A method for monitoring dairy cow feeding behavior

Abstract

This invention discloses a method for monitoring dairy cow feeding behavior, belonging to the field of dairy farming technology. Based on historical factors influencing feeding and historical feed intake, a dairy cow feed intake prediction model is trained, enabling the model to predict feed intake and providing a reference for feed distribution. After feed is distributed, feed intake data can be collected to monitor feeding behavior, effectively improving feed intake and allowing staff to manage dairy cows based on feeding behavior, thus increasing management efficiency.

Description

Technical Field

[0001] This invention belongs to the field of dairy farming technology, specifically relating to a method for monitoring dairy cow feeding behavior. Background Technology

[0002] Feed intake in dairy cows refers to the amount of forage (feed) actually consumed by a dairy cow per unit of time. Feed intake directly affects the cow's physical development and milk production. Maintaining a consistently high feed intake not only ensures that dairy cows receive comprehensive and abundant nutrients for normal growth and development but also achieves ideal milk production. Accurately detecting and calculating the real-time feed intake of dairy cows is the foundation for scientifically developing dairy cow nutrition programs.

[0003] In modern agriculture, the main method for measuring feed intake is the empirical method. The empirical method estimates how much feed a dairy cow consumes based on past data. This method requires a high level of expertise from the farmers and has a large margin of error, making it unsuitable for precise measurement. Summary of the Invention

[0004] This invention provides a method for monitoring the feeding behavior of dairy cows, which solves the problem of large errors in the existing technology that relies on feeding estimation by feeders.

[0005] A method for monitoring dairy cow feeding behavior includes:

[0006] A dairy cow feed intake prediction model was constructed, historical feed intake influencing factors and historical feed intake were obtained from the database, and the dairy cow feed intake prediction model was trained based on the historical feed intake influencing factors and historical feed intake to obtain the trained dairy cow feed intake prediction model.

[0007] Based on the trained dairy cow feed intake prediction model, the next feed intake is predicted, and the predicted feed intake is displayed to the staff to provide a reference for feed feeding.

[0008] After receiving the instruction on the amount of feed to be fed by the staff through the terminal or human-computer interaction, the automatic feeder will feed the feed according to the feed feeding control.

[0009] After the feed is fed, the acceleration information transmitted by the accelerometer worn by the cow is collected, and the actual feed intake of the cow is identified based on the transmitted acceleration information.

[0010] The predicted feed intake and actual feed intake of dairy cows are correlated and stored in the cloud to monitor the feeding behavior of dairy cows.

[0011] Furthermore, a dairy cow feed intake prediction model is constructed, including: constructing a neural network model with an input layer, a hidden layer, and an output layer connected in sequence, and using this neural network model as the dairy cow feed intake prediction model.

[0012] Furthermore, historical factors influencing feed intake and historical feed intake amounts are obtained from the database and pre-stored. The dairy cow feed intake prediction model is then trained based on these historical factors and amounts to obtain a fully trained dairy cow feed intake prediction model, including:

[0013] Obtain historical feeding influencing factors pre-stored in the database. These historical feeding influencing factors include feeding time, lying time, number of lying times, number of steps, feed concentrate-to-roughage ratio, and body weight corresponding to the historical time period.

[0014] Retrieve historical feed intake data pre-stored in the database;

[0015] The input data for constructing a dairy cow feed intake prediction model is based on historical feed intake influencing factors, and the historical feed intake is used as the expected output data of the dairy cow feed intake prediction model.

[0016] Based on the input data and the expected output data, the dairy cow feed intake prediction model is trained to obtain the trained dairy cow feed intake prediction model.

[0017] Furthermore, based on the input data and the expected output data, the dairy cow feed intake prediction model is trained to obtain the trained dairy cow feed intake prediction model, including:

[0018] Set the iteration counter t=1, randomly initialize multiple particles, the value of each particle is within a predetermined range, and each particle includes D-dimensional parameters, where D represents the number of parameters to be trained in the dairy cow feed intake prediction model;

[0019] Determine the fitness of each particle, and take the position of each particle as its individual optimal position. Take the position of the optimal individual with the highest fitness among all particles as the global optimal position.

[0020] Each particle is updated based on its individual optimal position and global optimal position, while simultaneously handling out-of-bounds errors, resulting in the updated particle.

[0021] Obtain the fitness of the updated particles and determine whether the fitness of each updated particle is greater than the fitness of its individual optimal position. If so, update the individual optimal position; otherwise, proceed to the update step of the global optimal position.

[0022] Determine whether the fitness of each individual's optimal position is greater than the fitness of the global optimal position. If so, replace the global optimal position with the optimal position of that individual; otherwise, proceed to the determination step of the iteration counter.

