EP4152947A1 - Procédé de détermination d'une composition alimentaire pour un troupeau, programme d'ordinateur correspondant, systèmes locaux et fédérés - Google Patents

Procédé de détermination d'une composition alimentaire pour un troupeau, programme d'ordinateur correspondant, systèmes locaux et fédérés

Info

Publication number
EP4152947A1
EP4152947A1 EP21725543.9A EP21725543A EP4152947A1 EP 4152947 A1 EP4152947 A1 EP 4152947A1 EP 21725543 A EP21725543 A EP 21725543A EP 4152947 A1 EP4152947 A1 EP 4152947A1
Authority
EP
European Patent Office
Prior art keywords
milk
group
animal
animals
feed
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
EP21725543.9A
Other languages
German (de)
English (en)
Inventor
José Ahedo
Alejandro Bach
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dairy Information Technology Services SL
Original Assignee
Dairy Information Technology Services SL
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dairy Information Technology Services SL filed Critical Dairy Information Technology Services SL
Publication of EP4152947A1 publication Critical patent/EP4152947A1/fr
Pending legal-status Critical Current

Links

Classifications

    • A—HUMAN NECESSITIES
    • A23—FOODS OR FOODSTUFFS; TREATMENT THEREOF, NOT COVERED BY OTHER CLASSES
    • A23K—FODDER
    • A23K50/00—Feeding-stuffs specially adapted for particular animals
    • A23K50/10—Feeding-stuffs specially adapted for particular animals for ruminants
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/02—Agriculture; Fishing; Forestry; Mining
    • A—HUMAN NECESSITIES
    • A23—FOODS OR FOODSTUFFS; TREATMENT THEREOF, NOT COVERED BY OTHER CLASSES
    • A23C—DAIRY PRODUCTS, e.g. MILK, BUTTER OR CHEESE; MILK OR CHEESE SUBSTITUTES; MAKING THEREOF
    • A23C2230/00—Aspects relating to animal feed or genotype
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/60—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to nutrition control, e.g. diets

