Determining endurance performance by a linear model

Abstract

Example embodiments relate to a computer-implemented method for determining endurance performance of a subject based on sensor measurements; the computer-implemented method comprising: obtaining (201) workload data measured by at least one workload sensor, comprising a time series (212) of the workload exerted by the subject during one or more physical activities; obtaining (202) heart rate data measured by at least one heart rate sensor, comprising a time series (211) of the heart rate of the subject during the one or more physical activities; selecting (203) time intervals (213 - 219) within the workload data and the heart rate data based on a set of selection criteria (204); for the respective time intervals (213 - 219), determining (205) an average workload (207) and, for a final portion of the respective time intervals (213 - 219), determining (205) an average heart rate (206); and determining (208) at least a portion of a workload-heart rate relationship (221), indicative for the endurance performance of the subject, by fitting a linear equation (224) to the average workloads and the average heart rates (230) of the respective time intervals (213 - 219).

Description

[0001] DETERMINING ENDURANCE PERFORMANCE BY A LINEAR MODEL Field of the Invention

[0001] The present invention generally relates to determining the enduranceperformance of a subject using sensor data, in particular to determining at least aportion of a workload-heart rate relationship of the subject based on heart rate dataand workload data. Background of the Invention

[0002] The endurance performance of endurance athletes is, amongst others,characterized by an aerobic lactate threshold and an anaerobic lactate threshold, alsoreferred to as LT1 and LT2 respectively. The aerobic lactate threshold expresses theminimum workload an endurance athlete has to exert to reach lactate production levelsthat exceed his / her physiological baseline levels. The anaerobic lactate thresholdexpresses the workload at which lactate production starts to exceed lactate clearance, thereby resulting in rapid lactate accumulation.

[0003] Training plans and structured workouts are typically constructed based on thelactate thresholds of the athlete, in particular on the workload an athlete can exert atthe anaerobic threshold. However, accurately determining the lactate thresholdstypically requires performing an exercise test at incremental workloads during whichblood samples of the athlete are regularly collected. The lactate concentration withinthe blood samples can then be related to the incremental workloads. These exercisetests have the problem that they are costly, typically require the athlete to perform amaximal effort, have to be performed in a controlled environment, and are generallyperceived as unpleasant by most athletes. As such, these tests are rarely performedby recreational athletes and even infrequently performed by professional athletes.

[0004] Some methods exist to approximate the anaerobic lactate threshold with a proxyvalue based on a dedicated test protocol without collecting blood samples. Forexample, in cycling, a common approach is to perform a ramp-up test where the workload is incrementally increased until failure. This allows determining the power acyclist can hold for an hour, i.e. the functional threshold power, which can be used asan approximation of the workload at the anaerobic threshold. However, this has theproblem that it requires the athlete to perform a maximal effort during a dedicated test,which is undesired as it results in undue fatigue of the athlete thereby interfering withoptimal endurance training. This further makes it unsuitable for continuousperformance tracking as the negative effects of frequently repeating such maximalefforts are even more undesirable.

[0005] Instead of a dedicated test protocol, some methods attempt to approximate thelactate thresholds of an athlete based on data obtained during regular workouts, e.g.by determining a power-duration curve or by analysing the heart rate variability.However, this has the problem that athletes typically do not spend a lot of time in high-intensity zones during regular workouts and therefore still require incorporating regularmaximal efforts in their training plan to enable approximation of the anaerobicthreshold. Additionally, these approximation methods are typically sensitive to outliersand noise.

[0006] It is a further problem that these approximation methods only allowapproximating the anaerobic threshold in an individualized manner, and that theaerobic threshold is merely estimated in a non-individualized manner using population- based heuristics. Summary of the Invention

[0007] It is an object of the present invention, amongst others, to solve or alleviateabove identified problems and challenges by determining the endurance performanceof a subject based on sensor measurements obtained during regular unstructuredworkouts without requiring regular maximal efforts, regular high-intensity efforts, orinvasive blood sampling.

[0008] According to a first aspect, this object is achieved by a computer-implementedmethod for determining endurance performance of a subject based on sensormeasurements. The computer-implemented method comprising:‒ obtaining workload data measured by at least one workload sensor,comprising a time series of the workload exerted by the subject during one or more physical activities; ‒obtaining heart rate data measured by at least one heart rate sensor,comprising a time series of the heart rate of the subject during the one or more physical activities; ‒selecting time intervals within the workload data and the heart rate databased on a set of selection criteria; ‒for the respective time intervals, determining an average workload and, fora final portion of the respective time intervals, determining an average heart rate; and ‒determining at least a portion of a workload-heart rate relationship,indicative for the endurance performance of the subject, by fitting a linear equation to the average workloads and the average heart rates of the respective time intervals.

[0009] The workload data expresses an external effort the subject exerts during theone or more physical activities. The workload data may therefore depend on theperformed endurance sport, e.g. power in cycling or rowing, and pace in running orswimming. The workload data may further account for other external effort metricsduring an endurance sport, e.g. elevation may be considered in addition to power or pace. The workload data is obtained by at least one workload sensor such as, for example, crank-based power meters in cycling, pedal-based power meters in cycling,a power meter in rowing, or a smart watch configured to measure pace in running orswimming.

[0010] The heart rate data expresses an internal effort the subject exerts during theone or more physical activities. In other words, the heart rate data is the cardiovascular response of the subject to the workload of a physical activity. Thus, the workload data and the heart rate data are time series with corresponding time steps. The heart rate data is obtained by at least one heart rate sensor, e.g. a pulse oximeter or photoplethysmography, PPG, sensor provided within a smart watch; or an electrocardiographic sensor provided within a chest strap. The at least one workloadsensor and the at least one heart rate sensor may further be provided within a singledevice, e.g. a smart watch. It is an advantage that the computer-implemented methodis agnostic to the used at least one heart rate sensor and the at least one workload sensor.

[0011] The workload data and heart rate data are measured during respective one ormore physical activities, also referred to as workouts. A plurality of time intervals areselected from the obtained workload data and heart rate data. The time intervals maybe intervals spread across the one or more physical activities. The number of time intervals selected from each of the one or more physical activities may, but need not,be equal. For example, five time intervals may be selected from a first physical activitywhile only one time interval is selected from a second physical activity. The timeintervals may have a predetermined fixed length, e.g. 2 minutes. Time intervals areselected based on a set of selection criteria, thereby limiting the amount of heart ratedata and workload data to segments that are suitable for determining the workload-heart rate relationship. This has the advantage that it reduces the computational loadand, thus, improves the efficiency of determining the workload-heart rate relationship.

