data, and obtain normalized power data over the observation period; e.Establish, by learning from the corrected power data, a reactive power forecast model for said station in operation outside of a situation of the type of situation identified in step c).

[0011] The observation period can last from a few months to a few years. The power data obtained by measurement are historical power data that have passed through the transformer substation. A period outside the normal operating pattern is a period during which the power data are identified as being outside the normal operating pattern. The observation period can include one or more periods outside the normal operating pattern. A single type of situation or several types of situation can be at the origin of the identified power data outside the normal operating pattern.

[0012] Thus, thanks to this implementation, it is possible to predict the reactive power at a source station more accurately, said data processing method being implemented from historical electrical power data. The model obtained by learning from the corrected power data includes parameters adjusted so as to minimize the difference between the data estimated by the model and the corrected power data. Said processing method also makes it possible to establish a forecasting model circumventing the problem of spatio-temporal variability encountered at the level of a local mesh.

[0013] According to one embodiment, the situation causing the power data outside the normal operating pattern is a reactive power regulation or a load transfer to the electricity distribution network.

[0014] The step of identifying a type of situation allows the power data outside the normal operating pattern to be corrected according to the type of situation identified. Reactive power regulation and load transfer are situations linked to network operation and can significantly disrupt a normal operating pattern of the transformer substation. These types of situations occur during network interventions, incidents, works or other often unpredictable constraints on the network, during which the network manager must carry out a network reconfiguration. The network no longer operates according to a normal operating pattern and the reactive power measurements recovered at the source substations no longer correspond to the same users (consumers and producers) as before and are said to be "outside the normal operating pattern".These situations are considered to be “outside the normal operating pattern”, and should therefore not influence the reactive power forecasting model.

[0015] According to one embodiment, the step of identifying power data outside the normal operating pattern comprises the following sub-steps: b1. Obtain, from the power data measured over the observation period, for each of the active and reactive powers, a first envelope curve maximizing the power data measured over the observation period and a second envelope curve minorating the power data measured over the observation period; b2. Obtain, from the first and second envelope curves, a deviation curve by calculating a deviation between the first envelope curve and the second envelope curve; b3. Define a deviation interval, the upper limit of which corresponds to the deviation curve and the lower limit corresponds to the opposite of the deviation curve; b4. Obtain, from the first and second envelope curves, a derived curve by calculating a derivative of an average of the first and second envelope curves; b5.Identify two jumps over the observation period, the jump being defined by an instant when a value of the derived curve is outside the deviation interval, the jump being of the low type if the value is below the deviation interval and of the high type if the value is above the deviation interval; b6. Obtain, for each of the active and reactive powers, a curve by steps, a step being obtained by calculating the average of the power data measured over a time interval defined by two consecutive jumps, the step being of the low type if the first of the two jumps is low and of the high type if the first of the two jumps is high.

[0016] The step curve obtained at this stage of identifying power data outside the normal operating pattern makes it possible, on the one hand, to identify periods outside the normal operating pattern and, on the other hand, to more easily identify a type of situation associated with a period outside the normal operating pattern.

[0017] According to one embodiment, the situation causing the power data outside the normal operating pattern is a load transfer to the electricity distribution network.

[0018] According to this embodiment, steps of the same type occurring simultaneously on the step curves of each of the active and reactive powers are identified as load transfer steps.

[0019] Indeed, a load transfer at the transformer station produces a specific effect on the electrical power transmitted at the said station. This type of situation can be identified by studying the type of step on each of the active and reactive power step curves.

[0020] According to one embodiment, the step of processing power data outside the normal operating pattern comprises the following sub-steps: d1. Calculation of corrected powers Y t by a relation of the type: Y t = Z t + η / σ . σ Or Z t = X t − η x / σ x X t designating power data outside the normal operating pattern over a period outside the normal operating pattern during one of the load transfer stages, η x the average of X t, σ x the standard deviation of X t; η designating the average of the corrected powers obtained by calculation by linear interpolation between the average of the preceding stage and the average of the stage following said stage; σ designating the standard deviation of the corrected powers obtained by calculation by linear interpolation between the standard deviation of the preceding stage and the standard deviation of the stage following said stage; d2. Obtaining corrected active and reactive power data by substituting said power data outside the normal operating pattern X t by the corrected powers Y t on the load transfer stage.

[0021] Since the load transfer produces a specific effect on the electrical power transmitted at the source station, the associated power data outside the normal operating pattern are processed and corrected and / or invalidated according to a specific processing step.

