T 0639/20 (Wind-farm optimisation/GENERAL ELECTRIC) 30-03-2023

1. Technical background

1.1 The opposed patent relates to a wind farm, such as wind farm 100 having wind turbines 102 and rotor blades 104 (see Figure 1 reproduced below).

FORMULA/TABLE/GRAPHIC

1.2 The opposed patent particularly concerns wind-energy capture optimisation in such a wind farm. It aims to do so by maximising the wind farm's "annual energy production" while minimising negative effects such as the fatigue load on wind turbines 102. This (constrained) maximisation is supposed to output optimal control settings for wind turbines 102. Using these optimal control settings, rotor blades 104 may, for instance, be swivelled and aligned relative to the prevailing wind direction.

1.3 The constrained maximisation is based on "farm-level wake models" which involve "wake parameters". These parameters will typically include ambient conditions, geometrical layout of the wind farm and/or operational information of the wind turbines. According to the invention, some of these wake parameters, such as the ambient conditions, are continuously monitored, e.g. by monitoring devices 110. These monitored values are then used to identify aerodynamically interacting sets of wind turbines and to estimate corresponding (set-wise) wake interactions. The set-wise estimation of these wake interactions, in turn, would, according to the opposed patent, make it possible to generate the

farm-level wake models in real-time.

2. Main request: claim 1 - construction

Throughout the appeal proceedings, the parties have put a particular emphasis on the construction of features A, (b) to (d) and (f). The board construes these features as follows.

2.1 The board agrees with the respondent that the "optimizing" step according to feature A can relate to a plurality of goals. As the respondent correctly brought forward, the wind-farm operation optimisation according to the method of granted claim 1 could aim

- to increase the lifetime of wind turbines of the "wind farm" of feature A,

- to maximise the wind farm's energy output or

- to reduce the wind farm's maintenance.

Granted claim 1, however, does not specify any cost function, which the skilled reader would typically expect to be minimised and yield an optimised operation in the sense of feature A. As a result, granted claim 1 does not allow to identify which goal the "optimizing" step of this feature is directed to. Similar comments apply to the "farm-level predictive wake model" of features (c) and (d), about which granted claim 1 also specifies no details. The farm-level predictive wake model of these features is a mere "black box" that does not allow to draw any conclusion as to which of the wind farm's operational aspects are supposed to be optimised in the claimed method.

2.2 For the reasons set out in points 2.2.1 and 2.2.2 below, the board does not share the respondent's view that the meaning of features (b) and (d) is to be confined to a grouping, resulting in "sets", of wind turbines that interact with each other. Likewise, the respondent could not convince the board that feature (d) would require to adjust the control settings for these interacting wind turbines in a

set-wise way, rather than for each of the wind turbines individually. The board acknowledges that the first sentence of paragraph [0015] of the opposed patent mentions a "set-wise evaluation" in this respect, but no such evaluation is apparent from claim 1 as granted taken by itself.

2.2.1 In the board's view, feature (b) merely requires to identify, i.e. "to indicate", sets of interacting wind turbines, rather than to (actively) group these wind turbines with the purpose of adjusting their control settings "en masse" within a particular set. The board notes in this respect that the term "set" can refer to a "pair". The second sentence of paragraph [0045] of the opposed patent is, moreover, in line with this interpretation. This means that feature (b) can, within the wind farm of feature A, boil down to a mere identification of those pairs of wind turbines that are interacting under currently prevailing weather conditions.

2.2.2 Moreover, feature (d) is silent about whether the control settings are adjusted for the new sets of interacting wind turbines on an "individual basis", i.e. wind turbine per wind turbine, or whether the control settings are adjusted for all of the interacting wind turbines of a particular new set in one go.

2.3 The board holds that the phrase "developing a

farm-level predictive wake model" according to feature (c) can, in combination with feature (f), be taken to mean that, for each repetition in accordance with the latter feature, a new farm-level predictive wake model is "developed" in the sense that it is "generated" by the claimed method. For each repetition, this "new farm-level predictive wake model" could either be the same as before or could be a different one. The board acknowledges that, by virtue of feature (c), such generating would have to be "based on the new values" of the wake parameters. However, this does not necessarily mean that the mathematical function underlying the newly generated farm-level predictive wake model must depend, taken by itself, on all of these new values. This can be illustrated by an example which the respondent brought forward during the oral proceedings before the board. In this example, two farm-level predictive wake models were used in the claimed method, one for a dominant wind direction (e.g. "North") and another one for a secondary wind direction (e.g. "East"). If the wind changes from "North" to "East", feature (c) could result in a mere change of farm-level predictive wake model, without this model being explicitly dependent on the wind direction.

