A method for predicting voltage sensitivity of a power grid

Abstract

A method for predicting voltage sensitivity of a power grid is disclosed, the power grid being coupled to a power generating unit at a point of interconnection Data (1) related to a plurality of operating points of the power grid is obtained, including a point in time where the data (1) related to the operating point was obtained. For each operating point, a voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (IQ) / reactive power (Q), corresponding to the operating point is calculated, using a model representation of the power grid. The data (1) related to the plurality operating points, including the points in time where the data (1) was obtained, and the corresponding calculated voltage sensitivities are supplied to a trained artificial intelligence (AI) model (3), identifying patterns related to voltage sensitivity of the power grid. Time dependent predictions of voltage sensitivity of the power grid are derived, based on the identified patterns.

Description

[0001] A METHOD FOR. PREDICTING VOLTAGE SENSITIVITY OF A POWER GRID

[0002] FIELD OF THE INVENTION

[0003] The present invention relates to a method for predicting voltage sensitivity of a power grid in response to changes in active current and / or active power supplied to the power grid from a power generating unit, and / or in response to changes in reactive current and / or reactive power supplied to the power grid from the power generating unit. The method according to the invention provides accurate time dependent predictions of voltage sensitivity of the power grid, thus allowing for proactively initiating measures for maintaining a stable power grid.

[0004] BACKGROUND OF THE INVENTION

[0005] Power grids rely on power feed from various power generating units. Traditionally, stability of power grids has been maintained by means of large conventional power plants. However, renewable power generators, such as wind turbines, wind farms, photovoltaic panels, solar farms, etc., provide a continuously increasing part of the power feed to power grids, thus displacing the stable conventional power plants. The inherent fluctuating nature of the power output of such renewable power generators introduces a risk of unstable conditions in the power grid when the power feed provided by the renewable power generators reaches a certain fraction of the total power feed to the power grid. Therefore, it is becoming increasingly common to require that renewable power generators are capable of providing grid stabilizing services to the power grid, e.g. in terms of voltage control and / or frequency control.

[0006] In order to enable renewable power generators to provide such grid stabilizing services, it may be desirable to be able to predict when there is a risk of conditions occurring that might result in instability of the power grid. However, it has proved difficult to make accurate predictions by means of a model based approach, due to the inherent non-linear and dynamic nature of the power grid, e.g. caused by the presence of other power generators nearby. DESCRIPTION OF THE INVENTION

[0007] It is an object of embodiments of the invention to provide a method for predicting voltage sensitivity of a power grid, which is accurate and reliable.

[0008] According to a first aspect the invention provides a method for predicting voltage sensitivity of a power grid, the power grid being coupled to a power generating unit at a point of interconnection, the method comprising the steps of:

[0009] - obtaining data related to a plurality of operating points of the power grid, each operating point defining a voltage (Vmeas) across the point of interconnection, active current (Ip) / active power (P) delivered by the power generating unit to the power grid, reactive current (Iq) / reactive power (Q) delivered by the power generating unit to the power grid, and a point in time where the data related to the operating point was obtained,

[0010] - for each operating point, calculating a voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (Iq) / reactive power (Q), corresponding to the operating point, based at least on the voltage (Vmeas), the active current (Ip) / active power (P) and the reactive current (Iq) / reactive power (Q), and using a model representation of the power grid,

[0011] - supplying the data related to the plurality operating points, including the points in time where the data was obtained, and the corresponding calculated voltage sensitivities to a trained artificial intelligence (Al) model,

[0012] - identifying patterns related to voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), by means of the Al model, and deriving time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (IQ) / reactive power (Q), based on the identified patterns.

[0013] Thus, the method according to the first aspect of the invention is a method for predicting voltage sensitivity of a power grid. In the present context the term 'voltage sensitivity' should be interpreted to mean a sensitivity of the voltage of the power grid in response to changes in other parameters related to the power grid, in particular changes in active and / or reactive current and / or power delivered to the power grid from one or more power generators. If the sensitivity of the voltage of the power grid is high, i.e. if it appears that the voltage of the power grid changes significantly in response to changes in relevant parameters related to the power grid, then there might be a risk of the power grid becoming unstable.