[0012] For the respective time intervals, a workload-heart rate pair is then obtainedcomprising an average workload and an average heart rate. The average workloadsof the respective workload-heart rate pairs are obtained by determining the averageworkload during the entire respective time intervals. The average heart rates of therespective workload-heart rate pairs are obtained by determining the average heartrate during a final portion of the respective time intervals. This final portion refers to atrailing portion or ending portion of the time interval where the heart rate hassubstantially stabilized, i.e. where the heart rate is substantially free of delayedphysiological responses such as cardiac lag or heart rate lag. The final portion may,for example, be the second half of the time interval.

[0013] In doing so, a set of workload-heart rate pairs is obtained. A linear equation isthen fitted to this set of workload-heart rate pairs to determine or approximate at leasta portion of a workload-heart rate relationship. A workload-heart rate relationshipexpresses the relationship between the workload exerted by a subject during a physical activity and the resulting heart rate of the subject. The workload-heart rate relationship is a good indicator of the endurance performance of the subject, as the workload-heart rate relationship can provide insight into the efficiency, endurance, fatigue resistance, training adaptations, and lactate thresholds of the subject. By fittingthe linear equation to the workload-heart rate pairs, the typically substantially linearportion of the workload-heart rate relationship between the aerobic threshold and theanaerobic threshold can be approximated.

[0014] The computer-implemented method thus allows to determine the enduranceperformance of a subject without interfering with endurance training, as performingdedicated maximal efforts and / or high-intensity efforts is unnecessary. This has thefurther advantage that it makes the computer-implemented method suitable for continuous performance tracking. It is a further advantage that the performance of the subject can be determined and tracked based on regular physical activities performed in real-world environments, i.e. without structured training sessions or withoutperforming dedicated test protocols in a controlled lab environment.

[0015] According to an example embodiment, the set of selection criteria may compriseat least one of a maximum variation in the workload data; a maximum workloadimbalance; a maximum altitude during the time interval; an increase in average heartrate relative to a preceding interval; a minimum energy expenditure; a maximumenergy expenditure; a minimum average heart rate; a maximum average heart rate; amaximum rate of change in heart rate; and a maximum rate of change in workload.

[0016] According to an example embodiment, the workload data and the heart ratedata may comprise a history of workload data and heart rate data measured during a predetermined number of previous physical activities or measured during previous physical activities performed within a predetermined preceding period.

[0017] In other words, the one or more physical activities containing the workload andheart rate data may have been performed by the subject within a historical timewindow. This historical time window may be defined as a predetermined number of previous physical activities, e.g. the last 15 workouts. Alternatively, the historical time window may be defined as a predetermined period, e.g. the last 28 days. The historicaltime window may preferably go back far enough in time as to include enough data forfitting the linear equation reliably, without going too far back in time such that changesin the subject’s endurance performance remain substantially undetectable within thedata.

[0018] According to an example embodiment, the computer-implemented method mayfurther comprise updating the workload-heart rate relationship based on workload data and heart rate data measured during a next physical activity.

[0019] Thus, the workload-heart rate relationship may initially be determined based ona history of workload and heart rate data measured during the historical window. Thereafter, the determined workload-heart rate relationship may be updated based on workload and heart rate data of a most recent physical activity, e.g. after the subjectcompletes a new workout and saves his / her measurements of the activity to a server.Updating the workload-heart rate relationship may be achieved by selecting time intervals within the most recent physical activity, determining new workload-heart ratepairs for those selected time intervals, and refitting the linear equation to an updatedset of workload-heart rate pairs that comprises the historical workload-heart rate pairsin addition to the new workload-heart rate pairs. This allows continuous tracking of theendurance performance without interfering with endurance training.

[0020] According to an example embodiment, the computer-implemented method mayfurther comprise discarding at least one time segment from the workload data and the heart rate data that includes one or more outliers within the workload data and / or the heart rate data.

[0021] In other words, if a time segment of the workload data and / or a time segmentof the heart rate data include one or more outliers, that time segment may be discardedfrom both the workload time series and the heart rate time series as these time serieshave corresponding time steps. A time segment may correspond to the entire durationof a physical activity or to a portion of the physical activity. An example of such a timesegment may be a segment of the heart rate data with null values or extremely lowvalues indicative of a sensor fault or malfunction. Detecting the one or more outliersmay be achieved by any outlier detection technique, e.g. heuristics based on domainknowledge, deep learning models, univariate detection methods, multivariate detectionmethods, or a combination thereof. Outliers may be detected within the heart rate databy itself, within the workload data by itself, or within a combination of the heart rate and workload data, e.g. when the combination of the heart rate and workload data indicatesthat the subject exerted a very high workload at an unrealistic low heart rate.

[0022] According to an example embodiment, the computer-implemented method mayfurther comprise obtaining an estimated workload of a next physical activity based on the heart rate data measured during the next physical activity and the determined workload-heart rate relationship; and detecting one or more outliers within the workload data by comparing the estimated workload with the workload data measured during the next physical activity.

[0023] The heart rate measured during a most recent physical activity may thus beused to predict the workload exerted during that most recent physical activity based onthe previously determined workload-heart rate relationship, i.e. based on the fittedlinear equation. This predicted workload may then be compared to the measuredworkload during the physical activity. An outlier may be detected if the differencebetween the predicted workload and the measured workload, i.e. the prediction error,exceeds a predetermined threshold. The entire physical activity may further bediscarded if the average prediction error exceeds a predetermined absolute average threshold.

[0024] This can further allow detecting overperformance and underperformance duringthe most recent physical activity, i.e. when the subject performed particularly good orbad during a workout. Positive average prediction errors may indicate that the subjectwas able to perform better than expected for the same heart rate, i.e. overperformance. Negative average prediction errors may indicate that the subject performed worse than expected for the same heart rate, i.e. underperformance.

[0025] According to an example embodiment, the computer-implemented method mayfurther comprise obtaining an estimated heart rate of a next physical activity based on the workload data measured during the next physical activity and the determined workload-heart rate relationship; and detecting one or more outliers within the heart rate data by comparing the estimated heart rate with the heart rate data measured during the next physical activity.

[0026] The workload measured during a most recent physical activity may thus be usedto predict the heart rate as a result of exerting said workload during that most recentphysical activity based on the previously determined workload-heart rate relationship, i.e. based on the fitted linear equation. This predicted heart rate may then be compared to the measured heart rate during the physical activity. An outlier may be detected if the difference between the predicted heart rate and the measured heart rate, i.e. the prediction error, exceeds a predetermined threshold. The entire physical activity may further be discarded if the average prediction error exceeds a predetermined absolute average threshold.

