Classifications

• • A—HUMAN NECESSITIES
  • A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
  • A01K—ANIMAL HUSBANDRY; AVICULTURE; APICULTURE; PISCICULTURE; FISHING; REARING OR BREEDING ANIMALS, NOT OTHERWISE PROVIDED FOR; NEW BREEDS OF ANIMALS
  • A01K29/00—Other apparatus for animal husbandry

Definitions

• TITLE A system and a method for monitoring the physical condition of a herd of livestock
• the invention relates to a system and a method for automatically monitoring the physical condition of a herd of livestock. Automated monitoring of the physical condition of a herd of livestock is helpful for increasing productivity by timely identifying animals which are likely to be in heat or ill, particularly in dairy farming, where increasing herd sizes and automated milking systems make it more difficult and cumbersome for farmers to identify these animals by means of visual observation.
• an experimental method for automatically monitoring the physical condition of a herd of livestock which includes the steps of: measuring a value of a property at regular intervals from each individual, identified animal, storing measurement data in accordance with the measured values of the measured property for each individual, identified animal, determining a prediction for a subsequent measured value of that property for the respective individual, identified animal from the stored measurement data regarding that individual, identified animal, and generating an attention signal in response to an error between the value of the measured property and the prediction for that value above a predetermined level.
• the measured property was the milk yield.
• a time-series model was formulated for predicting the milk yield of each milking or set of three successive milkings with sets of parameters each generally applicable in a particular period of time during a lactation for either heifers or multiparous cows.
• a disadvantage of this described method is, that it is cumbersome in that for each cow the appropriate set of parameters has to be selected. This also forms a potential source of errors. Furthermore, it is unlikely that the determined parameters will also apply to herds of cows of different races or even herds of other animals (e.g. goats), herds kept in other climates or fed with different types of feed.
• this object is achieved by providing a system as described in claim 1 and a method as described in claim 5.
• the method automatically assesses the significance of an error between a prediction and a measured value for each animal individually from data collected during the respective lactation.
• the confidence interval can be determined automatically for each individual measurement and each individual animal, so there is no need to input different selected confidence intervals for different periods of the lactation, for different categories of animals and for different measured properties. Furthermore, the need for separate research to obtain such confidence intervals is obviated.
• the width of the confidence interval is automatically adjusted on-line to the empirically found accuracy of fit of the time-series model and can be signalled separately to indicate the reliability of the predictions.
• the measured property can for example be one of the following properties: milk yield, milk temperature, milk conductivity, animal activity and intake of at least one type of feed.
• the method according to the invention has a prophylactic and accordingly productivity-increasing effect in that it allows an earlier and more reliable identification of individual animals likely to be ill. Firstly, the sooner animals to be checked by a veterinarian can be identified, the better the chances of recovery and the avoidance of adverse effects on the animal are and the better the chances are that further spread of a contagious disease through the herd can be avoided. Secondly, animals having a bad physical condition are accordingly more prone to catching diseases or, if already ill, further diseases. The sooner such animals are identified, the sooner action can be taken to improve the physical condition of such animals and to avoid that the identified animal catches a disease or a further disease.
• the error data are used to characterize the mutual dependence between errors in the predictions of the conductivities of milk obtained from different teats (quarters if the animals are cows). The data regarding this dependence are subsequently used for assessing the significance of errors in the prediction of the conductivity of milk obtained from any one of the teats.
• the error data collected during a lactation are also used to estimate the parameters of the time-series model underlying the predictions of the measured values during that same lactation for each animal individually.
• the time-series model is automatically tailored to an optimal fit to the characteristics of the variations in time of the respective property of that individual animal as the lactation progresses.
• Fig. 1 is a schematic representation of a system according to an embodiment of the invention
• Fig. 2 is a flow chart of a mode of carrying out the method according to the invention.
• the system and the method represented by Figs. 1 and 2 are presently the most preferred modes of carrying out the present invention.
• this method and this system are described in the context of monitoring a herd of cows, but in principle, the invention can also be used for monitoring other animals, provided that at least one property of each individual animal can be measured at regular intervals.
• the system shown in Fig. 1 is integrated with a milking stand 1 for milking individual cows one by one.
• the milking stand further includes a milking device with four suction cups 2-5, to be connected to a cow for withdrawing milk from that cow.
• Milk channels 5-8 are connected at their upstream ends to the suction cups 2-5 and at their downstream ends to conductivity sensors 9-12.
• the conductivity sensors 9-12 are part of a conductivity measurement unit 13. In the conductivity measurement unit 13, the milk channels 5-8 merge downstream from the conductivity sensors 9-12 into a single milk channel 14 passing through a flow meter 15 for measuring the milk yield.
• one of the milk channels 8 passes through a temperature sensor 16 mounted closely adjacent the suction cup 5 to reduce the influence of the ambient temperature on the temperature measurement.
• the milking stand is further provided with a feed dispenser 17 for offering each milked cow a ration of concentrate.
• feed dispensers may be provided in feeding stations located outside the milking area.
• the feed dispenser 17 is adapted to dispense feed as it is consumed by a cow until a predetermined ration has been consumed.
• the feed dispenser 17 further includes a sensor 18 for monitoring the quantity of feed dispensed to each cow. If the full ration has not been consumed when a cow leaves the feed dispenser, the weight of the left over of the respective ration is calculated.
• Each cow of the herd is provided with an activity meter (not shown) which registers a value related to or identical to the cow's activity pattern.
• the milking stand is provided with an activity meter reader 19 for reading a registered value from the activity meter of each cow which is milked, is to be milked or has been milked.
• Each cow of the herd is further provided with an identity tag (not shown).
• the milking stand 1 is provided with a cow identification structure 20 adapted for reading the identity tag of each cow which is milked, is to be milked or has been milked.
• Animal identification systems and systems for monitoring the activity of animals are commercially available and therefore not further described here.