As an aside, the board observes that the opposed patent is not specific regarding the precise form of this mathematical function. The appellant correctly pointed out that the opposed patent discloses, at most, one single equation that may underlie the "farm-level predictive wake model" of feature (c), namely equation (1) indicated in paragraph [0049] of the opposed patent. The board emphasises that the term "others" in this equation (1) does not necessarily encompass all of the other wake parameters considered in the opposed patent. Moreover, the opposed patent is silent about how the different parameters mentioned in the argument of function f of equation (1) are actually weighted. In sum, the skilled reader would understand the "developing" step of feature (c) to encompass more than a mere "updating" in view of currently prevailing weather conditions as suggested by paragraphs [0049] to [0051] of the opposed patent.

2.4 The board further agrees with the respondent's interpretation of the term "reference sets of interacting wind turbines" of feature (c) as meaning "certain groupings of upstream and downstream turbines" in the sense that the interacting wind turbines comprised in these reference sets will typically involve one or more upstream wind turbines that impact the operation of at least one downstream wind turbine. The skilled reader would also readily understand from the expression "historical values of the wake parameters corresponding to reference sets of interacting wind turbines" according to feature (c) that these "reference sets" must relate to sets that have been determined previously, e.g. when calibrating the "wind farm" of feature A or as determined via the "identifying" step according to feature (b) in a previous iteration of the claimed method.

2.5 Regarding feature (f), the board does not agree with Reasons 12 of the appealed decision that the clause "when a change in a monitored value of one or more of the wake parameters is outside a corresponding threshold" necessarily means that the repeating step of feature (f) would take place only if this change is outside the corresponding threshold or when this change would be "significant".

In this respect, the respondent, on the one hand, convincingly argued that the interpretation of the term "when" can have a temporal as well as a causal aspect. The board holds that the interpretation of the term "when" in the sense of "while" fits within this temporal aspect. On the other hand, the board does not share the respondent's view as to how this clause would limit feature (f): the mere presence of this clause in feature (f) does not necessarily convey to the skilled reader that the "threshold" could not be chosen arbitrarily. Granted claim 1 is, after all, silent about which value to attribute to this threshold. It does, in particular, not mandatorily require to set this threshold to 5% as suggested by the respondent based on paragraph [0054] of the opposed patent.

The skilled reader would, in the board's view, readily understand that this "threshold" according to feature (f) determines the sensitivity of the repetitive scheme underlying granted claim 1: a value close to zero means a high sensitivity to changes in the monitored value according to feature (f), whereas a high value represents a low sensitivity, i.e. a more robust scheme.

The skilled reader could, in fact, even eliminate the repetition involved in feature (f) altogether. It would make technical sense for the skilled reader to do so, for instance, when performing a test run of the claimed optimisation method. The respondent correctly argued in this respect that any method for optimising the operation of a wind farm will typically consist of several stages. The time lapse between these stages is not necessarily stringently defined. In one scenario, it could be that, in a first "initial-test" stage, the wind turbines of a particular wind farm are operated with a first set of wake parameters underlying the "farm-level predictive wake model" of feature (c). The board concurs with the respondent that such an "initial-test" stage may last for several weeks and that it can be used

- to gather data on, for instance, blade fouling,

- to perform sensor calibration or

- to study corrosion.

To allow for technically meaningful test results within such a scenario, the skilled reader would, in the board's view, readily choose the "threshold" of feature (f) to be sufficiently high. In this way, the change in the monitored value of any of the wake parameters would always be below this threshold and the control settings would remain the same for a sufficiently long period. The board therefore does not agree with the respondent that a high threshold would be contrary to the purpose of the claimed method to optimise a wind-farm operation.

3. Main request: claim 1 - novelty

3.1 The board holds that paragraphs [0004], [0006], [0008], [0028] and [0030] to [0036] as well as Figure 5 of document D2 indeed disclose features A and (a) to (f) of present claim 1.