Definitions

  • the invention lies in the field of dairy industry and the optimization of the production of milk from animals bred to produce milk.
  • the invention relates to a method for determining a feed composition to be fed in one day to a group of animals that are bred to produce milk, in particular dairy cattle, sheep or goats, said group having one or more animals.
  • the invention also relates to a corresponding computer program for implementing said method and to corresponding local and federated systems.
  • Nutritional models for animals like dairy cattle are known in the art. Said know models calculate nutrient requirements for an animal in order to obtain a specific quantity of milk, using pre-determined algorithms that take into account different parameters of the animal itself, including the energy needed for maintaining the body temperature.
  • a target quantity of milk is determined. This can be done according to the experience of the farmer or by giving the animal plenty of feed and observing how much milk it produces. Afterwards, a target quantity of milk is determined, usually a bit over the normal production. For example, the farmer observes that a particular animal produces 40 litres of milk in one day and determines the target quantity to 42 litres. Using that target quantity, the nutritional model is used for determining the minimum amount of nutrients that have to be fed to the animal in order to obtain said target quantity.
  • the known methods use Linear Programming in order to obtain a feed composition having at least said minimum amount of nutrients with a maximum of quantity of feed, not exceeding what the animal is able to eat, and minimizing the cost.
  • the method has been described for only one animal, nevertheless, the usual application is for a group of animals. In the latter case, the farmer calculates the food required for all of the animals in the group using the target quantity of milk for each animal.
  • the feed being a list of different ingredients like corn silage, alfalfa hay, barley, etc. and their quantities. It is usual that the different animals in a herd are grouped by pens.
  • the solutions that are known in the art calculate the total feed intake required for all the animals of the group, so the farmer can prepare a feed mix containing the ingredients determined by the nutritional model.
  • these nutritional models are a rough estimation of the real situation of the animals in a farm. For example, many of the nutritional models assume that each animal has ad libitum access to feed and water, and that they are kept under dry and clean conditions. Therefore, these models disregard any competition for the food among the different animals in the group. Environmental influence is sometimes disregarded too, even if some nutritional models further incorporate correcting factors considering the ambient temperature, humidity or housing conditions.
  • the methods known in the art have other limitations for the dairy industry.
  • One of the limitations is that they assume that all animals behave the same way, regardless of factors such are genetics and farm management, that can also be related with stress levels.
  • Another important limitation is that they disregard many types of nutrients, as well as the possible local differences in the nutrient content of each ingredient.
  • the models also disregard non-nutritional factors in the feed composition, for example, the palatability, synergistic or antagonist combinations of ingredients and/or nutrients, etc.
  • the nutritional models are aimed for productivity and/or cost minimization, it is not easy or even possible to use a different criterion. For example, calculating a feed composition that maximizes not the cost of the milk production but benefit obtained from said milk considering the cost of the ingredients and the selling price of the milk, said cost of the milk being often dependent on the milk quality and, in particular, dependent on the nutrients present in the milk. Therefore, it is necessary to provide a method for determining feed compositions for animals bred to produce milk that allows different criteria and provide a better adaptation for real-life conditions.
  • the invention is aimed to provide method for determining a feed composition of the type stated at the beginning, being able to avoid the problems that have been identified above.
  • a method for determining a feed composition of the type indicated at the beginning characterized in that said method comprises the steps of: a) determining an optimization criterion, said optimization criterion depending on at least one milk parameter; and b) for each animal of said group of animals, determining an optimized feed composition as the feed composition that optimizes said optimization criterion using a local milk prediction model; c) determining said feed composition to be fed to said group of animals as the aggregation of said optimized feed compositions for each of the animals of said group; wherein said local milk prediction model is configured for obtaining a prediction of milk production of one animal in one day depending on input data; said input data comprising a feed composition fed to said animal in said day; said prediction of milk production comprising a prediction of said at least one milk parameter.
  • the method can be executed periodically, for example daily, thus obtaining daily optimized group feed compositions.
  • it can also be executed for different groups of animals, thus obtaining optimized group feed compositions for different groups of animals, or for an entire herd in a farm or a group of farms.
  • the method of the invention uses a model that is not directed to determine a feed composition for obtaining a predeterminate amount of milk production. Instead, the method uses a predictive model of the expected milk production given a set of input data parameters comprising, at least, a feed composition.
  • the milk prediction model can be interpreted as a function having an n-dimensional input and m-dimensional output.
  • the n-dimensional input corresponds to the different parameters of the input data including, at least, the feed composition fed to the animal
  • the m-dimensional output includes said at least one milk parameter. If only one milk parameter is obtained as output, the function output will be one-dimensional, if two milk parameters are obtained, it will be bi-dimensional, etc.
  • the possible optimization criterion is only meeting a target milk production, and maybe minimizing the cost.
  • different optimization criteria can be determined.
  • Each optimization criterion can be simple, i.e. having just one parameter like the quantity of milk produced by the animal, or complex, i.e. having multiple parameters, for example, including the cost of the ingredients, the expected protein content in the milk, etc.
  • step b) can be performed in different ways according to different optimization methods for one variable or multi-variable optimization.
  • said optimization criterion is defined by a cost function and said step b) comprises a mathematical minimization of said cost function using said predicted milk parameters predicted for said input data.
  • the optimized feed composition is thereby determined for each animal of the group.
  • the feed composition is the aggregation of the optimized feed composition determined for each of the animals in the group.
  • the group contains only one animal, it is considered that the feed composition will be given only to that animal, so the aggregation only contains one optimized feed composition.
  • the method can be applied pen by pen, in the case that all the animals in the pen fed together, or animal by animal, in the case that each animal is fed independently.
  • the skilled person will understand that, in the former case, it is preferred that all the animals in the pen have similar characteristics.
  • the same local milk prediction model is used for all the herd in a farm, irrespective of the groups of animals.
  • a local milk prediction model is aimed for a single group of animals.