[0027] This can further allow detecting overperformance and underperformance duringthe most recent physical activity, i.e. when the subject performed particularly good orbad during a workout. Positive average prediction errors may indicate that the subjectwas able to perform better than expected for the same heart rate, i.e. overperformance. Negative average prediction errors may indicate that the subject performed worse than expected for the same heart rate, i.e. underperformance.

[0028] According to an example embodiment, the determined portion of the workload-heart rate relationship may comprise the workload-heart rate relationship between anaerobic threshold and an anaerobic threshold.

[0029] According to an example embodiment, the computer-implemented method mayfurther comprise determining the workload of the subject at an aerobic threshold based on the determined workload-heart rate relationship and a predetermined heart rate of the subject at the aerobic threshold.

[0030] In other words, the heart rate of the subject at the aerobic threshold should beknown. This predetermined heart rate may, for example, be obtained by performing a single lactate test in a lab. Thereafter, the predetermined heart rate at the aerobic threshold may be used to derive the workload at the aerobic threshold from thedetermined workload-heart rate relationship, i.e. by means of the fitted linear equation.This allows determining the workload at the aerobic threshold without repeating alactate test and without performing dedicated test protocols in controlledcircumstances, i.e. based on regular physical activities. This has the advantage that the workload at the aerobic threshold can be determined reliably and in anindividualized manner. It is a further advantage that repeated invasive blood samplingcan be avoided.

[0031] According to an example embodiment, the computer-implemented method mayfurther comprise determining the workload of the subject at an anaerobic threshold based on the determined workload-heart rate relationship and a predetermined heart rate of the subject at the anaerobic threshold.

[0032] In other words, the heart rate of the subject at the anaerobic threshold shouldbe known. This predetermined heart rate may, for example, be obtained by performing a single lactate test in a lab. Thereafter, the predetermined heart rate at the anaerobic threshold may be used to derive the workload at the anaerobic threshold from the determined workload-heart rate relationship, i.e. by means of the fitted linear equation. This allows determining the workload at the anaerobic threshold without repeating alactate test and without performing dedicated test protocols in controlledcircumstances, i.e. based on regular physical activities. This has the advantage that the workload at the anaerobic threshold can be determined reliably without having toperform dedicated maximal efforts. It is a further advantage that repeated invasiveblood sampling can be avoided.

[0033] According to an example embodiment, the computer-implemented method maycomprise determining parameter values of a fitness-fatigue model by minimizing a difference between the workload at the anaerobic threshold estimated by the fitness- fatigue model and the determined workload at the anaerobic threshold.

[0034] A fitness-fatigue model is a conceptual framework that describes how anathlete’s performance is influenced by two competing factors: fitness and fatigue. A fitness-fatigue model predicts how training, i.e. performing physical activities, affectsendurance performance over time by balancing the positive effects of fitness gains withthe negative effects of accumulated fatigue. A typical fitness-fatigue model predicts aperformance metric based on a relationship defined by parameter values and training load. Training load refers to the amount of stress exerted on a subject during a single physical activity or workout.

[0035] A problem with fitness-fatigue models is that the parameter values are typicallydetermined in a non-individualized manner using population-based heuristics or averages, as there is no continuous performance metric available to be used as a target to fit the parameter values. As such, fitness-fatigue models with an individualized fit are rare. Available individualized fitness-fatigue models are impractical as they require a long period of frequent field or lab testing to obtain target performance metrics. Moreover, this frequent testing typically requires the subject to performdedicated maximal efforts which is undesirable.

[0036] By using the workload at the anaerobic threshold determined based on the fittedlinear equation, a continuous performance metric target can be obtained withoutperforming frequent maximal efforts, long periods of structured training, frequent field or lab testing, or additional measuring protocols outside regular physical activities or training.

[0037] According to an example embodiment, minimizing the difference may beperformed by a differential evolution algorithm.

[0038] This allows determining the parameter values of the fitness-fatigue model in astable and reliable manner.

[0039] According to an example embodiment, the computer-implemented method mayfurther comprise determining a training load of at least one physical activity based on the workload data and / or the heart rate data of the at least one physical activity, wherein the training load is indicative for stress exerted on the subject by the at least one physical activity; and wherein the fitness-fatigue model determines the performance of a subject based on the parameter values and the training load.

[0040] According to an example embodiment, the determined parameter values of thefitness-fatigue model may comprise: ‒a first scaling factor indicative for an increase in performance due to thefitness effect of the physical activity; ‒a second scaling factor indicative for a decrease in performance due tothe fatigue effect of the physical activity;‒ a first time factor indicative for a decay of the impact of training load onthe increase in performance over time; and ‒a second time factor indicative for a decay of the impact of training loadon the decrease in performance over time.

[0041] According to an example embodiment, determining the parameter values of thefitness-fatigue model further comprises constraining the values of the first scaling factor and the second scaling factor between at least 0.01 and at most 3.00, constraining the value of the first time factor between at least 10 days and at most 60 days, and constraining the value of the second time factor between at least 1 day and at most 10 days.

[0042] According to an example embodiment, the second scaling factor may be at leastequal to the first scaling factor; and a difference between the first time factor and the second time factor may be at least 10 days.

[0043] According to an example embodiment, a performance metric estimated by thefitness fatigue model is defined as ^^ + ^^ or as ^^ + ^^ wherein ^^is indicative for the model offset, ^^is the first scaling factor, ^^is the second scaling factor, is the first time factor, ^^is the second time factor, ^^is the training load of physical activity ^, and^ is a unit of time.

[0044] According to a second aspect, the object is achieved by a data processingsystem configured to perform the computer-implemented method according to the first aspect.

[0045] According to a third aspect, the object is achieved by a computer programcomprising instructions which, when the computer program is executed by a computer, cause the computer to perform the computer-implemented method according to the first aspect.