[0022] In this embodiment, the step of processing power data outside the normal operating pattern can be applied to each of the periods outside the normal operating pattern. More particularly, a period outside the normal operating pattern can correspond to the duration of a load transfer stage.

[0023] According to one embodiment, the situation giving rise to the power data outside the normal operating pattern is a reactive power regulation on the electricity distribution network.

[0024] According to this embodiment, reactive power regulation is identified when, in the reactive power step curve, two consecutive steps have a value difference equal to or greater than 900 kVAr.

[0025] In fact, a reactive power regulation at the transformer station produces a specific effect only on the reactive power at the said station (there is no effect on the active power). This type of situation can be identified by studying the differences in value between the levels of the reactive power step curve.

[0026] According to one embodiment, the step of processing power data outside the normal operating pattern comprises the following sub-steps: d1'. Calculation of corrected powers Y t by a relation of the type: Y t = X t − η x + η X t designating power data outside the normal operating pattern over a period outside the normal operating pattern of a low level among the two consecutive levels identified as a reactive power regulation level, η x the average of X t; η designating the average of the corrected powers obtained by calculation by linear interpolation between the average of the preceding level and the average of the level following said level; d2'. Obtaining corrected reactive power data by substituting said power data outside the normal operating pattern X t by the corrected powers Y t on the reactive power regulation level.

[0027] Since reactive power regulation produces a specific effect on the reactive power passing through the substation, the associated power data outside the normal operating pattern are processed and corrected and / or invalidated according to a specific processing step.

[0028] In this embodiment, the step of processing power data outside the normal operating pattern can be applied to each of the periods outside the normal operating pattern linked to a reactive power regulation. More particularly, a period outside the normal operating pattern can correspond to the duration of said low level among the two consecutive levels.

[0029] According to one embodiment, sub-step d2') is carried out when the period outside the normal operating pattern is equal to or greater than one hour. The time threshold beyond which it is chosen to carry out step d2') is determined according to the impact of the period outside the normal operating pattern on the adjustment of the model and its prediction capacity in the normal operating pattern, i.e. outside of situations at the origin of the power data outside the normal operating pattern.

[0030] According to one embodiment, the normalized power data are obtained by: calculating from the corrected power data, for a type of day and / or a type of season, a maximum operating limit at a time t by a relationship of the type: BORNEmax t = PROFIL t + 2 . MAD t and a minimum operating limit by a relation of the type: BORNEmin t = PROFIL t − 2 . MAD t where MAD denotes the mean absolute deviation from the median and is calculated as follows: MAD x i = m é diane x i − m é diane x i xi denoting corrected power data corresponding to the same type of day and / or type of season, and where PROFILE is obtained by calculating the median of the corrected power data corresponding to the same type of day and / or type of season.

[0031] The standardized power data makes it possible, in particular, to identify, on the corrected active and reactive power data, day / night cycles, sometimes summer / winter cycles and other regularities linked to the activity of businesses and households.

[0032] According to one embodiment, the reactive power forecasting model is established by Group Lasso type linear regression as a function of explanatory variables.

[0033] Group Lasso linear regression learning is advantageous in high dimension, when the number of explanatory variables is greater than or equal to the number of observations, and only a limited number of these variables have a significant influence on the observations, these variables being called "variables of interest". Group Lasso linear regression learning is also advantageous for performing a consistent selection of a restricted subset of variables of interest, often allowing a better interpretation of a model. These selected variables of interest are grouped into groups of variables of interest. Finally, another advantage of Group Lasso linear regression learning is the simultaneous processing of curves from hundreds of source stations, in a high-performance and fast manner.

[0034] According to one embodiment, the explanatory variables are grouped by type of variables, said type of variables being chosen from a list comprising: autoregressive variables representative of historical data, exogenous variables linked to temperature, exogenous variables linked to active power data, exogenous variables linked to a period of the year.

[0035] Another object of the present invention relates to a device for processing electrical power data passing through an electrical transformer station of an electricity distribution network, the device comprising a processing circuit for implementing the method according to the invention.

[0036] Another object of the present invention relates to a computer program comprising instructions for implementing the method according to the invention, in which said instructions are executed by a processor of a processing circuit.

[0037] The explanatory variables of the reactive power forecasting model by the substation include autoregressive variables representative of historical data. When implementing the forecasting model, the historical data of these variables may include power data outside the normal operating pattern.