3.2 The respondent questioned the disclosure of features (b) to (d) and (f) by document D2. The board makes the following comments in that regard:

3.2.1 In particular, feature (b) is disclosed in paragraph [0033] of D2 in view of three clauses, namely (emphasis added in all three clauses):

- "[t]he second component of the algorithm determines which upstream turbine 12 causes a wake";

- "[a]ny upstream turbine 12 that does not cause a wake that impacts a downstream turbine 14 will not be adjusted";

- "[t]he sequence of turbines to be switched is also determined by the algorithm".

These clauses necessarily imply the identification of new sets of interacting wind turbines. Otherwise, the wake optimisation algorithm shown in paragraphs [0032] to [0035] of D2 could not recognise which turbines need to be adjusted and in which sequence this should occur.

The board was not convinced by the respondent's argument that the "set-wise analysis" according to paragraph [0015] of the opposed patent differed from what the respondent referred to as the "classical analysis" of paragraph [0033] of D2. This is because the first two sentences of this latter paragraph identify those downstream wind turbines 14 that are affected by the wake of a particular upstream wind turbine 12. In agreement with the board's construction of feature (b) as explained in point 2.2.1 above, this means that "new sets of interacting wind turbines" are likewise identified in the system of D2.

3.2.2 Feature (c) is shown by the wake optimisation algorithm described in paragraphs [0032] to [0035] and illustrated by Figure 5 of D2. The latter unequivocally teaches in combination with paragraph [0034] of D2 calculating new turbine settings based on a look-up table which is based on inputs such as "ambient wind conditions" and "turbine operating parameters". These inputs are in turn continuously measured and (at least partly) predicted as set out in paragraphs [0028] and [0031] of D2. The new turbine settings allow "to increase overall windpark [i.e. farm-level] energy capture", as is apparent from paragraph [0034] of D2. The board holds that this look-up table is the place where the "farm-level predictive wake model" underlying D2's wake optimisation algorithm is stored. This is similar to how "historical wake models" are stored according to paragraph [0052] of the opposed patent. Moreover, the "look-up table" according to paragraph [0034] of D2 is updated in an iterative fashion. By definition, an iterative approach calculates new values of a particular parameter (set) based on previous values of this parameter (set). The skilled reader would thus understand that the iterations in D2 must correspond to different cycles of executing the wake optimisation algorithm in which the meteorological and turbine operating conditions, i.e. the "wake parameters" of features (a) and (c), are continuously detected as explained in paragraph [0031] of D2. It would further be immediately apparent for the skilled reader that, during normal operation, only a sub-set of the look-up table's inputs will have to be updated during a particular cycle. As a result, in each cycle, the look-up table will be based on "new values" and on "historical values of the wake parameters". The latter constitute, by virtue of being stored in the look-up table, "historical wake models" which were determined during the previous cycles of the iterative approach. These previous cycles must, again by definition of an iterative approach, have identified their own sets of interacting wind turbines as set out in paragraph [0033] of D2, which correspond to the "reference sets" of feature (c) as construed in point 2.4 above. It goes without saying that these reference sets will include "historical values of the wake parameters" associated with them. These historical values will, in turn, determine "historical wake models" according to feature (c).

The respondent contested that the term "iterative fashion" of the last sentence of paragraph [0034] of D2 would necessarily involve a calculation of new values based on previous values. Rather, in the respondent's view, this term was to be construed, on the basis of the Merriam-Webster dictionary and within the context of D2, as a mere "repetition" without any renewed identification of "sets of interacting wind turbines". The respondent concluded that D2 did not directly and unambiguously disclose the "reference sets" of feature (c).

The board is not convinced that the skilled reader would indeed consider the terms "iterative" and "repetitive" to be equivalent, even if the notorious "man in the street" would arguably do so. This is because the skilled reader from the field of optimisation algorithms such as the algorithm disclosed in paragraphs [0032] to [0034] of D2 would have no doubt that the former term comprises the connotation of obtaining successively closer approximations to the solution of a particular problem, which the latter term does not have. Moreover, it would also not make any technical sense for the skilled reader to simply repeat the three components of the wake optimisation algorithm of paragraphs [0032] to [0034] of D2 for each and every change in the meteorological (i.e. ambient-wind) conditions and wind-turbine operating parameters that play a role in that wake optimisation algorithm. For instance, if a downstream wind turbine's operating status is changed from not running to running, the skilled reader would immediately understand that the second component of the wake optimisation algorithm need not be repeated for each and every wind turbine of the wind farm. Rather, it would be unequivocal for the skilled reader that only those upstream wind turbines 12 that could cause a wake impacting the now running downstream wind turbine will be identified to form part of the new sets of interacting wind turbines. All other sets of interacting wind turbines which are not affected by switching on the downstream wind turbine would remain the same as determined before this switching on. Therefore, the skilled reader would immediately dismiss the respondent's interpretation of the term "iterative" as an academic rather than a practical one. Therefore, D2 in fact discloses the "developing" step according to feature (c) in the same sense of "updating" as defined in paragraphs [0049] to [0051] of the opposed patent (cf. the last paragraph of point 2.3 above).