[0046] According to a fourth aspect, the object is achieved by a computer-readablemedium comprising instructions which, when executed by a computer, cause the computer to perform the computer-implemented method according to the first aspect. Brief of the

[0047] Fig. 1 shows an example of a lactate curve indicative for the relationshipbetween the workload exerted by a subject and the blood lactate concentration of the subject during a physical activity;

[0048] Fig. 2 shows steps of a computer-implemented method for determining theperformance of a subject without interfering with endurance training, according to example embodiments;

[0049] Fig. 3 shows further steps of the computer-implemented method, according toexample embodiments;

[0050] Fig.4 shows further steps of the computer-implemented method for determiningthe workload of a subject at the aerobic threshold and / or the anaerobic threshold, according to example embodiments;

[0051] Fig. 5 shows steps of a computer-implemented method for obtaining anindividualized fitness-fatigue model, according to example embodiments; and

[0052] Fig. 6 shows an example embodiment of a suitable computing system forperforming steps according to example aspects of the invention. Detailed

[0053] Fig. 1 shows an example of a lactate curve 117 indicative for the relationshipbetween the workload 112 exerted by a subject and the blood lactate concentration111 of the subject during a physical activity, e.g. cycling, running, swimming, rowing,or cross-country skiing. The lactate curve 117 comprises two points of particularinterest for determining the endurance performance of an athlete: the aerobic lactate threshold 118 and the anaerobic lactate threshold 119. The aerobic lactate threshold118 refers to the minimum workload an endurance athlete has to exert to reach bloodlactate production levels that exceed his / her physiological baseline levels. Theanaerobic lactate threshold 119 refers to the workload at which blood lactateproduction starts to exceed lactate clearance, thereby resulting in rapid lactateaccumulation within the athlete. The anaerobic lactate threshold 118 typically occursat a blood lactate concentration 113 of around 2.0 mmol / L, while the anaerobic lactatethreshold 119 typically occurs at a blood lactate concentration 114 of around 4.0mmol / L. However, this can vary between individuals.

[0054] The lactate thresholds 118, 119 are typically used to construct training plansand structured workouts for athletes. In particular, the workload 115, 116 an athletecan exert at the respective thresholds 118, 119 is typically used to plan structured training as it is a good indicator of the current endurance performance of the athlete. For example, in cycling, a structured training session may have the athlete perform 4sets of 15 minute intervals between 70% – 85% of the workload 116 at the anaerobicthreshold 119 with 5 minute recoveries between the intervals. As such, accurate andup-to-date knowledge of the lactate thresholds 118, 119 is important as it enablesathletes to train optimally by correctly adjusting the intensity of their workouts to theircurrent performance level. However, accurately determining the lactate thresholds118, 119 typically requires performing an exercise test at incremental workloads duringwhich blood samples of the athlete are regularly collected. This then allowsdetermining the workload 112 the athlete can sustain at certain blood lactate levels111, thereby obtaining the lactate curve 117

[0055] These exercise tests have the problem that they are costly, require the athleteto perform a maximal effort, have to be performed in a controlled environment, and aregenerally perceived as unpleasant by most athletes. Generally, it is undesirable toperform frequent maximal efforts, as these result in undue accumulation of fatigue andprevent the athlete to train optimally around the moment when the maximal effort hasto be performed. In other words, the maximal effort of the exercise test to determinethe anaerobic lactate threshold 119 interferes with optimal endurance training. This further makes the blood lactate exercise test unsuitable for continuous performance tracking, as the negative effects of frequently repeating such maximal efforts are even more undesirable. As such, these tests are rarely performed by recreational athletes and even infrequently performed by professional athletes.

[0056] Alternatives include approximating the lactate thresholds with a proxy valuewithout collecting blood samples based on dedicated test protocols, e.g. the ramp-uptest in cycling, or based on data obtained during regular workouts, e.g. by determining a workload-duration curve or by analysing the heart rate variability.

[0057] The approximation methods based on dedicated test protocols have the sameproblem as the blood lactate test in that they still require a maximal effort during a dedicated test, which interferes with optimal endurance training. The approximationmethods based on regular training data have the problem that athletes typically do notspend a lot of time in high-intensity zones, i.e. around the anaerobic threshold 119, during regular workouts and therefore still require incorporating regular maximal efforts in their training plan to enable approximation of the anaerobic threshold 119. Additionally, these approximation methods are typically sensitive to outliers and noise.

[0058] It is a further problem that these approximation methods only allowapproximating the anaerobic threshold in an individualized manner, and that theaerobic threshold is merely estimated in a non-individualized manner using population- based heuristics.

[0059] It can thus be desirable to determine and monitor the performance of a subjectwithout interfering with optimal endurance training, i.e. without performing maximalefforts or high-intensity efforts solely for the purpose of determining the subject’s performance.

[0060] Fig. 2 shows steps 200 of a computer-implemented method for determining theendurance performance of a subject without interfering with endurance training, according to example embodiments.

[0061] In a first step 201, 202, workload data and heart rate data are obtained. Theworkload data and heart rate data are measured during one or more physical activities.The workload data 212 expresses an external effort the subject exerts during the oneor more physical activities. The workload data 212 that is measured may thereforedepend on the performed endurance sport, e.g. power in cycling or rowing, and pace in running or swimming. The workload data may further account for other external effort metrics during an endurance sport, e.g. elevation may be considered in addition to the pace or power. The workload data is obtained by at least one workload sensor. In cycling, power can for example be measured with a crank-based power meter or a pedal-based power meter. In running or swimming, pace can for example be measured with a smart watch.

[0062] The heart rate data 211 expresses an internal effort the subject exerts duringthe one or more physical activities. The heart rate data thus represents the internal effort of the athlete to exert the external effort of the physical activity. In other words,the heart rate data is the cardiovascular response of the subject to the workload 212of the physical activity. It will thus be apparent that the workload data 212 and the heart rate data 211 are time series with corresponding time steps, i.e. the data points in the workload data 212 and heart rate data 211 are aligned in time such that each data point in the workload data 212 corresponds with a specific data point in the heart rate data 211. The heart rate data 211 is obtained by at least one heart rate sensor, e.g. a pulse oximeter or photoplethysmography, PPG, sensor provided within a smart watch; or an electrocardiographic sensor provided within a chest strap. The at least one workload sensor and the at least one heart rate sensor may further be provided within a single device, e.g. a smart watch. It is an advantage that the computer-implemented method is agnostic to the at least one heart rate sensor and the at least one workload sensor used.

[0063] In a next step 203, time intervals 213 – 219 are selected within the obtainedworkload data and heart rate data. These time intervals 213 – 219 are selected basedon a set of selection criteria 204. The set of selection criteria 204 may comprise amaximum variation in the workload data during the time interval 213 – 219, e.g. thestandard deviation of the smoothed workload data may not exceed an absolutethreshold of 50W and a relative threshold of 20%. The set of selection criteria 204 mayfurther comprise, a maximum workload imbalance during the time interval 213 – 219indicative for a maximum difference between the workload exerted by opposing limbsof the subject, e.g. the power imbalance measured at the left and right pedal of a cyclistmust be between 0.3 and 0.7. The set of selection criteria 204 may further comprise amaximum altitude during the time interval 213 – 219, e.g. at most 1000 m above sealevel. The set of selection criteria 204 may further comprise an increase in averageheart rate during the time interval 213 – 219 relative to the average heart rate of apreceding interval. The set of selection criteria 204 may further comprise a minimumenergy expenditure and a maximum energy expenditure, e.g. the total energyconsumed at the start of the time interval must be between 0kJ and 1500kJ. The setof selection criteria 204 may further comprise a minimum average heart rate; amaximum average heart rate; a maximum rate of change in heart rate; and a maximumrate of change in workload. The set of selection criteria 204 may, for example, comprisethat the average heart rate of a final portion of the time segment must be within(^^^^^ − ^1, ^^^^^ + ^2) wherein ^^^^^ is the heart rate of the subject at the aerobicthreshold, ^^^^^ is the heart rate of the subject at the anaerobic threshold, ^1 is apadding value, and ^2 is a padding value. The padding values ^1 and ^2 may, forexample, be 10 and 5 respectively.