[0038] According to one embodiment, the program further comprises instructions for forecasting, for an estimation day, reactive power passing through an electrical transformer station of an electricity distribution network, said forecasting instructions comprising the following steps: f1. Identification of operation outside the normal operating pattern or operation in the normal operating pattern of said station, by comparing active power and reactive power data from the day before the estimation day, with the standardized power data; f2. If operation outside the normal operating pattern is identified in step f1), calculation of the reactive power forecast for the estimation day: according to a first mode by applying the reactive power forecast model obtained in step e) of the method, from standardized power data; according to a second mode by applying the reactive power forecast model obtained in step e) of the method, from historical power data.

[0039] These forecast instruction steps are used to assess whether, on the day before the estimation day, the reactive power flowing through the substation is outside the normal operating pattern. This assessment is performed by comparing the reactive power measurement on the day before the estimation day and the normalized power data corresponding to the type of day and / or the type of season on the day before the estimation day. If it is identified that the reactive power on the day before the estimation day is outside the normal operating pattern, the reactive power forecast on the estimation day is calculated using the first mode where the historical data of the autoregressive variables of the model are replaced by their corresponding normalized power data, and using the second mode where the historical data of the autoregressive variables of the model are not replaced.The calculation of the reactive power forecast according to the second mode therefore includes the application of the forecast model from the historical data of the autoregressive variables, which are historical power data obtained by measurement at the transformer station level. Brève description des dessins

[0040] The invention will be better understood, and its advantages will appear better on reading the following detailed description of embodiments of the invention represented as non-limiting examples: Fig. 1 [ Fig. 1 ] is a functional diagram representing an electrical transformer station of an electricity distribution network; Fig. 2a [ Fig. 2a ] represents a time curve of reactive power at a source station; Fig. 2b [ Fig. 2b ] represents a time curve of active power at a source station; Fig. 3 [ Fig. 3 ] is a functional diagram of a method for processing measured active and reactive power data; Fig. 4 [ Fig. 4 ] is a functional diagram of a step of identifying power data outside the normal operating diagram; Fig. 5 [ Fig. 5 ] represents a reactive power curve during the implementation of a sub-step of the process shown schematically in figure 4 ; Fig. 6 [ Fig. 6 ] represents a reactive power curve during the implementation of another sub-step of the process shown schematically in figure 4 ; Fig. 7 [ Fig. 7 ] is a functional diagram of a step of identifying the type of situation at the origin of the power data outside the normal operating diagram; Fig. 8a [ Fig. 8a ] is a functional diagram of power data processing steps outside the normal operating diagram; Fig. 8b [ Fig. 8b ] is a functional diagram of power data processing steps outside the normal operating diagram; Fig. 8c [ Fig. 8c ] represents a reactive power curve following the implementation of a sub-step of the process shown schematically in figure 3 ; Fig. 9 [ Fig. 9 ] diagrams the execution of a computer program on a device; Fig. 10 [ Fig. 10 ] is a block diagram of prediction instructions in the computer program. Description des modes de réalisation

[0041] The following examples will allow a better understanding of the present invention, without limiting its scope.

[0042] In the various figures, the same references designate identical or similar elements.

[0043] There figure 1 is a functional diagram representing an electrical transformer station Po of an electricity distribution network, also called a source station. The source station Po lowers the high voltage B (HTB) to high voltage A (or HTA), or to medium voltage (MV), and allows electricity to pass from a transmission network to industrial sites or customers which are directly connected to the network, or to a distribution network which connects with consumers, particularly individuals. The transmission network allows electricity to be transported from electricity producers to a transformer station such as Po, thanks to an electricity transmission network manager.

[0044] Active and reactive power sensors SENS (P; Q) are located at the LIV delivery point of the electricity transmission system operator. The Po source substation includes adjustable coupling transformers allowing REGU regulation to be carried out in order to lower high voltage electricity B, exceeding 50,000 volts in alternating current, into high voltage A (or HTA) or medium voltage (MV), exceeding 1,000 volts without exceeding 50,000 volts in alternating current.

[0045] Reactive compensation at the Po source substation is carried out by equipment generally comprising varmetric regulators allowing the automatic switching of capacitor banks Gr in order to maintain a target power factor. The capacitor banks are generally made up of several single-phase or three-phase unit capacitors, assembled and interconnected to create power sets called "capacitor banks". Electromechanical or static contactors make it possible to establish or interrupt the flow of current, from an electrical or pneumatic control, in particular to the capacitor banks.

[0046] The Po source substation includes a DÉIE device, which is an Operating Information Exchange Device between an electricity distributor and a Producer Site connected to the HTA network, allowing the immediate transmission of information necessary for both reliable and responsive operation of the HTA network. REGUP voltage regulation by the producer can also be carried out at this level.