  • said feed composition comprises a list of quantities of ingredients, a list of quantities of nutrients or a combination thereof, i.e. a list of one or more ingredients and their quantities, a list of one or more nutrients and their quantities or a combination thereof.
  • the quantity can be, for example a weight, volume or a percentage thereof. Therefore, the resulting list of ingredients and/or nutrients can be used in the farm for creating a feed mix to be fed to the group of animals.
  • the model is adapted for predicting the milk parameters as a function of the ingredients. The latter allows to adapt the method to local variations in the nutrient compositions and availability of each ingredient.
  • determining an optimized feed composition comprises a first optimization step using said local milk prediction model, and a second optimization step using further constraints.
  • This embodiment is particularly advantageous for complex optimization criteria, because it allows to divide the optimization in a part that relates to the milk prediction model, and a part related to other constrains or parameters.
  • the milk prediction model can be used for obtaining a the predicted milk production based on a list of nutrients and their quantities
  • this embodiment allows to, first, obtaining a list of nutrients that maximize the milk production for each animal and, second, obtaining a list of the ingredients that have said required nutrients but that minimize the cost.
  • the first optimization step is performed using the local milk prediction model
  • the second optimization step is performed using further constraints that relate to the available ingredients, their nutritional contents and cost.
  • said at least one milk parameter comprises at least one of the following milk parameters: quantity of milk, thus allowing optimization criteria like maximizing the production of milk; protein content of the milk; fat content of the milk; lactose content of the milk; calcium content of milk; and fatty acid profile of milk; or combinations thereof.
  • the parameters relating to milk contents are particularly advantageous for optimization criteria that include quality-related information of the milk.
  • said input data of said local milk prediction model further comprises at least one animal parameter, preferably, at least one of the list consisting in: body weight; stage of lactation; stage of pregnancy; lactation number; and genomic value of the animal.
  • animal parameter preferably, at least one of the list consisting in: body weight; stage of lactation; stage of pregnancy; lactation number; and genomic value of the animal.
  • genomic value refers to the DNA sequence of the animal.
  • the model is able to make a prediction of the milk, not only based in feed intake but also in other parameters that have been found relevant for the production of milk.
  • said input data of said local milk prediction model further comprises at least one environmental parameter, preferably at least one of temperature, humidity and hours of exposure to daylight, thus accounting the influence of the environment in the production of milk.
  • said optimization criterion is one of: maximizing milk production; thus being able to determine feed compositions that lead to a maximum of milk production by the animals, preferably further minimizing the production cost; obtaining a predetermined milk protein content value or range; thus being able to produce milk with a quality criterion based on the protein content; obtaining a predetermined milk fat content value or range; thus being able to produce milk with a quality criterion based on the fat content; and maximizing milk production benefit.
  • the optimization criterion includes not only the milk production but also the milk selling price and the price of the ingredients to be used in the feed composition.
  • the price of the milk often depends on several parameters, for example, the quantity of milk sold, and quality parameters of the milk. Therefore, it is also possible that the optimization criterion is a combination of the criteria defined above, which is particularly advantageous if the optimization criterion comprises not only quantity or benefit, but also quality parameters of the milk.
  • the model used is not itself aimed for determining the feed composition, instead, the milk prediction model is aimed for predicting milk parameters given a set of input data, and the feed composition is determined using an optimization according to the inputs and outputs of the model.
  • the method further comprises the steps of: d) for each animal of the group, receiving a set of measures, comprising: a measure of consumed feed, said consumed feed being the feed composition that has been consumed by said animal in one day; a measure of milk production, said milk production being the milk that has been produced by said animal in said day, said measure of milk production comprising said at least one milk parameter; preferably, at least a measure of an animal parameter; and preferably, at least a measure of an environmental parameter; e) updating said local milk prediction model by incorporating said set of measures for each animal of the group.
  • a set of measures comprising: a measure of consumed feed, said consumed feed being the feed composition that has been consumed by said animal in one day; a measure of milk production, said milk production being the milk that has been produced by said animal in said day, said measure of milk production comprising said at least one milk parameter; preferably, at least a measure of an animal parameter; and preferably, at least a measure of an environmental parameter; e) updating said local milk prediction model by incorporating said
  • the model is updated based on a set of measures corresponding to input data of the model (feed intake and preferably other parameters relating to the animal and/or the environment) and to the output data of the model (the parameter or parameters related to the milk production).
  • the animal and environmental parameters have been disclosed above. Even if the predictions are accurate, the real measures can differ, therefore, updating the model according to the measures tends to improve its results. Moreover, this updating process results in a model that can, potentially, take into account local variations or complex dependencies in the milk production, for example, those relating to the nutrient profiles of the ingredients, the metabolic profiles of the animals, the environmental effects, management effects such as stocking density, etc.
  • the steps d) and e) are repeated daily, so that the milk prediction model is updated with according to the interval of time of reference, i.e. the model predicts the milk production in one day according to the input data in that day.
  • the method further comprises an initial learning phase, wherein only steps d) and e) are repeated daily, and a production phase comprising steps a) to e).
  • said measures of consumed feed and said measures of milk production are, independently, obtained using respective moving averages, in order to filter transitory results or brief variations.
  • Moving averages also referred as rolling averages, are thus used in order to obtain representative data having less variance.
  • the window for the average is determined according to the variance or the standard deviation of the sample, so that, if the variance is high, a long average window is selected, preferably, up to ten days, while, if the variance is low, a short average window is selected, the minimum being one day.
  • the length of the average window further depends on the size of the group, so that, for groups having a large number of animals, the average windows is shorter than for groups having a small number of animals.
  • said measure of consumed feed is an estimated measure based in a measure of feed consumed by all the animals in the group.
  • This is particularly advantageous in the case that all the animals of the group feed from the same feeding point, when it is not possible to measure the feed intake of each animal independently.