[0064] By selecting 203 the time intervals 213 – 219 based on the set of selectioncriteria 204, the amount of heart rate data and workload data can be limited tosegments that are suitable for determining the workload-heart rate relationship. This has the advantage that it reduces the computational load and, thus, improves the efficiency of determining the workload-heart rate relationship.

[0065] The time intervals 213 – 219 may be defined by a start time ^^^^^^^and an end time ^^ ^^^ within respective physical activities ^. The time intervals 213 – 219 may havea predetermined fixed length, e.g. 2 minutes. The selected time intervals may bespread across one or more physical activities. The number of time intervals selected from each of the one or more physical activities may, but need not, be equal. Forexample, five time intervals may be selected from a first physical activity while only onetime interval is selected from a second physical activity.

[0066] In a following step 205, a workload-heart rate pair is obtained for the respectivetime intervals 213 – 219. A workload-heart rate pair (^^ ^^^^ , ^^^ ^ ^^ ) comprises anaverage workload ^^^^^^206 and a corresponding average heart rate ^^^^^^207. The respective average workloads ^^^^^^ 206 are obtained by determining the averageworkload during the entire respective time intervals 213 – 219, i.e. during [^^ ^^^^^ , ^^^^^]. The average heart rates ^^^^^^207 are obtained by determining the average heart rateduring a final portion of the respective time intervals 213 – 219, i.e. during [^^ ^^^^^^^ , ^^^^^] wherein ^^ ^^^^^ < ^^ ^^^^^^^ < ^^ ^^^ . This final portion thus refers to a trailing portion orending portion of the time interval where the heart rate has substantially stabilized, i.e. where the heart rate is substantially free of delayed physiological responses such ascardiac lag or heart rate lag. The final portion may, for example, be the second or lasthalf of the time interval. This allows avoiding that the average heart rate , ^^^^^^is influenced by the initial portion of the time interval when the heart rate of the subject isstill adapting to the workload. In doing so, a set 231 of workload-heart rate pairs isobtained.

[0067] In a next step 208, at least a portion of a workload-heart rate relationship 221 isdetermined by fitting a linear equation 224, 209 to the average workloads and theaverage heart rates of the respective time intervals 213 – 219, i.e. by fitting a linearequation 224 to the workload-heart rate pairs 231. A workload-heart rate relationship221 expresses the relationship between the workload 226 exerted by a subject duringa physical activity and the resulting heart rate 225 of the subject. The workload-heartrate relationship 221 is a good indicator of the endurance performance of the subject,as the workload-heart rate relationship can provide insight into the efficiency,endurance, fatigue resistance, training adaptations, and lactate thresholds 222, 223 ofthe subject. By fitting the linear equation 224 to the workload-heart rate pairs 231, thesubstantially linear portion of a typical workload-heart rate relationship 221 betweenthe aerobic threshold 222 and the anaerobic threshold 223 can be approximated. Inother words, the determined portion of the workload-heart rate relationship 221 maypreferably comprise the portion of the workload-heart rate relationship between theaerobic threshold 222 and the anaerobic threshold 223.

[0068] The computer-implemented method thus allows to determine the enduranceperformance of a subject without interfering with endurance training, as performingdedicated test protocols with maximal efforts and / or high-intensity efforts isunnecessary. This has the further advantage that it makes the computer-implemented method suitable for continuous performance tracking. It is a further advantage that the performance of the subject can be determined and tracked based on regular unstructured physical activities performed in real-world environments, i.e. withoutstructured training sessions or without performing dedicated test protocols in acontrolled lab environment.

[0069] Determining 208 or approximating the portion of the workload-heart raterelationship may initially be performed on a history of workload data and heart ratedata obtained during a historical time window. This is illustrated in Fig. 3, which showsa schematic timeline 310 of physical activities 312 – 320 performed by a subject. Whenthe computer-implemented method described in relation to Fig. 2 is executed a firsttime for a certain subject at timestep the workload data 201 and heart rate data 202of the physical activities 314 – 319 performed during the historical window 311 may beobtained. This historical time window 311 may be defined as a predetermined numberof previous physical activities, e.g. the last 15 workouts. Alternatively, the historical time window 311 may be defined as a predetermined period, e.g. the last 28 days. The historical time window may preferably go back far enough in time as to include enoughdata for fitting the linear equation 224 reliably, without going too far back in time suchthat changes in the subject’s endurance performance remain substantially undetectable within the data.

[0070] Fig. 3 further shows additional steps 301, 302 of the computer-implementedmethod according to further example embodiments. In step 301, the obtained workloaddata and heart rate data may be checked for outliers. Outliers can, for example, be nullvalues or extraordinary values caused by faulty sensors, malfunctioning sensors, orenvironmental factors. Detecting an outlier may be achieved by any outlier detection technique such as, for example, heuristics based on domain knowledge, deep learning models, univariate detection methods, multivariate detection methods, or a combination thereof. If a segment of the workload data and / or the heart rate datacomprises one or more outliers, it may be discarded from the data. In other words, if atime segment of the workload data and / or a time segment of the heart rate data include one or more outliers, that time segment may be discarded from both the workload time series and the heart rate time series as these time series have corresponding timesteps. Outliers may be detected within the heart rate data by itself, e.g. due to a lowbattery of the heart rate sensor; within the workload data by itself, e.g. due to amalfunction of the workload sensor; or within a combination of the heart rate andworkload data, e.g. when the combination of the heart rate and workload data indicatesthat the subject exerted a very high workload at an unrealistic low heart rate. It will beapparent that a discarded time segment due to outliers may have any length and, thus,an entire physical activity may be discarded. For example, physical activity 315 mayshow an excessive number of outliers and may be discarded completely.