[0014] The power grid is coupled to a plurality of power generators and a plurality of power consumers, thus enabling distribution of power produced by the power generators to the power consumers for consumption. In the present context the term 'power generating unit' should be interpreted to mean a unit which is capable of generating electrical power, and which supplies all or part of the produced power to the power grid. The power generating unit may be a conventional unit, such as a power plant, a combustion engine, etc., or it may be or include a renewable power generating unit, such as a wind turbine, a photovoltaic cell, etc.

[0015] Thus, the power grid is coupled to a power generating unit at a point of interconnection, and in the following the interaction between the power grid and this specific power generating unit is considered.

[0016] In the method according to the first aspect of the invention, data related to a plurality of operating points of the power grid is initially obtained. Each operating point defines at least four quantities, i.e. (i) a voltage (Vmeas) across the point of interconnection; (ii) active current (IP) and / or active power (P) delivered by the power generating unit to the power grid; (iii) reactive current (IQ) and / or reactive power (Q) delivered by the power generating unit to the power grid; and (iv) a point in time where the data related to the operating point was obtained.

[0017] The voltage (Vmeas) is the voltage across the point of interconnection between the power grid and the power generating unit, i.e. across the point where the power generating unit is coupled to the power grid. Accordingly, the voltage, Vmeas, is the voltage in front of the power generating unit and towards the power grid.

[0018] The active current (IP) represents an amount of active current (IP) delivered by the power generating unit to the power grid. Similarly, the active power (P) represents an amount of active power delivered by the power generating unit to the power grid. Thus, the active current (IP) as well as the active power (P) represents the active part of the power which the power generating unit provides to the power grid.

[0019] Similarly, the reactive current (IQ) represents an amount of reactive current (IQ) delivered by the power generating unit to the power grid, and the reactive power (Q) represents an amount of reactive power delivered by the power generating unit to the power grid. Thus, the reactive current (IQ) as well as the reactive power (Q) represents the reactive part of the power which the power generating unit provides to the power grid.

[0020] Thus, each operating point represents interrelated, correlated or corresponding values of the quantities (i)-(iii), related to a specific point in time (iv).

[0021] Accordingly, the plurality of operating points forms a collection or library of such interrelated, correlated or corresponding values, e.g. related to various points in time.

[0022] For each of the operating points, a voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), corresponding to the operating point, is calculated. The calculation is performed based at least on the voltage (Vmeas), the active current (Ip) / active power (P) and the reactive current (Iq) / reactive power (Q) defined by the relevant operating point. Furthermore, the calculation is performed using a model representation of the power grid.

[0023] The model representation of the power grid applied for calculating the voltage sensitivity may, e.g., represent an approximation of the behaviour of the power grid under various operating conditions. It may take static as well as dynamic behaviour into account, at least to a certain extent. The model representation may be a mathematical representation of the physical power grid, e.g. in the form of a state space representation.

[0024] Thus, in addition to quantities (i)-(iv) defined above, each operating point now also defines a calculated voltage sensitivity of the power grid, related to the point in time specified by the operating point.

[0025] The data related to the plurality of operating points, including the points in time where the data was obtained, and the corresponding calculated voltage sensitivities, is supplied to a trained artificial intelligence (Al) model. Thus, the Al model receives correlated information regarding voltage across the point of interconnection, the active and reactive parts of the power delivered by the power generating unit to the power grid and the voltage sensitivity of the power grid, at a plurality of specified discrete points in time.

[0026] The Al model has been previously trained, e.g. based on simulations and / or previously obtained data, possibly supplemented by reinforced learning during usage of the Al model. This will be described in further detail below.

[0027] The Al model is then applied for identifying patterns related to voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q). For instance, the Al model may identify patterns in the correlated information regarding the voltage across the point of interconnection, the active and reactive parts of the power delivered by the power generating unit to the power grid and voltage sensitivity of the power grid, as a function of time. Finally, time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (IQ) / reactive power (Q) are derived, based on the identified patterns. For instance, the identified patterns may reveal that certain combinations of voltage, active current / power and / or reactive current / power may be likely to result in high or low voltage sensitivity of the power grid.