[0047] Circuit breakers are provided at the Po source substation and protect the network against possible overloads due to fault currents caused, for example, by lightning or by the ignition of a tree branch located too close to the line. Circuit breakers are used to switch sections of the circuit on or off. Disconnectors can be provided at the Po source substation to ensure a visible break in an electrical circuit and to direct the current into the substation.

[0048] THE figures (2a , 2b) represent two time curves.

[0049] The curve in figure 2a represents reactive power data Q measured at the SENS sensor of an electrical transformer station Po of an electricity distribution network, over an observation period T of one year at a time step of 10 minutes. The reactive power values Q are expressed in volt-ampere reactive (VAr), and vary between -10000 VAr and 5000 VAr on the curve represented in figure 2a . The periods referenced PGr on the curve indicate periods of sudden drop in reactive power measured at the observed transformer station. The PGr periods of drop in reactive power can be of variable duration. The PGr periods of drop in reactive power represented in figure 2a are due to the activation of capacitor steps, to achieve reactive power regulation at the source station Po.

[0050] The curve in figure 2b represents active power data P measured at the SENS sensor of the same electrical transformer station Po, over the same observation period. The active power values P are expressed in Watts (W), and vary between 0 and 2*10 4< W on the curve represented in figure 2b . The values of active power P do not undergo sudden changes in periods Gr since the activation of the capacitor steps only acts on the reactive power. On the other hand, a progressive decrease in active power is observed in the middle of the year when there is no electricity consumption for electric heating.

[0051] There figure 3 represents a method for processing electrical power data passing through the Po substation, with a view to establishing a forecasting model Mp of the electrical power at the Po substation, outside the following situations: reactive power regulation, carried out in particular by capacitor banks; load transfers between source stations.

[0052] A load shift corresponds to a situation where, during an incident, work or intervention, part or all of the load of a source substation, called a backup substation, is taken over by one or more other source substations, called backup substations. Load shifts allow active power P to be shifted to the source substations, and have an impact on reactive power Q. Load shifts must therefore be detected when building a reactive electrical power forecast model Mp.

[0053] A reactive power regulation Q corresponds to a situation where capacitor banks, which are reactive power banks located downstream of the transformers of the source stations, are activated either automatically by a varmetric regulator when the reactive power exceeds a certain threshold, or manually by a network operator.

[0054] As shown in the diagram of the figure 3 , said data processing method comprises a first step S1 where active power and reactive power data Dm measured at said station are obtained over the observation period T.

[0055] In step S2, power data outside the normal operating pattern Dhsne are identified, during said period T, by monitoring the variation over time of the reactive power Q and / or the active power P and by detecting a jump in said variation.

[0056] More particularly, the identification of power data outside the normal operating pattern Dhsne during said period T requires the detection of sudden drops in the measured power data Dm, representative of a situation Shsne at the origin of the power data outside the normal operating pattern.

[0057] Step S3 makes it possible to identify at least one type of situation Shsne at the origin of the power data outside the normal operating pattern Dhsne identified.

[0058] The Shsne situation causing the power data outside the normal operating pattern may be reactive power regulation or load transfer to the electricity distribution network.

[0059] The power data outside the normal operating pattern Dhsne are then processed in step S4 according to the type of situation identified, in order to obtain corrected power data Dcc over the period T.

[0060] In step S5, a model is established, by learning from the corrected power data Dcc, and makes it possible to predict the reactive power of said station in operation outside of a situation of the type of situation identified in step S3.

[0061] The interest in developing such a model is the forecasting of the power (active P and reactive Q) transiting at the source substations, with a view to anticipating constraints on the distribution network, adequately dimensioning the source substations, the transmission and distribution networks and, where appropriate, implementing anticipatory resolution levers. The construction of a learning forecasting model consists of identifying consumption patterns and influential factors, based on a history of electrical power data.

[0062] The reactive power Q measured at the source substations Po is subject to uncertainties linked to the consumption and production of reactive power by users on the network downstream of the substations Po. Thus, the reactive power Q curves often present day / night cycles, sometimes summer / winter cycles and other regularities linked to the activity of businesses and households. The power data collected in this context are said to be in "normal operating pattern" (SNE).