[0071] From the remaining or retained workload data and heart rate data, e.g. data ofphysical activities 312, 316, 317, 318, 319 in the example of Fig. 3, time intervals maybe selected in step 203 and workload-heart rate pairs may be determined in step 204, as described in relation to Fig. 2. In a next step 302, one or more preconditions for determining the workload-heart rate relationship may be checked. A precondition may be that a minimum number of intervals are selected in step 203, e.g. at least 10 time intervals. Another precondition may be that the selected intervals must belong to aminimum number of unique physical activities 314, 316, 317, 318, 319, e.g. at least 2distinct physical activities. Another precondition may be that a minimum number of selected intervals have an average heart rate above a certain threshold, e.g. anaverage heart rate within the upper half of the range (^^^^^ − ^1, ^^^^^ + ^2) describein relation to Fig.2. If the preconditions are met, the computer-implemented methodmay proceed to step 208 to determine a portion of the workload-heart rate relationship.Otherwise, the computer-implemented method may be interrupted until sufficient high-quality data is available to determine the endurance performance of the subject more reliably.

[0072] Thus, after completing step 208, at least a portion of the initial workload-heartrate relationship has been determined by fitting an initial linear equation 224 to theworkload-heart rate pairs 331 of previous physical activities 314, 316, 317, 318, 319.The computer-implemented method may further comprise updating 305 this initialworkload-heart rate relationship based on workload data and heart rate data measuredduring a next physical activity 320 or next workout 320. For example, some time afterinitially determining the endurance performance of the subject at timestep thesubject completes a new physical activity 320 at timestep ^^. The initial workload-heartrate relationship may then be updated 305 based on this most recent physical activity320. Updating 305 the workload-heart rate relationship may be achieved by selectingtime intervals within the most recent physical activity 320, determining new workload-heart rate pairs 333 for those selected time intervals, and refitting the linear equation332 to an updated set of workload-heart rate pairs that comprises at least a portion ofthe historical workload-heart rate pairs 331 in addition to the new workload-heart ratepairs 333. In other words, updating 305 the workload-heart rate relationship may beachieved by at least repeating steps 203, 204, and 205. It will be apparent that theupdated set of workload-heart rate pairs may represent a moving window, i.e. theworkload-heart rate pairs of the oldest physical activity 314 may be discarded from theset upon adding the new workload-heart rate pairs 333. This allows continuous tracking of the endurance performance without interfering with endurance training.

[0073] When a next physical activity 320 becomes available, the computer-implemented method may further also detect outliers within the workload data or heart rate data of that next physical activity 320 based on the initial workload-heart rate relationship, i.e. based on the initial linear equation 224. Detected outliers, or time segments comprising the outliers, may be discarded.

[0074] This can be achieved as described in relation to step 301. Alternatively orcomplementary, the heart rate measured during the most recent physical activity 320 may be used to predict the workload exerted during that activity 320 based on the previously determined workload-heart rate relationship, i.e. based on the initial linearequation 224. This predicted workload may then be compared to the measuredworkload during the physical activity. An outlier may then be detected if the difference between the predicted workload and the measured workload, i.e. the prediction error, exceeds a predetermined threshold.

[0075] Alternatively or complementary, the workload measured during the most recentphysical activity 320 may be used to predict the heart rate as a result of exerting said workload during that activity 320 based on the previously determined workload-heartrate relationship, i.e. based on the initial linear equation 224. This predicted heart ratemay then be compared to the measured heart rate during the physical activity. An outlier may be detected if the difference between the predicted heart rate and themeasured heart rate, i.e. the prediction error, exceeds a predetermined threshold. Itwill further be apparent that steps 304 and / or 305 may be repeated 340 for each or at least some of the new physical activities 320 that are recorded for a subject.

[0076] Comparing the predicted heart rate with the measured heart rate of a mostrecent physical activity 320 can further allow detecting overperformance andunderperformance during the most recent physical activity 320, i.e. when the subjectperformed particularly good or bad during the most recent workout 320. This can beachieved by selecting time intervals within the most recent physical activity 320, as instep 203, and determining measured workload-heart rate pairs for the selected timeintervals. Next, determining predicted workload-heart rate pairs for the selected timeintervals by predicting the average heart rates corresponding to the average measuredworkload for the respective time intervals. This can be achieved by inputting themeasured average workloads to the linear equation 224 for the respective timeintervals. Thereafter, prediction errors may be determined as the respectivedifferences between the measured workload-heart rate pairs and the predictedworkload-heart rate pairs for the respective time intervals. The predictions errors mayfurther be averaged to obtain a single average prediction error for the most recentphysical activity 320. A positive average prediction errors may indicate that the subjectwas able to perform better than expected for the same heart rate, i.e. overperformance. A negative average prediction error may indicate that the subject performed worse than expected for the same heart rate, i.e. underperformance.

[0077] Fig. 4 shows further steps 400 of the computer-implemented method fordetermining the workload 412, 414 of the subject at the aerobic threshold 421 and / orthe anaerobic threshold 422 based on the workload-heart rate relationshipapproximation, i.e. linear equation 224 ^^ = ^ + ^ ∗ ^^ determined in step 208. Thiscan be achieved by determining the inverse of the linear equation 224, i.e. ^^ = . This inverse relationship can then be used to determine the workload 412 of the subject at the aerobic threshold 421 in step 401 based on a predetermined heart rate 411 of the subject at the aerobic threshold 421. In other words, the heart rate of the subject at the aerobic threshold should be known.

[0078] This predetermined heart rate 411 may, for example, be obtained by performinga single lactate test in a lab. Thereafter, the predetermined heart rate at the aerobicthreshold 411 may be used to derive the workload 412 at the aerobic threshold 421based on the inverse of the fitted linear equation 224. This allows determining theworkload 412 at the aerobic threshold 421 without repeating a lactate test and withoutperforming dedicated test protocols in controlled circumstances, i.e. based on regularphysical activities. This has the further advantage that the workload at the aerobic threshold can be determined reliably and in an individualized manner.

[0079] Alternatively or complementary, the inverse linear equation can be used todetermine the workload 414 of the subject at the anaerobic threshold 422 in step 402,based on a predetermined heart rate 413 of the subject at the anaerobic threshold 422. In other words, the heart rate of the subject at the anaerobic threshold should beknown. This predetermined heart rate 413 may, for example, be obtained byperforming a single lactate test in a lab. Thereafter, the predetermined heart rate at theanaerobic threshold may be used to derive the workload 414 at the anaerobic thresholdbased on the inverse of the fitted linear equation 224. This allows determining theworkload 414 at the anaerobic threshold 422 without repeating a lactate test andwithout performing dedicated test protocols in controlled circumstances, i.e. based onregular physical activities. This has the advantage that the workload at the anaerobic threshold can be determined reliably without having to perform maximal efforts.

[0080] The computer-implemented method thus allows continuously tracking theendurance performance of a subject by periodically determining the workload at theaerobic and / or anaerobic threshold without interfering with endurance training, asfrequently performing dedicated maximal and / or high-intensity efforts is unnecessary.The determined workload at the anaerobic threshold can further be used as a continuous performance metric target for fitting an individualized fitness-fatigue model.