[0023] Determine if the count value of the iteration counter is greater than the maximum number of updates. If so, output the global optimal position and use this global optimal position as the network parameter of the cow feed intake prediction model to obtain the trained cow feed intake prediction model. Otherwise, return to the step of updating particles.

[0024] Furthermore, the fitness of the particle is obtained through a fitness function, which is:

[0025]

[0026] Where f represents fitness, d nj This represents the actual output data of the j-th neuron in the output layer of the dairy cow feed intake prediction model under the n-th input, where n = 1, 2, ..., N, N represents the batch size of the data used to train the dairy cow feed intake prediction model, and j = 1, 2, ..., J, J represents the total number of neurons in the output layer. nj d nj The corresponding expected output data.

[0027] Furthermore, based on the individual optimal position and the global optimal position, each particle is updated, and out-of-bounds handling is performed simultaneously, resulting in updated particles, including:

[0028] Based on the individual optimal position and the global optimal position, each particle is updated as follows:

[0029]

[0030] in, This represents the d-th dimension parameter of the i-th particle in the t-th update. Indicates the updated ω t Let c1 represent the inertia weight in the t-th update, c1 represent the first update coefficient, r1 represent the first random number between (0,1), and p represent the first random number between (0,1). ad c1 represents the mean of the optimal positions of all individuals, c2 represents the second update coefficient, r2 represents the second random number between (0,1), and p g Indicates the globally optimal position;

[0031] The particles are processed to avoid exceeding the limit, resulting in updated particles.

[0032] Furthermore, the inertia weight ω in the t-th update t for:

[0033]

[0034]

[0035] Among them, w maxw represents the maximum value of the inertia weight. max Set to 0.9; w min w represents the minimum inertia weight. min Set to 0.4; b represents the damping factor, k t f represents the population evolution factor in the t-th update. ad t f represents the average fitness of all particles in the t-th update. ad m This represents the average fitness of all particles in the m-th update.

[0036] Furthermore, based on the trained dairy cow feed intake prediction model, the next feed intake is predicted to obtain the predicted feed intake, including:

[0037] Obtain the target feed intake influencing factors for dairy cows, use these factors as input to the trained dairy cow feed intake prediction model, and predict the next feed intake to obtain the predicted feed intake.

[0038] Furthermore, after feed is administered, acceleration information transmitted from accelerometers worn by the cows is collected, and based on this acceleration information, the cows' current actual feed intake is identified, including:

[0039] After the feed is fed, the acceleration information transmitted by the accelerometer worn on the cow is collected. The acceleration information is input into a pre-trained neural network for recognizing feeding behavior categories to obtain the cow's feeding state, which includes the cow's feeding state, chewing state, and non-feeding state.

[0040] The target acceleration information of dairy cows in the feeding state of curling and chewing is determined, and the target acceleration information is input into a pre-trained feeding rate estimation neural network to obtain the feeding rate prediction value.

[0041] Based on the predicted feeding rate and the corresponding duration, the actual current feed intake of the dairy cow is obtained.

[0042] Furthermore, the predicted feed intake and actual feed intake of dairy cows are correlated and stored in the cloud to monitor the feeding behavior of dairy cows, including:

[0043] The system obtains the identity information of the current dairy cow, and uses this information as a basis to link the predicted feed intake and actual feed intake of the current dairy cow into a data set. This linked data is then stored in the cloud to monitor the dairy cow's feeding behavior.

[0044] This invention provides a method for monitoring dairy cow feeding behavior. Based on historical factors influencing feeding and historical feed intake, a dairy cow feed intake prediction model is trained, enabling the model to predict feed intake and providing a reference for feed distribution. After feed is distributed, feed intake data can be collected to monitor the cows' feeding behavior, effectively improving feed intake and allowing staff to manage the cows based on their feeding behavior, thus increasing management efficiency. Attached Figure Description

[0045] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0046] Figure 1 This is a flowchart illustrating a method for monitoring dairy cow feeding behavior according to an embodiment of the present invention.

[0047] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0048] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0049] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0050] like Figure 1 As shown, a method for monitoring dairy cow feeding behavior includes:

[0051] S101. Construct a dairy cow feed intake prediction model, obtain historical feed intake influencing factors and historical feed intake pre-stored in the database, and train the dairy cow feed intake prediction model based on the historical feed intake influencing factors and historical feed intake to obtain the trained dairy cow feed intake prediction model.

[0052] A dairy cow feed intake prediction model can be set up as a classification neural network to learn the correlation between historical feed intake influencing factors and historical feed intake.