[0063] However, as represented in figure 2 , hazards related to network operation can significantly disrupt these diagrams. During incidents, works or other often unpredictable constraints, the distribution network manager must reconfigure the network. The electrical power passing through the Po substation then leaves the "normal operating pattern" (SNE) and the reactive power measurements recovered at the source substation no longer correspond to the same users (consumers and producers) as in SNE. These Shsne situations lead to operation "outside the normal operating pattern" (HSNE) and must not influence the Mp forecast model.

[0064] When detecting power outside the normal operating scheme, two types of operating situations outside the Normal Operating Scheme in particular are processed in step S4: load reports and reactive power regulations, particularly by capacitor steps. Step S4 allows operating situations outside the Normal Operating Scheme to be corrected or invalidated, and operating limits to be calculated in the Normal Operating Scheme.

[0065] As represented in figure 4 , the step of identifying the power data outside the normal operating pattern Dhsne comprises a first step S21 of obtaining, from the power data measured over the period T, for each of the active powers P and reactive powers Q, a first envelope curve E max increasing the power data measured Dm over the period T and a second envelope curve E min reducing the power data measured Dm over the period T.

[0066] This is an image processing technique, called "opening and closing", which allows the calculation of "envelope" curves, one majorant (E max ) and the other minorant (E min ), which frame (cf. figure 5 ) the power curves (active P or reactive Q).

[0067] The identification of power data outside the normal operating diagram Dhsne comprises step S22 during which, for each of the active powers P and reactive powers Q, a deviation curve E between the two envelope curves is obtained, by calculating a deviation between the first envelope curve E max and the second envelope curve E min according to a relationship of the type: E = Emax − Emin

[0068] A deviation interval [-E, E] is defined in step S23, the upper limit of which corresponds to the deviation curve E and the lower limit of which corresponds to the opposite of the deviation curve -E.

[0069] A derivative curve D is obtained in step S24, from the first and second envelope curves E max and E min , by calculating a derivative of an average of the first and second envelope curves E max and E min through a relation of the type D = d é riv é e Emax + Emin 2

[0070] Step S25 allows ST jumps to be identified over the period T. A jump is defined by an instant when a value of the derivative curve D leaves the deviation interval [-E, E].

[0071] The jump is of the low type if the value D is below the deviation interval [-E, E] and corresponds to a sudden jump with a drop in reactive power.

[0072] The jump is of the high type if the value D is above the deviation interval [-E, E] and corresponds to a sudden jump with an increase in reactive power.

[0073] In step S26, a step curve Cp is obtained for each of the active P and reactive Q powers. A step Pa is obtained by calculating the average of the measured power data Dm over a time interval Is defined by two consecutive ST jumps, the step being of the low type Pab if the first of the two ST jumps is low and of the high type Pah if the first of the two ST jumps is high. The step curve Cp is represented in figure 6 , and includes several drops in reactive power represented by low Pab levels. The drop or increase in reactive power can also be assessed with reference to reactive power values expected in a Normal Operating Scheme.

[0074] As represented in figure 7 , in step S31, a load transfer type situation Ac CH on the electricity distribution network is identified on the step curve when two steps of the same type occur simultaneously on the step curves of each of the active and reactive powers. Said steps of the same type occurring simultaneously are identified as load transfer steps.

[0075] When the load transfer levels are low Pab, it is a period of load decrease. When the load transfer levels are high Pah, it is a period of load increase.

[0076] The power data outside the normal operating pattern Dhsne are the data X t , where t is a 10-minute time step included in the period outside the normal operating pattern.

[0077] In the case of a load transfer type situation, the step of processing power data outside the normal Dhsne operating scheme is represented in figure 8a and includes step S411 of calculating the corrected powers Y t by a relation of the type: Y t = Z t + η σ . σ Or Z t = X t − η x / σ x X t denotes the power data outside the normal operating pattern over a period during one of the load transfer stages. η x denotes the average of the X t data over the duration of the load transfer stage, and σ x the standard deviation of the X t data over this same period. η is the result of the calculation by linear interpolation between the averages of the stages preceding and following said stage, and σ is the result of the calculation by linear interpolation between the standard deviations of the stages preceding and following said stage.

[0078] Corrected active and reactive power data Dcc over the period outside the normal operating pattern are obtained in step S421 by substituting the power data outside the normal operating pattern X t with the corrected powers Y t on each level outside the normal operating pattern, preferably on each load transfer level.

[0079] In a variant of the method, a load deferral period is corrected only if its duration is less than one week.

[0080] As represented in figure 7 , in step S31, a situation of the reactive power regulation type Ac GR , in particular by capacitor steps on the electricity distribution network, is identified when, in the reactive power step curve Cp(Q), two consecutive steps have a value difference equal to or greater than 900 kVAr. Said consecutive steps are identified as reactive power regulation steps.