• the conductivity measurement unit 13 the flow meter
• the temperature sensor 16, the feed sensor 18, the activity meter reader 19 and the cow identification structure 20 are each connected to a central data processing structure 21 for processing the measured data regarding each cow.
• the data processing structure 21 is connected to a display 22.
• devices for generating audible alarms can for example be operated via a dedicated connection to the data processing structure, via a network (e.g. via the telephone network) or be remotely controlled.
• the data processing structure is also provided with the necessary peripherals.
• 16, 18, 19, 20 and the data processing structure 21 can also be realised in the form of a wired or wireless bus-structure, in which each station has a distinct address.
• the data processing structure 21 is programmed for storing measurement data in accordance with the measured properties for each individual, identified animal and for determining a prediction for subsequent measured values of these properties for the respective individual, identified animal from the stored measurement data regarding the respective individual, identified animal.
• the data processing structure 21 is programmed for storing error data in accordance with errors between predicted values and measured values for each individual, identified animal, for determining data characterizing the distribution of the prediction errors for each individual, identified animal, for determining a confidence interval for a prediction for each individual, identified animal from the data characterizing the distribution of the errors in the predictions of the measured values, and for activating the signalling device 22 to generate a selected attention signal if an error between the value of a measured property and the prediction for that value is outside the confidence interval.
• the invention is implemented as set forth below with reference to the flow chart shown in Fig. 2.
• the algorithm according to this flow chart is preferably repeated at each milking.
• the identification tag is read to identify the respective cow as is denoted by step 23.
• step 24 values of the milk yield, the milk temperature, the conductivity of the milk obtained from each quarter, the registered value of the activity meter (which may be read with regular intervals upon milkings or at other moments) and the amount of concentrate feed consumed or left over are measured as is denoted by step 24. These measurements are taken at each milking and for each individual, identified animal, i.e. at regular intervals. The measured values are read by the data processing structure 21 as is denoted by step 25.
• step 25 On the basis of earlier measurements and predictions, or at the first milking of a lactation as an initial set of values and parameters, status data which determine the prediction for each next value to be measured and characterize the distribution of errors in previous predictions have been stored in a memory of the data processing structure 21 for each individual, identified cow. In step 25, this status of the respective identified cow is read by the data processing structure 21.
• a prediction for the values measured in step 24 is made as is denoted by step 26. Furthermore, on the basis of the error data - available in the form of the variance and the covariance of earlier predictions and measured values - a variance-covariance matrix of the error is determined for the respective sets of predictions and measured values as is denoted by step 27. ith these error-standardization data, the errors between the current predictions and the measured values can be standardized.
• feed left overs mostly equal zero and are sometimes higher.
• experimentally obtained data suggest that successive left overs are independent and that for each individual animal there is a distinct probability distribution for the percentage of the left over of the concentrates ration, which is preferably defined by:
• This distribution can be used to calculate the probability
• calculations are preferably each time be based on the ombined predicted and measured values of two or more successive milkings.
• step 28 it is decided that an attention signal must be generated.
• attention signals based on a single error are generated if at least one measured value is outside a confidence interval which, after standardization - and assuming the errors for the animal in healthy condition and not in heat have a normal distribution - are outside predetermined confidence intervals.
• a "*" mark can be added to the identification code of a cow if at least one of the measured values is outside a 95% confidence interval
• a "**" mark can be added to the identification code of a cow if at least one of the measured values is outside a 99% confidence interval
• a "***" mark can be added to the identification code of a cow if at least one of the measured values is outside a 99.9% confidence interval.
• the values on the basis of which the attention signal has been generated and the deviation relative to the predicted value are displayed and/or printed as well.
• attention signals based on combinations of errors to indicate the likelihood of heat are generated if activity is rather high and the combination of activity, yield and temperature falls outside a certain confidence interval.
• Attention signals to indicate the likelihood of mastitis are preferably generated if the conductivity error is rather high and the combination of conductivity, yield and temperature falls outside a certain confidence interval.
• An attention signal indicating the likelihood of other illnesses is preferably generated if the combined error of yield, temperature and activity falls outside certain confidence intervals and the concentrate intake is at a level having a low probability under normal circumstances.
• step 29 the display 22 is controlled to display the selected attention signal in association with the identification data of the respective cow.
• step 30 the monitoring of a cow on the basis of the collected data is in principle stopped in response to an attention signal regarding the respective animal, or at least in response to an attention signal above a certain confidence level regarding the respective animal.
• step. 31 If it is decided that no attention signal or no attention signal above a predetermined confidence level is to be generated, the status data for the respective individual cow are updated using two of the following three sets of data: the latest measured values, the latest predictions and the latest errors between the predictions and the corresponding measured values. In the flow chart, this is denoted by step. 31. If, after an attention signal has been generated, verification by the farmer or by a veterinarian reveals that the attention signal was unjustified, the measured values are preferably replaced by the predicted values, so the monitoring of the checked animal can be continued on the basis of the previously collected data and the data entered instead of the latest set of measured values. Thus, step 30 can be overruled in the event of a false positive attention signal.
• impossible measurement results such as a milk temperature of more than 50 °C
• a warning signal indicating that a measured value has been skipped is displayed or printed.
• warning signals indicating the likelihood of malfunction of the measurement structure are obtained as well.