3.2.3 Furthermore, feature (d) is palpably disclosed in paragraphs [0030], [0031] and [0034] of D2.

3.2.4 Lastly, feature (f) is also disclosed by virtue of the "iterative fashion" according to paragraph [0034] of D2. As apparent from the term "turbine operating parameters" in the penultimate sentence of this paragraph, this iterative update is triggered, for instance, if there is a discrete change in the wind-turbine operating status, e.g. from running to not running or vice versa. Such a discrete change will evidently be outside of some non-zero value. The board also notes, for the sake of completeness, that the "repeating" step according to feature (f) is not necessarily performed and therefore does not necessarily limit the claimed method under all circumstances that would occur to the skilled reader (cf. the last two paragraphs of point 2.5 above).

3.3 Hence, the board agrees with the appellant that claim 1 as granted lacks novelty (Article 54 EPC), contrary to the opposition division's conclusions.

4. Auxiliary requests 1 to 5: admittance

4.1 The appealed decision was not based on auxiliary requests 1 to 5 within the meaning of Article 12(2) RPBA 2020 ("on which the decision under appeal was based") for the simple reason that the opposition division rejected the opposition, i.e. allowed the respondent's higher-ranking main request (cf. appealed decision, point 7). Therefore, the respondent was supposed to demonstrate that these requests were admissibly filed and maintained in the proceedings before the opposition division (Article 12(4), first sentence, RPBA 2020). If this is done, these auxiliary requests form part of the appeal proceedings and cannot be regarded as an "amendment". Moreover, Article 12(3) RPBA 2020 requires that the written reply to the appeal contains the respondent's "complete appeal case" and specifies "expressly all the requests relied on".

4.2 In the case in hand, however, the respondent did not make any submissions to that effect in its written reply to the appeal. Only in its response to the board's preliminary opinion under Article 15(1) RPBA 2020 did the respondent address that matter by contesting the board's discretion not to admit auxiliary requests 1 to 5 (letter of 3 March 2023, pages 4 to 6). Therefore, the five auxiliary requests on file are to be regarded as "amendments", and may be admitted only at the discretion of the board (Article 12(4) RPBA 2020).

4.3 Given that, in the present case, the board is convinced without any burden that the respondent indeed admissibly filed and maintained the present auxiliary requests during the opposition proceedings (albeit not demonstrated with the written reply to the statement of grounds of appeal), and since the fact that the opposition division did not need to examine and decide upon those requests cannot be imputed to the respondent (cf. point I above), it seems to be just and fair to admit them into the proceedings.

4.4 Hence, the board decided to admit auxiliary requests 1 to 5 into the appeal proceedings (Article 12(4), second sentence, RPBA 2020).

5. Remittal of the case to the opposition division

5.1 Given the extensive claim construction conducted during the appeal proceedings (which deviated substantially from that of the appealed decision), the respondent favoured a remittal of the case to the opposition division. The opposition division could then, in the respondent's view, decide upon the case using the claim construction as adopted by the board (see points 2.1 to 2.5 above). Also, further prior-art documents on file may potentially become relevant for the assessment of novelty and/or inventive step as regards auxiliary requests 1 to 5.

5.2 The appellant had no objections against a remittal to the opposition division.

5.3 In view of the above, it is justified to remit the case to the opposition division, so that the latter may assess the allowability of auxiliary requests 1 to 5 first. In that context, the board only adds that, where Article 56 EPC is considered in that assessment, it could in particular be relevant, for establishing which technical effect e.g. feature (k) credibly achieves over the whole claimed scope, to determine how the skilled reader would construe the term "sparse tree structure" used therein and to which extent this term is indeed reflected in Figure 3 of the opposed patent.

5.4 Hence, "special reasons" present themselves for remitting the case to the opposition division for further prosecution (Article 11 RPBA 2020 and Article 111(1) EPC).