[0081] A fitness-fatigue model is a conceptual framework that describes how anathlete’s performance is influenced by two competing factors: fitness and fatigue. A fitness-fatigue model predicts how training, i.e. performing physical activities, affectsendurance performance over time by balancing the positive effects of fitness gains withthe negative effects of accumulated fatigue. A typical fitness-fatigue model predicts aperformance metric, e.g. the workload at the anaerobic threshold, based on arelationship defined by parameter values and training load. Training load refers to the amount of stress exerted on a subject during a single physical activity or workout.

[0082] A problem with fitness-fatigue models is that the parameter values are typicallydetermined in a non-individualized manner using population-based heuristics oraverages, as there are no methods to continuously track a performance metric to beused as a target to fit the parameter values without unduly interfering with the trainingof the subject. As such, fitness-fatigue models with an individualized fit are rare.Available individualized fitness-fatigue models are impractical as they require a long period of frequent field or lab testing to obtain a target performance metric. Moreover,this frequent testing typically requires the subject to perform a maximal effort which isundesirable.

[0083] Fig. 5 shows steps 500 of a computer-implemented method for obtaining anindividualized fitness-fatigue model according to example embodiments. To this end,the workload at the anaerobic threshold determined as described in relation to Fig. 4may be obtained in step 500a. In a next step 501, the training load ^^ 513 of at leastone physical activity may be determined based on workload data 510 and heart ratedata 511 measured during the at least one physical activity. Preferably, the trainingload 513 is determined based on internal load techniques that capture thepsychophysiological impact of the workload through subjective scores and heart rate, e.g. Banister’s training impulse, bTRIMP, score. However, the training load 513 may also be determined based on external load techniques that quantify the objective workload, e.g. training stress score, TSS.

[0084] In the following steps 502 the parameter values of a fitness-fatigue model 514may be determined. The fitness-fatigue model may be an equation that predicts the workload of a subject at the anaerobic threshold ^^^^based on an initial baseperformance ^^, the training load ^^ of one or more physical activities ^, a unit of time^, and parameter values 514 – 517. The parameter values 514 – 517 may preferablycomprise a first scaling factor ^^ 514 indicative for an increase in performance due tothe fitness effect of the physical activity; a second scaling factor ^^ 515 indicative for adecrease in performance due to the fatigue effect of the physical activity; a first timefactor 516 indicative for a decay of the impact of training load on the increase in performance over time; and a second time factor ^^517 indicative for a decay of theimpact of training load on the decrease in performance over time. The used fitness-fatigue model may be defined as Alternatively, the used fitness-fatigue model may be defined as

[0085] The parameter values of the fitness-fatigue model may then be determined byminimizing 503 the difference between the workload at the anaerobic thresholdestimated by the fitness-fatigue model ^^^^ and the determined workload 512 at theanaerobic threshold ^^^^^. In other words, the parameter values 514 – 517 may beiteratively adjusted as to minimize the difference between ^^^^ and ^^^^^. In doing so,an individualized fitness-fatigue model can be obtained without performing maximal efforts, long periods of structured training, frequent field or lab testing, or additional measuring protocols outside regular physical activities or training.

[0086] It will be apparent that steps 500 may be performed regularly as to continuouslyupdate the determined workload at the anaerobic threshold ^^^^^ and the parametervalues 514 – 517 of the fitness-fatigue model. This has the advantage thatpersonalized, and properly scaled fitness and fatigue variables can be determined andprovided to athletes or coaches. This can offer useful applications in endurance training such as, for example, tracking the fitness / fatigue ratio as an indicator of overtraining.

[0087] Minimizing the difference in step 503 may further be performed by a differentialevolution algorithm. This allows determining the parameter values 514 – 517 in a stableand reliable manner. Minimizing the difference in step 503 may further be subject toconstraints based on physiological domain knowledge to limit the search space for theparameter values 514 – 517, thereby improving the convergence speed. The valuesof the first scaling factor ^^ 514 and the second scaling factor ^^ 515 may beconstrained between at least 0.01 and at most 3.00, i.e. ^^ and ^^ ∈ [0.01; 3.00]. Thevalues for the first time factor 516 may be constrained between at least 10 days andat most 60 days, i.e. ∈ [10, 60]. The values for the second time factor ^^ 517 maybe constrained between at least 1 day and at most 10 days, i.e. ^^ ∈ [1, 10].

[0088] To further improve the identified parameter values 514 – 517 for theindividualized fitness-fatigue model, minimizing the difference in step 503 may besubject to further constraints that only allow physiologically relevant combinations ofthe parameter values 514 – 517. These constraints may include that the second scalingfactor ^^ 515 may be at least equal to the first scaling factor ^^ 514, i.e. ^^ ≥ ^^; anda difference between the first time factor ^^ 516 and the second time factor ^^ 517 maybe at least 10 days, 10.

[0089] Fig. 6 shows a suitable computing system 600 enabling to implementembodiments of the above described computer-implemented method according to the invention. Computing system 600 may in general be formed as a suitable general- purpose computer and comprise a bus 610, a processor 602, a local memory 604, one or more optional input interfaces 614, one or more optional output interfaces 616, a communication interface 612, a storage element interface 606, and one or more storage elements 608. Bus 610 may comprise one or more conductors that permit communication among the components of the computing system 600. Processor 602 may include any type of conventional processor or microprocessor that interprets and executes programming instructions. Local memory 604 may include a random-access memory (RAM) or another type of dynamic storage device that stores information and instructions for execution by processor 602 and / or a read only memory (ROM) or another type of static storage device that stores static information and instructions for use by processor 602. Input interface 614 may comprise one or more conventional mechanisms that permit an operator or user to input information to the computing device 600, such as a keyboard 620, a mouse 630, a pen, voice recognition and / or biometric mechanisms, a camera, etc. Output interface 716 may comprise one or more conventional mechanisms that output information to the operator or user, such as a display 640, etc. Communication interface 612 may comprise any transceiver-like mechanism such as for example one or more Ethernet interfaces that enables computing system 600 to communicate with other devices and / or systems such as forexample, amongst others, at least one heart rate sensor 651 and at least one workloadsensor 652. The communication interface 612 of computing system 600 may beconnected to such another computing system by means of a local area network (LAN) or a wide area network (WAN) such as for example the internet. Storage element interface 606 may comprise a storage interface such as for example a Serial Advanced Technology Attachment (SATA) interface or a Small Computer System Interface (SCSI) for connecting bus 610 to one or more storage elements 608, such as one or more local disks, for example SATA disk drives, and control the reading and writing of data to and / or from these storage elements 608. Although the storage element(s) 608 above is / are described as a local disk, in general any other suitable computer-readable media such as a removable magnetic disk, optical storage media such as a CD or DVD, -ROM disk, solid state drives, flash memory cards, etc. could be used.