  • Different estimations can be envisaged, by the way of a non-limiting example, simply dividing the feed consumed by all the animals by the number of animals in the group.
  • said estimated measures are obtained by: obtaining an expected consumption for each animal of the group; obtaining a total expected consumption for the group, being the aggregation of each expected consumption for each of the animals of the group; dividing said measure of feed consumed by all the animals in the group by said total expected consumption for the group, thus obtaining a correction factor; and for each animal in the group applying said correction factor to said expected consumption for said animal.
  • said estimated measure is obtained by applying a consumption factor of said animal to said measure of feed consumed by all the animals in the group; wherein said consumption factors are determined by: obtaining an expected consumption for each animal of the group; obtaining a total expected consumption for the group, being the aggregation of each expected consumption for each of the animals of the group; and for each animal in the group determining said consumption factor as the ratio between said expected consumption for said animal and said total expected consumption.
  • the factor for each animal is obtained in a previous step as the ratio between the feed expected to be consumed by each animal and the total feed expected to be consumed.
  • an expected consumption is obtained, for example, using a model of feed intake, using previous observations of feed intake by said animal, etc.
  • the total expected consumption corresponds to the addition of the expected consumptions of all the animals. Therefore, the factor for each animal is related to the portion of the total expected feed consumption that it is expected to be consumed by said animal. According to this definition, the sum of all the factors for all the animals in the group is 1. Nevertheless, other equivalent relationships can be envisaged, for example, using percentages instead of factors or other weighted relations. Even if this embodiment is still an approximation, it leads to more accurate results, in special if the expected consumptions are close to the reality.
  • said expected consumption for each animal is obtained by using the optimized feed composition for said animal, determined in step b), thus using the milk prediction model in order to estimate the consumption.
  • This option is particularly advantageous when the model is well trained.
  • said expected consumption for each animal is obtained by using a prediction model configured to estimate the feed intake of one animal based on its milk production and/or animal characteristics. Since the prediction model for the estimation is not the same than the local milk prediction model, this embodiment is less likely to show positive feedbacks in the adaptation, and is, therefore, particularly suitable for the initial updating iterations when the milk prediction model is not fully trained.
  • a prediction model configured to estimate the feed intake of one animal based on its milk production and/or animal characteristics.
  • said local milk prediction model is a neural network, thus being able to adapt to the updates based on measures, with a limited a priori considerations.
  • said neural network is a Long-Short Term Memory, LSTM neural network.
  • LSTM neural networks each node has a feedback, therefore they show a memory effect that keeps the variations and improvements during some time. This effect makes the particularly well suited for the type of usage in the milk prediction model.
  • the method further comprises: sending to a central module: said set of measures; and the internal status of the neural network nodes of said local milk prediction model; receiving from said central module an updated milk prediction model; and replacing the current local milk prediction model with said updated milk prediction model.
  • Said central module is preferably one or more servers that can be accessed remotely, for example, using internet as the communication platform.
  • said updated milk prediction model is generated using federated learning which is a known strategy of distributed learning used in the field of the neural networks.
  • federated learning is a known strategy of distributed learning used in the field of the neural networks.
  • this kind of learning strategy is particularly advantageous in order to take advantage of the information coming from multiple sources (i.e. different local milk prediction models), but at the same time, maintain a local adaptation that can take into consideration local particularities for each farm.
  • the invention also refers to a computer program containing program instruction codes that, when executed by a computer, lead to said computer to perform the method described above.
  • the invention also refers to a local system for determining feed compositions to be fed to at least one group of animals that are bred to produce milk, in particular dairy cattle, sheep or goats, each group of animals having one or more animals, characterized in that said system comprises a control module comprising a local milk prediction model and is configured for performing the method according to any of the embodiments disclosed above for each group of animals, thus having the same technical effects and advantages that have already been explained.
  • said local system further comprises a mobile device to assist in the creation of a feed mix that will be given to one group of animals of said at least one group of animals, said mobile device comprising: display means, for displaying information to a user; communication means, for communication with said control module; and weight measure receiving means, for receiving measures of weight of ingredients; wherein said mobile device is configured for: receiving from said control module a feed composition for one group of animals of said at least one group of animals, said feed composition comprising a list of ingredients and the weight of each of said ingredients that has be added to a feed mix that will be given to said group of animals; for each of said ingredients of said list of ingredients: displaying said weight of said ingredient using said display means; and receiving a measure of the weight of said ingredient that has been added to said feed mix; and, sending to said control module said measures of the weight of each ingredient; wherein said control module is configured for determining a measure of feed consumed by all the animals in said group according to said measures of the weight of each ingredient.
  • the animals are fed using a feed mix containing different ingredients, for example, corn silage, alfalfa hay, barley, etc. said ingredients are taken from their respective storage locations, mixed, and then offered to the animals.
  • the feed mix is often transported using trucks or mixing wagons.
  • the mobile device disclosed herein assist in the creation of the feed mix using the feed composition that has been determined by the method by displaying the quantities of each ingredient.
  • the device further comprises location means, for determining the location of the mobile device regarding the place of storage of each ingredient.
  • location means for determining the location of the mobile device regarding the place of storage of each ingredient.
  • the latter allows displaying of the quantities of each ingredient when the mobile device is located where said ingredient is.
  • Different options can be envisaged for said location means, by the way of non-limiting examples, using GPS, NFC or beacons.
  • the mobile device is also used for receiving the measures of the actual quantity of each ingredient which are often slightly different from the quantity that has been determined for the feed composition. Using these measures, the control module can determine the real amount of feed offered to the animals that then used for determining the measure of consumed feed, the latter being used for updating the local milk prediction model.