[0028] Furthermore, since the information supplied to the Al model includes the point in time where the data related to the respective operating points was obtained, the identified patterns will also reveal points or periods in time where such combinations of voltage, active current / power and / or reactive current / power are likely to occur, thus allowing the time dependent predictions of voltage sensitivity to be derived.

[0029] Based on the discrete information supplied thereto, the Al model may be able to 'fill in gaps', so as to allow for prediction of the behaviour of the power grid at operating points and / or points in time where no real or measured data exists.

[0030] The time dependent predictions resulting from performing the method according to the first aspect of the invention are accurate, since they rely on patterns derived by the Al model, based on real, possibly measured, data. Thus, nonlinearities, dynamic behaviour, changes in the power grid over time, etc., are automatically taken into account.

[0031] The time dependent predictions of voltage sensitivity of the power grid may reflect estimated variations in voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (Iq) / reactive power (Q) during a diurnal cycle, a weekly cycle and / or a yearly cycle.

[0032] According to this embodiment, the time dependent predictions provide information regarding specific times during a day / night, specific times during a week and / or specific times during a year, where the probability of high voltage sensitivity of the power grid, and thus potentially an unstable power grid, is high or low. This may, e.g., allow for appropriate measures to be taken in a timely manner, so as to avoid incidents of an unstable or weak power grid. For instance, power generators and / or power consumers connected to the power grid may be requested to behave in a specified manner during specific periods during a relevant time cycle, such as increasing or decreasing power production or power consumption. This will be described in further detail below.

[0033] The method may further comprise the step of predicting periods of highly voltage sensitive power grid, based on the derived time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q). As described above, such periods could, e.g., be specific time periods during a relevant time cycle.

[0034] In the present context the term 'highly voltage sensitive power grid' should be interpreted to mean situations where changes in active current, active power, reactive current and / or reactive power delivered from the power generating unit to the power grid are likely to result in significant changes in the voltage of the power grid, thus potentially increasing the risk of an unstable or weak power grid. Various detectable parameters or combinations of parameters may be indicative for the occurrence of a highly voltage sensitive power grid, and such parameters or combinations of parameters may be readily derivable from the patterns identified by the Al model. This will be described in further detail below. Accordingly, the identified patterns also reveal time periods where such parameters or combinations of parameters occur, thus allowing the relevant periods of time to be identified.

[0035] The method may further comprise the step of proactively initiating measures for counteracting highly voltage sensitive power grid prior to predicted periods of highly voltage sensitive power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q).

[0036] Once periods time where highly voltage sensitive power grid is likely to occur have been identified, it is possible to take appropriate measures before such periods of time occur. For instance, in the case that the derived time dependent predictions indicate that the power grid is likely to be highly voltage sensitive at specific recurring time periods, e.g. during a diurnal, weekly and / or yearly cycle, it is possible to request power generators and / or power consumers connected to the power grid to behave in a specified manner immediately before such time periods, in order to prevent that the predicted periods of highly voltage sensitive power grid result in an unstable or weak power grid. Furthermore, it is possible to limit the need for the power generators and / or power consumers to react fast on instabilities in the power grid, such as voltage deviations and / or frequency deviations.

[0037] The measures taken before the predicted time periods of highly voltage sensitive power grid may, e.g., include requesting the power generating unit to increase or decrease the active power delivered by the power generating unit to the power grid, in the case that the frequency of the power grid is likely to decrease below or increase above, respectively, a specified frequency range. Similarly, the power generating unit may be requested to increase or decrease the reactive power delivered by the power generating unit to the power grid, in the case that the voltage of the power grid is likely to decrease below or increase above, respectively, a specified voltage range. Alternatively or additionally, other appropriate measures may be taken.

[0038] The method may further comprise the step of updating the Al model based on correlated measurements of voltage (Vmeas) across the point of interconnection, active current (Ip) / active power (P) delivered by the power generating unit to the power grid, and reactive current (IQ) / reactive power (Q) delivered by the power generating unit to the power grid, and using reinforced learning.