• the milking stand 1 includes a plurality of suction cups 2-5 and a plurality of milk channels 5-8, each connected to one of the suction cups 2-5 and a measurement sensor 9-12 for measuring the conductivity of milk passed through the respective milk channel 5-8 is provided, the conductivity of milk obtained from each quarter can be measured individually. Furthermore, the data structure 21 is programmed for generating an attention signal if the error between the predicted conductivity value and the conductivity value measured by any one of the measurement sensors 2-5 exceeds a threshold value. Thus, an increased conductivity which typically indicates an increased likelihood of mastitis, which typically occurs in one or two of the quarters at a time, can be indicated with an very high sensitivity and specificity.
• the threshold value of the error in the prediction of the conductivity of milk from any quarter is positively related to the average error of the corresponding predictions of all quarters. If the conductivity of milk obtained from all quarters is higher than predicted it is more likely that the deviations are caused by other factors than mastitis, since this disease rarely occurs in all quarters simultaneously. Therefore, a higher sensitivity and specificity can be obtained if the threshold value for any one quarter in response to which an attention signal is generated is higher in response to measured conductivities of milk obtrained from the other quarters which are higher than the predicted conductivity as well.
• the sensitivity and the specificity of the monitoring method can be further increased if it is also taken into account to what extent the conductivities of milk obtained from different quarters are mutually dependent for each individual animal. This is preferably achieved by providing that the dependence between the conductivity values of milk from different quarters is determined for each animal individually from the measured conductivity values of that individual animal, and that, for each individual identified animal, the influence of the average error on the threshold value is positively related to the dependence between the conductivity values of milk from different quarters determined for that individual identified animal.
• the average error in the predicted conductivity values is of less influence on the threshold level for any one conductivity value than for individual cows of which the variations in the conductivity values of milk obtained from different quarters show a closely related behaviour.
• the predictions are preferably made using a time-series model assumed to be valid for healthy cows that are not in heat; unduly great deviations indicate that this assumption is no longer valid, so the monitoring utilizing the model is in principle stopped as described before.
• Appropriate time-series models for the different properties can be established by plotting experimental data, examining the correlograms of the autocorrelations, selecting an appropriate ARIMA and fitting the chosen model.
• the parameters of the time-series model are preferably estimated for each cow individually from errors between values estimated for that individual cow and corresponding measured values as the respective lactation progresses.
• a time-series model is obtained, which automatically adapts itself to the characteristics of the respective cow (for example to more or less erratic variations in the measured properties) and other circumstances which influence the characteristics of the variations of the observations in time.
• no experiments are needed to establish the best parameter settings of the time-series model under different circumstances, for example to take into account the number of days in milk, the race of the cow, the climate, the feed, the milking habits, different categories of cows (heifers or multiparous cows) etc.
• the method is generally easier to manage and differences between individual cows and successive lactations of each cow are also taken into account. It is noted that the on-line estimation of the parameters of the time-series model is also advantageous in that it provides an automatic tailoring of the time-series model to each individual animal if the confidence interval is not individually determined for each individual animal.
• the parameters of the time-series model are preferably estimated using a Kalman filter including a state vector determining the prediction for the next measurement, in which state vector the parameters of the time-series model are included.
• the Kalman filter is a method to estimate the state of a system on-line.
• the state is a quantity that determines the coming behaviour of the system. The estimate is improved after each new observation by using the new information.
• the state comprises (1) the parameters in the time-series models and (2) the probability distribution of the percentage of the calculated concentrates left over.
• the system is described by state-space equations in the form of:
• xt + AtXt- 1 + w t (2) a system equation: xt + AtXt- 1 + w t (2).
• x t is the state vector
• y t the observation vector
• C and A t are system matrices
• v t is the random observation error
• w t is the random system error.
• the observation equation describes the relationship between the measurements and the state, which itself is not directly measurable in general.
• the system equation defines the relation between the state at successive observations.
• the distribution of v t is N(0, V t ) and the distribution of w t is N (0, W t ).
• the estimate of the state x t at observation t is obtained using the measured values obtained at the observations yi to yt-i•
• the Kalman filter provides a new estimate of the state after each set of observations and furthermore a variance-covariance matrix for the state estimate.
• the Kalman filter is a two-stage estimation procedure.
• an estimate of the state and the variance-covariance matrix is calculated on the basis of the previous state.
• this estimation is updated in accordance with the set of observations y t and the estimation error e t (representing the differences between the values obtained during the set of observations and the predictions). The updated estimates are used in connection with the next set of observations.
• the Kalman filter gives the minimum mean square linear estimator of X f Furthermore, the variance-covariance matrix of the estimation error e - on the basis of which the standardization discussed above can be carried out - can also be calculated.
• the state would consist of the measured variables.
• the Kalman filter is used to estimate the parameters of the time-series models of the cow variables, therefore the state includes these parameters.
• the Kalman filter gives a new estimate of the state after each milking, which means new estimates of the parameters of the time-series models. With these new estimates of the parameters, new measurement values are forecasted so that deviant measurements can be signalized reliably and without having to pre-select parameters of the time-series model which are believed to provide the best fit in the respective situation.
• the variance-covariance matrix of the estimated state can furthermore be used to relate the errors between predictions and measured values mutually as described above.
• the Kalman filter calculation method can also be used to fit the probability distribution of the predetermined concentrate consumption levels with the concentrate consumption levels. To achieve this, a description with state-space equations (1) and (2) is used. In this case the following definitions apply:
• the vector x t defines the state (here the probability distribution) and the vector y t is determined by the set of observations with r ⁇ defined as follows:
• Sensitivity for diseases mastitis excluded
• specificity of the detection model based on 263 cases and 40286 milkings outside illness periods.