  • said control module is further configured for obtaining a measure of weight of the remaining feed that has not been consumed by all the animals in said group, and wherein determining a measure of feed consumed by all the animals further includes said weight of the remaining feed, thereby adjusting the measure in the case that the animals have not consumed all the feed mix that has been provided to them, thus improving the accuracy in the measures and, consequently, of the updated model.
  • the invention also refers to a federated system for determining feed compositions to be fed to at least one group of animals that are bred to produce milk, in particular dairy cattle, sheep or goats, each group of animals having one or more animals, characterized in that said system comprises: at least one local system as disclosed above; and a central module, comprising a seed milk prediction model; wherein said seed milk prediction model and the local milk prediction models of said at least one local system are all neural networks, preferably LSTM neural networks, having the same node structure; each neural network having an internal status; said central module configured for receiving from each control module corresponding to each of said at least one local system: a set of measures; and said internal status of the neural network of said local milk prediction model; said set of measures comprising, for each animal: a measure of consumed feed, said consumed feed being the feed composition that has been consumed by said animal in one day; a measure of milk production, said milk production being the milk that has been produced by said animal in said day, said measure of milk production comprising said at least one milk parameter;
  • the federated system disclosed herein is thus able to improve the seed model using the data coming from different local systems.
  • the seed model is therefore able to accumulate correlations between the input data and the milk parameters relying on potentially many sources of information.
  • each local system can still maintain adaptations related to the local conditions, thanks to the federated learning, thus obtaining more accurate local results.
  • Fig. 1 is a simplified flowchart of the steps for determining a feed composition according to one embodiment of the method of the invention.
  • Fig. 2 is a simplified flowchart of the inputs and outputs of the local milk prediction model.
  • Fig. 3 is a representation of the update of the local milk prediction model using a set of measures for different animals.
  • Figs. 4A, 4B and 4C show a comparison between the real daily milk production (black line), the prediction according to current state of the art (dashed line), and the prediction using the local milk production model according to one embodiment of the invention (grey line).
  • Each graph corresponds to a different animal of a group of animals.
  • the horizontal axis is the day, and the vertical the yield of milk in kg per day.
  • Fig. 5 is a simplified diagram of an embodiment of the federated system according to the invention. Detailed description of embodiments o the invention
  • the figures 1 to 4 show one exemplary embodiment of the method for determining an optimized group feed composition 1 to be fed in one day to a group 2 of animals that are bred to produce milk, in particular dairy cattle, sheep or goats, said group 2 having one or more animals.
  • Figure 5 shows a corresponding federated system according to the same embodiment.
  • the animals are dairy cattle and the group 2 of animals is a pen 2 containing initially 23 animals.
  • the skilled person will understand that these numbers are non-limiting and used just for clarification purposes.
  • Fig. 1 shows that the method comprises the steps described hereinafter: a) Determining an optimization criterion 3, said optimization criterion 3 depending on at least one milk parameter.
  • the optimization criterion 3 used for the examples is milk production maximization at minimum cost for the ingredients.
  • the milk parameter is milk production in kilograms.
  • the milk production in one day is often referred as yield milk in the art.
  • the group 2 is a pen containing initially 23 dairy cows.
  • each optimized animal feed composition 4 is a list of ingredients and their quantities that have to be given to the corresponding animal in order to optimize the optimization criterion 3.
  • the optimized group feed composition 1 is a list containing all the ingredients of each of the optimized animal feed compositions 4 and the sum of their quantities for all the animals in the pen 2.
  • the method of the example is repeated daily, thus obtaining a daily optimized feed composition for the group of animals 2.
  • the local milk prediction model 5 is configured for obtaining a prediction of milk production 52 of one animal in one day depending on input data 51 as shown in Fig. 2.
  • the local milk prediction model 5 is a Long-Short Term Memory, LSTM, neural network, and the input data 51 comprises: A feed composition fed to said animal in said day, in this embodiment, said feed composition is the amount of dry feed provided to the cow in kg, the crude protein in kg, the neutral detergent fibre in kg, and the energy of the feed in Meal.
  • the following animal parameters body weight in kg, stage of lactation in days, and lactation number, the latter being the number of times that the cow has been in lactation.
  • the prediction of milk production 52 comprises milk output data comprising said at least one milk parameter, in this embodiment, the kilograms of milk production for said day.
  • Other units can be envisaged for the parameters described above, by the way of an example, using imperial units and/or giving the crude protein content and detergent fibre as percentages of the amount of dry feed.
  • determining the optimized animal feed composition 4 comprises a first optimization step using said local milk prediction model 5, and a second optimization step using further constraints.
  • the local milk production model 5 is tested with the following input data 51 :
  • the cow body weight (694kg), stage of lactation (51 days) and lactation number (3) in that day.
  • a plurality of seven different feed compositions each feed composition having different values of the amount of dry feed, crude protein, neutral detergent fibre, and the energy of the feed.
  • a linear programming is performed using the feed composition obtained in the first optimization step as constraints together with a list of available ingredients containing their nutritional compositions and prices. After this step, the result is the optimal combination of ingredients that satisfy the supply nutritional requirements for the cow at the least possible cost:
  • Fig. 3 shows further steps of the first embodiment of the method, aimed to the update of the local milk production model 5: d) For each animal of the group 2, receiving a set of measures 7. e) Updating the local milk prediction model 5 by incorporating said set of measures 7 for each animal of the group 2.
  • Steps e) and d) are repeated daily, and the measures are therefore corresponding to each day. Nevertheless, instead of single daily data, moving averages are used, using a rolling window of 30 days.
  • said set of measures 7 comprises measured data corresponding to the input data 51 and prediction of milk production 52 used by the local milk prediction model 5, in particular: A measure of consumed feed, said consumed feed being the feed composition that has been consumed by said animal in one day, i.e. the amounts of dry feed, crude protein and neutral detergent fibre, as well as the energy of the feed.
  • the following environmental parameters minimum temperature, maximum temperature and average temperature.
  • a measure of milk production said milk production being the milk that has been produced by said animal in said day, said measure of milk production comprising said at least one milk parameter, i.e. the amount of milk production in kg.