[0039] According to this embodiment, the Al model is continuously updated and improved during usage of the Al model as part of the method according to the invention. Thereby it is ensured that the Al model continues to accurately reflect the behaviour of the power grid, including the dynamic behaviour of the power grid, even if such behaviour changes over time.

[0040] The method may further comprise repeating the steps of identifying patterns related to voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (IQ) / reactive power (Q), and deriving time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (Iq) / reactive power (Q), based on the updated Al model.

[0041] According to this embodiment, once the Al model has been updated as described above, the updated Al model is applied for identifying the patterns, and eventually for deriving the time dependent predictions. Thereby it is ensured that the time dependent predictions accurately reflect the actual behaviour of the power grid, even if this behaviour changes over time.

[0042] The power generating unit may be a wind turbine, a wind farm or a hybrid renewable power plant, and the method may further comprises the step of controlling the power generating unit to provide a power feed into the power grid, based on the derived time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (IQ) / reactive power (Q).

[0043] In the present context, the term 'wind farm' should be interpreted to mean a plurality of wind turbines arranged within a specified geographical area, and which share some infrastructure, such as internal power grid, connection to an external power grid, substations, access roads, etc. Similarly, in the present context, the term 'hybrid renewable power plant' should be interpreted to mean a plurality of renewable power generators of at least two different kinds arranged within a specified geographical area, and which share some infrastructure, such as internal power grid, connection to an external power grid, substations, access roads, etc. For instance, the hybrid renewable power plant may include a plurality of wind turbines and a plurality of photovoltaic panels.

[0044] Thus, according to this embodiment, the power generating unit is a renewable power generating unit, and thus of a kind where the power production depends strongly on the available renewable resources, such as wind and / or solar influx. As described above, these types of power generating units are to an increasing extent requested to provide grid stabilising services to power grids. The time dependent predictions of voltage sensitivity of the power grid obtained by means of the method according to the invention allow the renewable power generating units to provide such services in a proactive manner.

[0045] The step of, for each operating point calculating a voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q) may comprise calculating dV / dP, dV / dlp, dV / dQ and / or dV / dlq.

[0046] According to this embodiment at least one of dV / dP, dV / dlp, dV / dQ and dV / dlq is calculated as part of calculating the voltage sensitivity of the power grid. The calculated voltage sensitivity may simply be one of these quantities or a combination of two or more of them. As an alternative, the voltage sensitivity may be calculated from one or more of these quantities in combination with one or more further quantities or parameters. Such further quantities or parameters may, e.g., include a Thevenin model voltage, a measured voltage, the active power, the reactive power, the grid frequency, dynamics of the system, e.g. of the power grid, such as damping ratio, etc. dV / dP represents expected change in grid voltage in response to changes in active power delivered to the power grid by the power generating unit, and dV / dlp represents expected change in grid voltage in response to changes in active current delivered to the power grid by the power generating unit. Thus, dV / dP as well as dV / dlp represents expected change in grid voltage in response to changes in the active part of the power delivered to the power grid by the power generating unit.

[0047] Similarly, dV / dQ represents expected change in grid voltage in response to changes in reactive power delivered to the power grid by the power generating unit, and dV / dlq represents expected change in grid voltage in response to changes in reactive current delivered to the power grid by the power generating unit. Thus, dV / dQ as well as dV / dlq represents expected change in grid voltage in response to changes in the reactive part of the power delivered to the power grid by the power generating unit. The expected change in grid voltage in response to changes in the active and / or the reactive part of the power delivered to the power grid by the power generating unit provides a suitable measure for the voltage sensitivity of the power grid.

[0048] As an alternative, the step of, for each operating point calculating a voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q) may comprise calculating a short-circuit ratio.

[0049] In the present context, the term 'short-circuit ratio' should be interpreted to mean the ratio of the short circuit apparent power (SCMVA) in the case of a line- line-line-ground (3LG) fault at the location on the grid where a power generator is connected, to the power rating of the power generator itself (GMW). Thus, the short-circuit ratio provides a suitable measure for the voltage sensitivity of the power grid.