[0081] In the case of a reactive power regulation type situation Ac GR, the step of processing power data outside the normal operating scheme Dhsne is represented in figure 8b and includes step S412 of calculating the corrected powers Y t by a relation of the type: Y t = X t − η x + η X t denotes the power data outside the normal operating pattern over a period outside the normal operating pattern during a low stage Pab among the two consecutive stages. η x denotes the average of the data X t over the stage Pab. η is the result of the calculation by linear interpolation between the averages of the stages preceding and following said stage.

[0082] Corrected reactive power data over the period outside the normal operating pattern are obtained in step S422 by substituting the power data outside the normal operating pattern Dhsne with the corrected powers Y t over the reactive power regulation stages in the measured power data.

[0083] According to one embodiment, sub-step S422 is carried out when the period outside the normal operating pattern is equal to or greater than one hour.

[0084] Obtaining corrected Dcc power data, an example of which is shown in figure 8c , is used to feed the construction of the Mp forecasting model by learning. The corrected power data Dcc also allows the calculation of normalized power data Dcn. The usefulness of the normalized power data Dcn will be explained during the implementation of the Mp forecasting model.

[0085] The normalized power data Dcn are obtained by calculation, for a type of day and / or a type of season of: PROFILE (t), obtained by calculating the median of the corrected power data corresponding to the same type of day and / or type of season; the mean absolute deviation of the median of corrected power data xi corresponding to the same type of day and / or type of season by a relation of the type: mAD x i = m é diane x i − m é diane x i a maximum operating limit at a time t by a relation of the type: BORNEmax t = PROFIL t + 2 . MAD t a minimum operating limit by a relation of the type: BORNEmin t = PROFIL t − 2 . MAD t

[0086] The BORNE max and BORNE min operating terminals are operating terminals in SNE, for each type of day of the week and each season. There are three types of day: type 1 from Monday to Friday, type 2 Saturday, type 3 Sunday and public holidays. There are three seasons: winter, summer and mid-season.

[0087] The forecasting model is established by Group-Lasso type linear regression carried out according to a matrix formula of the type: Q j + 1 t = X t ⋅ β t + ε t with : t being a time index ranging from 1 to 48, one model per half-hour being forecast; Q j+1 being the reactive power to be forecast for the following day, day j being the current day; X is a vector containing explanatory variables of the model; β are parameters to be estimated; ε is the modeling error.

[0088] There are as many models as there are source stations (2300) and times t (48).

[0089] In the context of Group-Lasso regression, the β parameters are estimated by an optimization problem of the type: arg min β 1 2 Q j + 1 − X ⋅ β 2 2 + λ ∑ j = 1 K w j . β G j 2

[0090] The explanatory variables are grouped into K groups Gj according to their nature, λ being a positive regularization parameter, wj being a strictly positive weight associated with the group G j of variables (generally this weight is: Card G j

[0091] and β Gj being, for G j a group of variables, the vector β restricted to the elements of the group G j: Autoregressive variables relating to the past of reactive powers; Exogenous variables linked to temperature; Exogenous variables linked to active power; Exogenous variables linked to the calendar.

[0092] In this embodiment, the model is trained for 2300 source stations. The basic model is identical for all 2300 source stations. Optimizing the Group-Lasso problem allows selecting the most explanatory variables for each individual source station, thanks to an estimation of the β parameters during the training process. Thus, if the reactive power of a source station is not thermosensitive, then temperatures will not be involved in the forecasting process, unlike for another thermosensitive station.

[0093] Parameter learning is performed on a history of corrected Dcc power data, on which load reports and reactive power regulations have been corrected as explained previously.

[0094] The forecast is made by applying the following formula: Q j + 1 ^ = X ⋅ β ^ with : β ^ the parameters estimated during model training; X the explanatory variables; Q j + 1 ^ the reactive power forecast for the following day.

[0095] There figure 9 represents a device for processing data Dm of electrical power from an electrical transformer station Po of an electricity distribution network. The device comprises a processing circuit CT for implementing one or other of the embodiments of the invention presented above. More particularly, the device is suitable for implementing the method presented above for processing data Dm of electrical power passing through the station Po, with a view to establishing the forecast model Mp of the electrical power from the station Po. By way of example, the processing circuit CT may typically comprise: an interface IN for receiving the measured power data Dm, a memory MEM suitable for storing instructions of a computer program to implement the method according to the invention, as well as temporary data (for example calculation data or others) or permanent data (for example instruction codes of a computer program within the meaning of the invention), a processor PROC connected to the memory MEM to read the instructions of the computer program and implement the above method, an output interface IHM (for example a graphical interface or more generally a human / machine interface) for example to display forecast reactive power data.