  • the measure of consumed feed for each animal of the group 2 is not a direct measure.
  • said estimated measures are obtained following the steps described herein.
  • an expected consumption is obtained for each animal of the group 2.
  • the expected consumption for each animal is obtained by using a prediction model configured to predict the feed intake of one animal based on its milk production and/or animal characteristics.
  • Different prediction models can be envisaged, for example the NRC 2001 , but is out of the scope of this document to describe in detail the characteristics of said models.
  • a total expected consumption for the group is obtained, being the aggregation of each expected consumption for each of the animals of the group 2.
  • the expected consumptions are then used for weighting the consumption of each animal from the measured consumption of all the pen.
  • a correction factor is obtained by dividing the measure of feed consumed by all the animals in the group 2 by the total expected consumption for the group 2. Finally, for each animal in the group 2, said correction factor is applied to the expected consumption for said animal.
  • the head count column corresponds to the number of animals in the group. Note that this number can be different from day to day, in the example, in the first and second days the group 2 has 23 animals, and in the third day, the group 2 has 24 animals.
  • the second column is the real measure of feed consumed by all the animals in the group 2.
  • the third column is the total expected consumption for the group 2 which is obtained as the aggregation of each expected consumption for each of the animals of the group 2.
  • the expected consumption in the exemplary table is calculated by a customized model, but other known models can be envisaged, for example the well-known NRC 2001.
  • the last column is the ratio between the values in the second and third columns, thus being a correction factor.
  • the three first columns of Table II correspond to the equivalent columns in Table I.
  • the expected consumption has been calculated using the customized model mentioned above.
  • the last column is the measure of consumed feed obtained for said animal in step d) and used for updating the local milk production model 5 in step e). It corresponds to the multiplication of expected consumption by the correction factor.
  • Fig. 4A, 4B and 4C show exemplary graphs comparing the milk production in kg for different consecutive days. Each figure represents a different animal in the pen.
  • the black line corresponds to real measures of milk production and the grey line the prediction of milk 52 for the animal using the local milk prediction model 5 of the invention.
  • the dotted line has been calculated using the common known model described by the Wool equation describing an algebraic model of the lactation curve in cattle.
  • Fig. 5 shows a diagram of the complete system corresponding to the first embodiment.
  • the figure depicts a federated system 200 having three local systems 100 and a central module 201.
  • Each of the local systems 100 has a control module 101 comprising a local milk prediction model 5 and is configured for performing the method according to the embodiment described above.
  • the central module 201 has a seed milk prediction model 6 that is a LSTM neural network. All the neural networks 5 and 6 have the same structure.
  • Each control module 101 is a computer running a software application having software instruction commands that, where are run by the computer execute the method described above.
  • Other kind of hardware can also be envisaged, by the way of non limiting examples, a plurality of computers, a portable device like a mobile phone or tablet, or a dedicated machine having computing capabilities.
  • the communication between the central module 201 and each of the control modules 101 is done using World Wide Web communications, but other channels can be envisaged, for example, direct radio communications.
  • the method that is performed by the control modules 101 of local systems 100 further comprises the daily steps of: sending to the central module 201 : the set of measures 7; and the internal status of the neural network nodes of the local milk prediction model 5; receiving from said central module 201 an updated milk prediction model that has been generated by the central module 201 using federated learning; and replacing the current local milk prediction model 5 with said updated milk prediction model.
  • each of the local systems 100 of the federated system 200 that is represented in Fig. 5 further comprises a mobile device 102 to assist in the creation of a feed mix that will be given to the animals.
  • the different ingredients to be used in the feed mix are stored in their respective storage locations.
  • a truck moves from one storage location to another, loading different ingredients that are mixed together and given to the animals.
  • the mobile device 102 moves together with the truck in order to assist in the creation of the feed mix, and is provided with a tracking system and a map having the storage locations of the ingredients, so when, the truck arrives to a storage location, the mobile device 102 it is able to determine the ingredient corresponding to said location.
  • Other location means can be envisaged, for example, using RFID, NFC or beacons together with corresponding elements in the storage locations of the ingredients.
  • the mobile device 102 comprises display means, in particular a display screen, for displaying information to a user. It is also provided with communication means, in particular a wireless communication module, capable of using cell-phone data networks and/or WIFI networks.
  • the mobile device 102 also has a weigh measure interface, for receiving measures of weights of ingredients.
  • control module 101 sends to the mobile device 102 the optimized group feed composition 1 , containing the list of ingredients and the weight of each ingredient.
  • the mobile device 102 receives said composition and, when the truck arrives to the storage location of each ingredient, it displays in the screen the quantity of the said ingredient that has to be used in the feed mix. The operator loads the quantity and the weight measurement interface receives the actual weight that has been loaded in the truck for said ingredient. The truck then moves to the next storage location, and the sequence is repeated until the feed mix has been completed. Using the communication means, the mobile device 102 sends the measures of the weight of each ingredient to the control module 101. At the end of the day, the control module 101 further receives a measure of the weight of the remaining feed that has not been consumed by the group 2 of animals.
  • control module 101 determines de feed consumed by all the animals in the group 2, which is latter used for determining the measure of the feed consumed by each animal according to the method described above.
  • the milk parameters obtained by the local milk prediction model comprise not only the quantity of milk but also at least one of the following: quantity of milk; protein content of the milk; fat content of the milk; lactose content of the milk; calcium content of milk; and fatty acid profile of milk; or combinations thereof.
  • the optimization criterion 3 depends on the milk parameters available, therefore, the skilled person will correspondingly select the milk parameters to be used.
  • the optimization criterion 3 is obtaining a predetermined milk protein content value or range.
  • Other possible optimization criteria 3 can be envisaged, for example, obtaining a predetermined milk fat content value or range, maximizing milk production benefit, or combinations thereof.
  • the input data 51 of the local milk prediction model 5 further comprises at the humidity.
  • the input data comprises the hours of exposition to daylight.
  • the expected consumption for each animal is obtained by using the optimized animal feed composition 4 for said animal, determined in step b).