[0050] The term 'grid strength' may be applied for describing the resiliency of the power grid to small changes in the vicinity of the grid location. From the side of a power generator, the grid strength may be regarded as related to the changes of the voltage the power generator encounters on its terminals, i.e. the voltage across the point of interconnection, as the power generator's current injection varies.

[0051] Alternatively or additionally, the step of calculating the voltage sensitivity may include calculating a grid impedance, a DC gain vector and / or any other suitable kind of quantity or parameter that provides information regarding the sensitivity of the grid voltage.

[0052] The method may further comprise the step of training the Al model. According to this embodiment the Al model is trained, based on suitable training data, before the steps described above are performed. The Al model may, e.g., be trained based on available historical data related to operation of power grids, in particular in interaction with relevant power generators. According to a second aspect the invention provides a controller for controlling at least one wind turbine based on a predicted voltage sensitivity of a power grid to which the at least one wind turbine is connected, wherein the controller is adapted to predict the voltage sensitivity of the power grid by means of a method according to the first aspect of the invention.

[0053] The controller according to the second aspect of the invention is, thus, adapted to perform at least part of the method according to the first aspect of the invention. The remarks set forth above with reference to the first aspect of the invention are therefore equally applicable here.

[0054] BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The invention will now be described in further detail with reference to the accompanying drawings in which

[0056] Fig. 1 is a block diagram illustrating a method according to an embodiment of the invention,

[0057] Figs. 2-8 are graphs illustrating various steps of a method according to an embodiment of the invention, and

[0058] Figs. 9 and 10 illustrate prediction of periods of highly voltage sensitive power grid in accordance with a method according to an embodiment of the invention.

[0059] DETAILED DESCRIPTION OF THE DRAWINGS

[0060] Fig. 1 is a block diagram illustrating a method according to an embodiment of the invention. A power generating unit, e.g. in the form of a wind turbine, a wind farm or a hybrid renewable power plant, is coupled to a power grid at a point of interconnection, and is operated so as to supply power to the power grid. During the operation of the power generating unit, data 1 related to a plurality of operating points is obtained, including data 1 in the form of active power (P) delivered by the power generating unit to the power grid, reactive power (Q) delivered by the power generating unit to the power grid, voltage (V) across the point of interconnection, nominal voltage (VnOm), i.e. a static voltage level defined by grid topology, and the point in time where these interrelated values were obtained, in the form of the hour of the day, the day of the week and / or the week of the year. Thus, for each of a plurality of points in time, interrelated values of active power (P), reactive power (Q), voltage (V) across the point of interconnection, and nominal voltage (VnOm) are obtained. Each of these interrelated collections of values represents an operating point.

[0061] The obtained values of active power (P), reactive power (Q), voltage (V) and nominal voltage (VnOm) are supplied to a first level estimator 2. In the first level estimator 2, a number of target values are calculated for each operating point. The target values include one or more of dV / dlq, dV / dlp, Thevenin voltage (Eo), an estimated short-circuit ratio (SCREst), a damping ratio for voltage dynamics of the power grid (Q, an oscillation frequency for voltage dynamics of the power grid (co), and quality of service (QoS). These target values, on their own or two of more of the target values in combination, represent a voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (IQ) / reactive power (Q).

[0062] The target values calculated by the first level estimator 2 are supplied to a second level neural network 3, comprising a trained Al model. The second level neural network 3 further receives the originally obtained data 1, including the points in time where the respective data 1 was obtained. Thus, the second level neural network 3 is in the position of interrelated or correlated values of active power (P), reactive power (Q), voltage (V) across the point of interconnection, nominal voltage (VnOm), voltage sensitivity of the power grid, and point in time, for a plurality of operating points, and thus for a plurality of discrete points in time.

[0063] Based on the available data, the second level neural network 3 identifies patterns related to voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), by means of the trained Al model. Based on the identified patterns, the second level neural network 3 derives time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q).