[0096] The vector of explanatory variables contains past powers D h,reg that may be in HSNE. These data are identified by comparing them to the operating limits in the normal operating pattern obtained during the calculation of the normalized power data Dcn. When the data from the day before the estimation day are identified as being in HSNE, the model provides two results; one in the normal operating pattern SNE by replacing the past data with their previously calculated profile, and the other outside the normal operating pattern by not modifying the past data in HSNE.

[0097] For this purpose, said computer program further comprises instructions for forecasting, for an estimation day j+1, the reactive power of an electrical transformer station Po of an electricity distribution network. Said forecasting instructions are shown diagrammatically by step S6 on the figure 10. The said forecasting instructions include the following steps: S61. Identification of operation outside the normal HSNE operating pattern or operation in the normal SNE operating pattern of said station, by comparison of measured power data Dc(j) of active power P and reactive power Q of the day before j of estimation day j+1, with the normalized power data Dcn(j) corresponding to the day and season of the day before j; S62.If an operation outside the normal HSNE operating pattern is identified in step S61, calculation of reactive power Q for the estimation day j+1: according to a first mode MODE1 by application of the reactive power forecast model Mp Q obtained in step S5 of the method, from the normalized power data Dcn, in particular in correspondence with historical data D h,reg contained in the autoregressive variables of the model; according to a second mode MODE2 by application of the reactive power forecast model Mp obtained in step S5 of the method, from the power data, in particular the historical power data D h,reg of the autoregressive variables of the model Mp. The historical power data D h,reg contained in the autoregressive variables of the model Mp are not replaced in this mode.If the operation identified in step S61 is an operation in normal SNE operating pattern, the historical power data D h,reg contained in the autoregressive variables of the model Mp are not replaced.

[0098] This data processing in step S6 makes it possible to produce forecast results consistent with the information contained in the autoregressive variables of the model Mp. For example, if the reactive power during the week of the estimation day j+1 was in HSNE, the forecast instructions described above make it possible to produce an estimate of reactive power in HSNE on day j+1 through MODE2, as well as an estimate of reactive power in SNE on day j+1 through MODE1. MODE2 is of interest, for example, in cases where on day j+1, the reactive power of the substation is still in HSNE. MODE1 is of interest, for example, in cases where on day j+1, the reactive power of the substation has ceased to be in HSNE and has switched to operation in SNE.

Claims

1. A method for processing electrical power data transiting at an electrical transformation station Po of an electricity distribution network, comprising the following steps: a. Obtaining data of active power P and reactive power Q measured Dm at said station Po over an observation period T; b. Identifying, from the measured power data Dm over the observation period T, power data outside the normal operating pattern Dhsne during said observation period T, by monitoring the variation over time of the reactive power Q and / or the active power P and by detecting jumps in said variation; c. Identifying at least one type of situation Shsne at the origin of the identified power data outside the normal operating pattern Dhsne, said type of situation being at least one type of situation among a reactive power regulation on the electricity distribution network and a load transfer on the electricity distribution network, the identification step comprising the following sub-steps: c1- Calculating, from the measured power data Dm over the observation period T, for each of the active P and reactive Q powers, a first envelope curve Emax increasing the measured power data Dm over the observation period T and a second envelope curve Emin reducing the measured power data Dm over the observation period T; c2- Calculating a difference between the first envelope curve (Emax ) and the second envelope curve (Emin ) to obtain a difference curve E; c3- Defining a difference interval [-E, E], the upper limit of which corresponds to the difference curve E and the lower limit corresponds to the opposite of the difference curve E; c4- calculating a derivative of an average of the first and second envelope curves (Emax, Emin ), said derivative being called a derived curve; c5- Identifying two jumps ST over the observation period T, the jump being defined by a time when a value of the derived curve D is outside the difference interval [-E, E], the jump being of the low type if the value is below the difference interval [-E, E] and of the high type if the value is above the difference interval [-E, E]; c6- Calculating a curve by steps Cp, for each of the active power P and reactive power Q, a step Pa being obtained by calculating the average of the measured power data Dm over a time interval Is defined by two consecutive jumps ST, the step being of the low type Pab if the first of the two jumps ST is low and of the high type Pah if the first of the two jumps ST is high; d. Calculating corrected power data Dcc from the power data outside the normal operating pattern Dhsne and depending on the type of identified situation Shsne, and calculating normalised power data Dcn over the observation period T; e. Establishing, by learning from the corrected power data Dcc, a reactive power prediction model Mp of said station in operation outside a situation Shsne of the type of situation identified in step c); f. providing, for an estimation day (d+1), a reactive power transiting at an electrical transformation station Po of an electricity distribution network, said prediction step comprising an identification of an operation outside the normal operating pattern or an operation in a normal operating pattern of said station, by comparing active power and reactive power data from the day (d) before the estimation day (d+1), with the normalised power data Dcn; and if an operation outside the normal operating schedule is identified, implementing, for the estimation day, the calculated reactive power prediction: - according to a first mode by applying the reactive power prediction model obtained in step e) of the method, from the normalised power data Dcn; - according to a second mode by applying the reactive power prediction model obtained in step e) of the method, from historical power data.