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Polymers & Plastics (AREA)
  • Marketing (AREA)
  • Animal Husbandry (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Quality & Reliability (AREA)
  • Agronomy & Crop Science (AREA)
  • Operations Research (AREA)
  • Marine Sciences & Fisheries (AREA)
  • Mining & Mineral Resources (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Birds (AREA)
  • Zoology (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Food Science & Technology (AREA)
  • Feeding And Watering For Cattle Raising And Animal Husbandry (AREA)

Abstract

L'invention concerne un procédé de détermination d'une composition alimentaire pour un troupeau, un programme d'ordinateur correspondant, des systèmes locaux et fédérés. Une composition alimentaire quotidienne pour un troupeau (1) destinée à être donnée à un troupeau (2) d'animaux qui sont élevés pour produire du lait est déterminée par : a) détermination d'un critère d'optimisation (3) en fonction d'au moins un paramètre de lait; b) pour chaque animal, la détermination d'une composition alimentaire pour animaux optimisée (4) comme étant celle qui optimise ledit critère d'optimisation (3) à l'aide d'un modèle de prédiction de lait local (5); et c) la détermination de ladite composition alimentaire pour troupeau optimisée (1) en tant qu'agrégation desdites compositions alimentaires pour animaux optimisées (4); ledit modèle de prédiction de lait local (5) étant conçu pour obtenir une prédiction de la production de lait (52) d'un animal en une journée en fonction de données d'entrée (51); lesdites données d'entrée (51) comprenant une composition alimentaire donnée audit animal au cours de ladite journée; ladite prédiction de production de lait (52) comprenant ledit paramètre de lait.
EP21725543.9A 2020-05-18 2021-05-18 Procédé de détermination d'une composition alimentaire pour un troupeau, programme d'ordinateur correspondant, systèmes locaux et fédérés Pending EP4152947A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
EP20382418.0A EP3912477A1 (fr) 2020-05-18 2020-05-18 Procédé de détermination d'une composition d'alimentation de groupe optimisée, programme informatique correspondant, systèmes locaux et fédérés
PCT/EP2021/063059 WO2021233865A1 (fr) 2020-05-18 2021-05-18 Procédé de détermination d'une composition alimentaire pour un troupeau, programme d'ordinateur correspondant, systèmes locaux et fédérés