Landscapes

• Life Sciences & Earth Sciences (AREA)
• Environmental Sciences (AREA)
• Animal Husbandry (AREA)
• Biodiversity & Conservation Biology (AREA)
• Management, Administration, Business Operations System, And Electronic Commerce (AREA)
• Testing Or Calibration Of Command Recording Devices (AREA)
• Investigating Or Analysing Biological Materials (AREA)
• Investigating Or Analyzing Materials By The Use Of Electric Means (AREA)

Priority Applications (6)

Applications Claiming Priority (1)

Publications (1)

Family

ID=8166239

Family Applications (1)

Country Status (6)

Cited By (10)

Families Citing this family (21)

Citations (3)

Family Cites Families (5)

• 1996
  • 1996-06-14 AU AU62245/96A patent/AU6224596A/en not_active Abandoned
  • 1996-06-14 JP JP50109098A patent/JP3856476B2/ja not_active Expired - Fee Related
  • 1996-06-14 US US09/202,380 patent/US6405672B1/en not_active Expired - Fee Related
  • 1996-06-14 WO PCT/EP1996/002604 patent/WO1997047187A1/en not_active Ceased
  • 1996-06-14 EP EP96920821A patent/EP0903980B1/en not_active Revoked
  • 1996-06-14 DE DE69610998T patent/DE69610998T2/de not_active Revoked

Patent Citations (3)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events