2. The method according to claim 1, wherein the situation Shsne is a load transfer on the electricity distribution network.

3. The method according to claim 2, taken in combination with claim 2, wherein steps of the same type occurring simultaneously on the step curves Cp of each of the active and reactive powers are identified as load transfer steps.

4. The method according to claim 3, wherein the step d) of calculating the corrected power data Dcc comprises the following sub-steps: d1. Calculation of corrected powers Yt by a relation of the type: Y t = Z t + η σ ⋅ σ where Z t = X t − η x / σ x Xt designating power data outside the normal operating pattern over a period outside the normal operating pattern during one of the load transfer steps, ηx the average of Xt, σx the standard deviation of X t; η designating the average of the corrected powers obtained by calculation by linear interpolation between the average of the previous step and the average of the step following said step; σ designating the standard deviation of the corrected powers obtained by calculation by linear interpolation between the standard deviation of the previous step and the standard deviation of the step following said step; d2. Obtaining corrected active and reactive power data Dcc by substituting said power data outside the normal operating diagram Xt with the corrected powers Yt on the load transfer step.

5. The method according to claim 1, wherein the situation Shsne is a reactive power regulation on the electricity distribution network.

6. The method according to claim 5, taken in combination with claim 2, wherein the reactive power regulation is identified when, in the reactive power step curve Cp (Q), two consecutive steps have a value difference equal to or greater than 900 kVAr.

7. The method of claim 6, wherein the step d) of calculating the corrected power data Dcc comprises the following sub-steps: d1'. Calculation of corrected powers Yt by a relation of the type: Y t = X t − η x + η Xt designating power data outside the normal operating pattern over a period outside the normal operating pattern during a low step Pab among the two consecutive steps identified as a reactive power regulation step, ηx the average of X t; η designating the average of the corrected powers obtained by calculation by linear interpolation between the average of the previous step and the average of the step following said step Pab; d2. Obtaining corrected reactive power data Dcc by substituting said power data outside the normal operating diagram Xt with the corrected powers Yt on the reactive power regulation step.

8. The method according to claim 7, wherein the sub-step d2') is carried out when the period outside the normal operating pattern is equal to or greater than one hour.

9. The method according to one of claims 4, 7 and 8, wherein the normalised power data Dcn is obtained by: calculating from the corrected power data Dcc, for a type of day and / or a type of season, a maximum operating boundary at a time t by a relationship of the type: LIMITmax t = PROFILE t + 2 . MAD t and a minimum operating terminal by a relationship of the type: LIMITmin t = PROFIL t − 2 . MAD t where MAD designates the mean absolute deviation from the median and is calculated as follows: MAD x i = median x i − median x i xi designating corrected power data corresponding to the same type of day and / or type of season, and where PROFIL is obtained by calculating the median of the corrected power data corresponding to the same type of day and / or type of season.

10. The method according to one of claims 1 to 9, wherein the reactive power prediction model is established by Group Lasso type linear regression as a function of explanatory variables (X).

11. The method according to claim 10, wherein the explanatory variables are grouped by type of variables, said type of variables being selected from among a list comprising: self-regressive variables representative of historical data D h,reg, exogenous variables related to the temperature, exogenous variables related to active power data, exogenous variables related to a period of the year.

12. A device for processing data of electrical powers transiting at an electrical transformation station Po of an electricity distribution network, the device including a processing circuit (CT) for implementing the method according to one of the preceding claims.

13. A computer program including instructions for implementing the method according to one of claims 1 to 11, wherein said instructions are executed by a processor of a processing circuit.