Publications (1)

Publication Number Publication Date
EP4152947A1 true EP4152947A1 (fr) 2023-03-29

Family

ID=70779669

Family Applications (2)

Application Number Title Priority Date Filing Date
EP20382418.0A Withdrawn EP3912477A1 (fr) 2020-05-18 2020-05-18 Procédé de détermination d'une composition d'alimentation de groupe optimisée, programme informatique correspondant, systèmes locaux et fédérés
EP21725543.9A Pending EP4152947A1 (fr) 2020-05-18 2021-05-18 Procédé de détermination d'une composition alimentaire pour un troupeau, programme d'ordinateur correspondant, systèmes locaux et fédérés

Family Applications Before (1)

Application Number Title Priority Date Filing Date
EP20382418.0A Withdrawn EP3912477A1 (fr) 2020-05-18 2020-05-18 Procédé de détermination d'une composition d'alimentation de groupe optimisée, programme informatique correspondant, systèmes locaux et fédérés

Country Status (2)

Country Link
EP (2) EP3912477A1 (fr)
WO (1) WO2021233865A1 (fr)

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060036419A1 (en) * 2004-07-29 2006-02-16 Can Technologies, Inc. System and method for animal production optimization
US7886691B2 (en) * 2005-09-13 2011-02-15 Wisconsin Alumni Research Foundation Method for optimizing health and productivity of milk producing animals
NL1033926C2 (nl) * 2007-06-03 2008-12-08 Maasland Nv Werkwijze en inrichting voor het beheren van een groep melkdieren, alsmede een computerprogrammaproduct daarvan.

Also Published As

Publication number Publication date
EP3912477A1 (fr) 2021-11-24
WO2021233865A1 (fr) 2021-11-25

Similar Documents

Publication Publication Date Title
US10354342B2 (en) Adaptive livestock growth modeling using machine learning approaches to predict growth and recommend livestock management operations and activities
Lansink et al. Investigating technical efficiency and potential technological change in Dutch pig farming
RU2464780C2 (ru) Способ, устройство и компьютерный программный продукт для управления группой молочного скота
Sauer et al. Sequential technology implementation, network externalities, and risk: the case of automatic milking systems
WO2018102592A1 (fr) Capture, traitement et adaptabilité asynchrones d'informations de poids de ration de bétail de dispositif d'alimentation en temps réel et transfert sur connexion sans fil pour dispositif mobile, commande de chaîne d'alimentation de machine à machine, et traitement d'application
Alqaisi et al. Optimal dairy feed input selection under alternative feeds availability and relative prices
CN108229739B (zh) 作物产量的预测方法、终端、计算机可读存储介质
Menendez et al. The characterization of the cow-calf, stocker and feedlot cattle industry water footprint to assess the impact of livestock water use sustainability
Davoudkhani et al. Economic optimization of feeding and shipping strategies in pig-fattening using an individual-based model
WO2017011755A1 (fr) Technologies génomiques pour la gestion de la production et des performances en agriculture
CN115409153A (zh) 一种基于注意力lstm的畜牧指标预测方法及预测系统
EP3912477A1 (fr) Procédé de détermination d'une composition d'alimentation de groupe optimisée, programme informatique correspondant, systèmes locaux et fédérés
Kulak et al. Comparisons of economic values with and without risk for livestock trait improvement
Lewis et al. Using a two-price market value method to value extra pasture DM in different seasons
Pomar et al. Fundamentals, limitations and pitfalls on the development and application of precision nutrition techniques for precision livestock farming
Žgajnar et al. Combination of linear and weighted goal programming with penalty function in optimisation of a daily dairy cow ration
Patil et al. Dairy cattle nutrition and feed calculator—an android application
Niemi et al. The economic and environmental value of genetic improvements in fattening pigs: an integrated dynamic model approach
Lehtonen Sector-level economic modeling as a tool in evaluating greenhouse gas mitigation options
Hanigan et al. Metabolic models of ruminant metabolism: recent improvements and current status
Woli et al. Modifying the National Research Council weight gain model to estimate daily gain for stockers grazing bermudagrass in the southern United States
Eckard et al. The challenges and opportunities when integrating animal models into grazing system models for evaluating productivity and environmental impact
Parker Standardisation between livestock classes: the use and misuse of the stock unit system
Whittemore et al. Nutrition management of growing pigs
AU2021105597A4 (en) A multivariate prediction method for nutrient requirement & cattle feed formulation

Legal Events

Date Code Title Description
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: UNKNOWN

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE

PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20221115

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

DAV Request for validation of the european patent (deleted)
DAX Request for extension of the european patent (deleted)