[0064] More particularly, from the information made available to the second level neural network 3, it is possible to identify, by means of the trained Al model, patterns related to which combinations of the obtained data 1 and points in time are likely to result in high or low voltage sensitivity of the power grid. Thus, the second level neural network 3 is able to provide such predictions, also for points in time and / or combinations of measured values that were not represented by the discrete operating points.

[0065] Figs. 2-8 are graphs illustrating various steps of a method according to an embodiment of the invention. Fig. 2 shows measured active power (P) delivered by a power generating unit to a power grid as a function of time. Fig. 3 shows measured reactive power (Q) delivered by the power generating unit to the power grid as a function of time, during the same time interval. Fig. 4 shows measured voltage (V) across a point of interconnection between the power generating unit and the power grid as a function of time, also during the same time interval. Thus, Figs. 2-4 represent interrelated values of active power (P), reactive power (Q) and volage (V) across the point of interconnection, as a function of time.

[0066] Fig. 5 shows voltage sensitivity of the power grid, in the form of estimated short-circuit ratio (SCREst), as a function of time, calculated from the information presented in the graphs of Figs. 2-4.

[0067] Fig. 6 shows active power (P) and estimated short-circuit ratio (SCREst) as a function of time, i.e. essentially a combination of the information represented by the graph of Fig. 2 and the information represented by the graph of Fig. 5. It can be seen that the graph of Fig. 6 includes a plurality of discrete points, each representing an operating point of the power grid.

[0068] Fig. 7 illustrates patterns identified from the information represented by the graph of Fig. 6. Essentially, the graph illustrated in Fig. 7 is a smoothed version of the graph of Fig. 6, with outliers removed. Thus, the graph of Fig. 7 represents expected voltage sensitivity of the power grid as a function of active power (P) and time, in the vicinity of the measured operating points represented by the interrelated data of Figs. 2-4.

[0069] Fig. 8 illustrates time dependent predictions of voltage sensitivity of the power grid, derived from the patterns represented in Fig. 7. It can be seen that the predictions include points in time and combinations of data which are not represented by the measured operating points. Furthermore, the graph illustrated in Fig. 8 essentially 'fills in the gaps' of Fig. 7, while taking into account the dynamic behaviour of the power grid underlying the information represented by the graph of Fig. 7. Thus, the predictions illustrated in Fig. 8 provides a significantly more complete picture than the information represented by the graph of Fig. 6, and particularly of the separated information represented by the respective graphs of Figs. 2-5.

[0070] Accordingly, the predictions of Fig. 8 can, e.g., be applied for predicting periods of time where there is a particularly high risk of highly voltage sensitive power grid, and it is therefore possible to initiate appropriate measures before such periods occur, thus reducing the risk of an unstable power grid. The predicted periods of time could, e.g., be specific times of the day / night, i.e. during a diurnal cycle, specific days of the week, specific weeks of the year, etc.

[0071] Figs. 9 and 10 illustrate prediction of periods of highly voltage sensitive power grid in accordance with a method according to an embodiment of the invention.

[0072] Fig. 9 shows active power (P) and voltage sensitivity of the power grid, in the form of dV / dlq, as a function of time, in the form of the hour of day, i.e. with regard to a diurnal cycle.

[0073] The point marked 'A' represents an operating point where dV / dlq is high, e.g. above 0.5. Thus, under these circumstances, the voltage (V) of the power grid changes significantly in response to changes in the reactive current (IQ) delivered by the power generating unit to the power grid. This indicates a highly voltage sensitive power grid, and point 'A' therefore represents an example of circumstances under which a highly voltage sensitive power grid can be expected.

[0074] The point marked 'B' represents an operating point where dV / dlq changes dramatically within a short period of time, and / or in response to small changes in active power (P). This also indicates a risk of a highly voltage sensitive power grid, and point 'B' therefore also represents an example of circumstances under which a highly voltage sensitive power grid can be expected, even though dV / dlq is not particularly high at this point.