[0090] Although the present invention has been illustrated by reference to specificembodiments, it will be apparent to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied with various changes and modifications without departing from the scope thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. In other words, it is contemplated to cover any and all modifications, variations or equivalents that fall within the scope of the basic underlying principles and whose essential attributes are claimed in this patent application. It will furthermore be understood by the reader of this patent application that the words "comprising" or "comprise" do not exclude other elements or steps, that the words "a" or "an" do not exclude a plurality, and that a single element, such as a computer system, a processor, or another integrated unit may fulfil the functions of several means recited in the claims. Any reference signs in the claims shall not be construed as limiting the respective claims concerned. The terms "first", "second", third", "a", "b", "c", and the like, when used in the description or in the claims are introduced to distinguish between similar elements or steps and are not necessarily describing a sequential or chronological order. Similarly, the terms "top", "bottom", "over", "under", and the like are introduced for descriptive purposes and not necessarily to denote relative positions. It is to be understood that the terms so used are interchangeable under appropriate circumstances and embodiments of the invention are capable of operating according to the present invention in other sequences, or in orientations different from the one(s) described or illustrated above.

Claims

1. CLAIMS1. A computer-implemented method for determining endurance performance of asubject based on sensor measurements; the computer-implemented method comprising: ‒obtaining (201) workload data measured by at least one workload sensor,comprising a time series (212) of the workload exerted by the subject during one or more physical activities; ‒obtaining (202) heart rate data measured by at least one heart rate sensor,comprising a time series (211) of the heart rate of the subject during the one or more physical activities; ‒selecting (203) time intervals (213 - 219) within the workload data and theheart rate data based on a set of selection criteria (204); ‒for the respective time intervals (213 - 219), determining (205) an averageworkload (207) and, for a final portion of the respective time intervals (213 -219), determining (205) an average heart rate (206); and‒ determining (208) at least a portion of a workload-heart rate relationship(221), indicative for the endurance performance of the subject, by fitting alinear equation (224) to the average workloads and the average heart rates (230) of the respective time intervals (213 - 219).

2. The computer-implemented method according to claim 1, wherein the set ofselection criteria (204) comprises at least one of: ^a maximum variation in the workload data;^ a maximum workload imbalance;^ a maximum altitude during the time interval;^ an increase in average heart rate relative to a preceding interval;^ a minimum energy expenditure;^ a maximum energy expenditure;^ a minimum average heart rate;^ a maximum average heart rate;^ a maximum rate of change in heart rate; and^ a maximum rate of change in workload.

3. The computer-implemented method according to any of the preceding claims,wherein the workload data (212) and the heart rate data (211) comprise a history of workload data and heart rate data measured during a predetermined number of previous physical activities or measured during previous physical activities performed within a predetermined preceding period.

4. The computer-implemented method according to claim 3, further comprisingupdating the workload-heart rate relationship (224) based on workload data and heart rate data measured during a next physical activity.

5. The computer-implemented method according to any of the preceding claims,further comprising discarding (301) at least one time segment from the workload data and the heart rate data that includes one or more outliers within the workload data and / or the heart rate data.

6. The computer-implemented method according to claim 5, further comprisingobtaining an estimated workload of a next physical activity based on the heart rate data measured during the next physical activity and the determined workload-heart rate relationship; and detecting one or more outliers within the workload data by comparing the estimated workload with the workload data measured during the next physical activity.

7. The computer-implemented method according to claim 5 or 6, further comprisingobtaining an estimated heart rate of a next physical activity based on the workload data measured during the next physical activity and the determined workload- heart rate relationship; and detecting one or more outliers within the heart rate data by comparing the estimated heart rate with the heart rate data measured during the next physical activity.

8. The computer-implemented method according to any of the preceding claims,wherein the determined portion of the workload-heart rate relationship comprises the workload-heart rate relationship between an aerobic threshold (222) and an anaerobic threshold (223).

9. The computer-implemented method according to claim 8, further comprisingdetermining (401) the workload (412) of the subject at the aerobic threshold (421)based on the determined workload-heart rate relationship (224) and a predetermined heart rate (411) of the subject at the aerobic threshold (421).

10. The computer-implemented method according to claim 8 or 9, further comprisingdetermining (402) the workload (414) of the subject at the anaerobic threshold(422) based on the determined workload-heart rate relationship (224) and a predetermined heart rate (413) of the subject at the anaerobic threshold (422).

11. The computer-implemented method according to claim 10, further comprisingdetermining (602) parameter values (614, 615, 616, 617) of a fitness-fatigue model (614) by minimizing (603) a difference between the workload at the anaerobic threshold (618) estimated by the fitness-fatigue model and the determined workload at the anaerobic threshold (614).

12. The computer-implemented method according to claim 11, further comprisingdetermining (601) a training load (613) of at least one physical activity based on the workload data (610) and / or the heart rate data (611) of the at least one physical activity, wherein the training load is indicative for stress exerted on the subject by the at least one physical activity; and wherein the fitness-fatigue model (614) determines the performance of a subject based on the parameter values (614, 615, 616, 617) and the training load (613).

13. The computer-implemented method according to any of claims 11 - 12, whereinthe determined parameter values of the fitness-fatigue model comprise: ^a first scaling factor (614) indicative for an increase in performance dueto the fitness effect of the physical activity; ^a second scaling factor (615) indicative for a decrease in performancedue to the fatigue effect of the physical activity; ^a first time factor (616) indicative for a decay of the impact of trainingload on the increase in performance over time; and ^a second time factor (617) indicative for a decay of the impact oftraining load on the decrease in performance over time.

14. The computer-implemented method according to claim 13, wherein determining(602) the parameter values (614, 615, 616, 617) of the fitness-fatigue model (614) further comprises constraining the values of the first scaling factor (614) and the second scaling factor (615) between at least 0.01 and at most 3.00, constraining the value of the first time factor (616) between at least 10 days and at most 60 days, and constraining the value of the second time factor (617) between at least 1 day and at most 10 days.

15. The computer-implemented method according to any of claims 13 - 14, whereina performance metric estimated by the fitness fatigue model is defined as ^^+^^∑^^^^^^ ^^(^^^) / ^^, wherein ^^is indicative for a model offset, ^^is the first scaling factor, ^^is the second scaling factor,is the first time factor, ^^is the second time factor, ^^ is the training load of physical activity ^, and ^ is a unit of time.

Images