[0075] The point marked 'C' represents an operating point where dV / dlq is not particularly high, and where dV / dlq does not change significantly as a function of time or in response to changes in active power (P). Thus, the voltage sensitivity of the power grid at this point may be regarded as low, and therefore point 'C' represents an example of circumstances under which the power grid is not expected to be highly voltage sensitive.

[0076] Fig. 10 illustrates classification of highly voltage sensitive power grid and normal grid situation, respectively, based on two target values of a trained Al model.

[0077] For instance, in the case that dV / dlq is higher than 0.1 and the Thevenin voltage is smaller than 0.98pu, and SCREst is above a specified complex curve, e.g. defined by a polynomial, then the power grid may be classified as 'highly voltage sensitive'. When one or more of these conditions are not fulfilled, the power grid may be classified as 'not highly voltage sensitive'.

Claims

CLAIMS1. A method for predicting voltage sensitivity of a power grid, the power grid being coupled to a power generating unit at a point of interconnection, the method comprising the steps of:- obtaining data (1) related to a plurality of operating points of the power grid, each operating point defining a voltage (Vmeas) across the point of interconnection, active current (Ip) / active power (P) delivered by the power generating unit to the power grid, reactive current (Iq) / reactive power (Q) delivered by the power generating unit to the power grid, and a point in time where the data (1) related to the operating point was obtained,- for each operating point, calculating a voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), corresponding to the operating point, based at least on the voltage (Vmeas), the active current (Ip) / active power (P) and the reactive current (Iq) / reactive power (Q), and using a model representation of the power grid,- supplying the data (1) related to the plurality operating points, including the points in time where the data (1) was obtained, and the corresponding calculated voltage sensitivities to a trained artificial intelligence (Al) model (3),- identifying patterns related to voltage sensitivity of the power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (Iq) / reactive power (Q), by means of the Al model (3), and- deriving time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), based on the identified patterns.

2. A method according to claim 1, wherein the time dependent predictions of voltage sensitivity of the power grid reflect estimated variations in voltage sensitivity of the power grid in response to changes in the active current(lp) / active power (P) and / or the reactive current (Iq) / reactive power (Q) during a diurnal cycle, a weekly cycle and / or a yearly cycle.

3. A method according to claim 1 or 2, further comprising the step of predicting periods of highly voltage sensitive power grid, based on the derived time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current(lq) / reactive power (Q).

4. A method according to claim 3, further comprising the step of proactively initiating measures for counteracting highly voltage sensitive power grid prior to predicted periods of highly voltage sensitive power grid in response to changes in the active current (IP) / active power (P) and / or the reactive current (Iq) / reactive power (Q).

5. A method according to any of the preceding claims, further comprising the step of updating the Al model (3) based on correlated measurements of voltage ( meas) across the point of interconnection, active current (Ip) / active power (P) delivered by the power generating unit to the power grid, and reactive current (Iq) / reactive power (Q) delivered by the power generating unit to the power grid, and using reinforced learning.

6. A method according to claim 5, further comprising repeating the steps of identifying patterns related to voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), and deriving time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q), based on the updated Al model (3).

7. A method according to any of the preceding claims, wherein the power generating unit is a wind turbine, a wind farm or a hybrid renewable powerplant, and wherein the method further comprises the step of controlling the power generating unit to provide a power feed into the power grid, based on the derived time dependent predictions of voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (IQ) / reactive power (Q).

8. A method according to any of the preceding claims, wherein the step of, for each operating point calculating a voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q) comprises calculating dV / dP, dV / dlp, dV / dQ and / or dV / dlq.

9. A method according to any of claims 1-7, wherein the step of, for each operating point calculating a voltage sensitivity of the power grid in response to changes in the active current (Ip) / active power (P) and / or the reactive current (Iq) / reactive power (Q) comprises calculating a short-circuit ratio.

10. A method according to any of the preceding claims, further comprising the step of training the Al model (3).

11. A controller for controlling at least one wind turbine based on a predicted voltage sensitivity of a power grid to which the at least one wind turbine is connected, wherein the controller is adapted to predict the voltage sensitivity of the power grid by means of a method according to any of the preceding claims.

Images