[0053] Optionally, neural network training algorithms such as gradient descent, genetic algorithm, and wolf pack optimization algorithm can be used to train the dairy cow feed intake prediction model to obtain a trained dairy cow feed intake prediction model.

[0054] S102. Based on the trained dairy cow feed intake prediction model, predict the next feed intake, obtain the predicted feed intake, and display the predicted feed intake to the staff to provide a reference for feed feeding.

[0055] The trained dairy cow feed intake prediction model has the ability to identify factors influencing feed intake. Therefore, these factors can be collected and identified using the trained model to obtain predicted feed intake. It is worth noting that, to ensure accuracy, the data format of the currently collected feed intake influencing factors should be the same as the historical data format to guarantee correct data identification.

[0056] S103. After receiving the instruction on the amount of feed to be fed by the staff through the terminal or human-machine interaction, the automatic feeder is fed according to the feed feeding control.

[0057] Automatic feeders are a relatively common device in existing technology. They can deliver a fixed amount of feed and identify the actual amount of feed consumed after feeding.

[0058] S104. After the feed is fed, the acceleration information transmitted by the accelerometer worn on the cow is collected, and the actual feed intake of the cow is identified based on the transmitted acceleration information.

[0059] The dairy cows are fed twice a day at set times. After each feeding, the actual feed intake can be recorded. At this time, the factors affecting feed intake can be the data from the day before the scheduled feeding, thus achieving data consistency.

[0060] Optionally, when dairy cows are kept in pens, the remaining feed in the trough should be weighed and recorded before each feeding. The amount of feed from the previous feeding should be subtracted from the remaining amount to obtain the actual feed intake. The sum of the two feed intakes is the total feed intake for one day.

[0061] However, in current technologies, dairy cows are often not fed individually, so multiple cows may feed simultaneously. It's also possible that cows graze freely in pastures without needing to be fed, making it cumbersome to collect feed intake data by weighing. Therefore, a solution is to equip dairy cows with smart collars. These collars possess unique identification information and can collect acceleration data, thereby enabling the determination of the cows' feeding behavior.

[0062] Applying the solution described in this embodiment to an environment where dairy cows are raised individually will yield even better results. It is worth noting that even when dairy cows are fed in groups, predicted feed intake can still be obtained, allowing staff to determine the total amount of feed to be given. Furthermore, predicting feed intake makes it easier for staff to manage the dairy cows.

[0063] S105. The predicted feed intake and actual feed intake of dairy cows are correlated and stored in the cloud to complete the monitoring of dairy cow feeding behavior.

[0064] To facilitate monitoring of dairy cows' feeding behavior, the predicted and actual feed intake of dairy cows are uploaded to the cloud. This allows staff to view historical data and remotely check data, making management more convenient.

[0065] In dairy cow production, monitoring feed intake is crucial. Maintaining a high feed intake ensures that dairy cows receive comprehensive and abundant nutrients, promoting normal growth and development, and ultimately achieving the desired milk yield. Accurately monitoring and calculating real-time feed intake satisfies the nutritional needs of dairy cows while preventing waste, thus contributing to good production efficiency and economic benefits.

[0066] This invention provides a method for monitoring dairy cow feeding behavior. Based on historical factors influencing feeding and historical feed intake, a dairy cow feed intake prediction model is trained, enabling the model to predict feed intake and providing a reference for feed distribution. After feed is distributed, feed intake data can be collected to monitor the cows' feeding behavior, effectively improving feed intake and allowing staff to manage the cows based on their feeding behavior, thus increasing management efficiency.

[0067] In this embodiment, constructing a dairy cow feed intake prediction model includes: constructing a neural network model with an input layer, a hidden layer, and an output layer connected sequentially, and using this neural network model as the dairy cow feed intake prediction model. For example, a BP (Back Propagation) neural network or an RBF (Radial Basis Function) neural network can be used as the dairy cow feed intake prediction model.

[0068] In this embodiment, historical feed intake influencing factors and historical feed intake amounts are obtained from a pre-stored database, and a dairy cow feed intake prediction model is trained based on these historical factors and amounts to obtain a trained dairy cow feed intake prediction model, including:

[0069] Obtain historical feeding influencing factors pre-stored in the database. These historical feeding influencing factors include feeding time, lying time, number of lying times, steps, feed concentrate-to-roughage ratio, and body weight corresponding to the historical time period.

[0070] Feeding time refers to the duration of feeding between the first feeding and the next feeding. Lying time, number of times lying down, and steps are all acquired through smart collars, while weight and feed concentrate-to-roughage ratio are uploaded through staff's terminal devices.

[0071] Retrieve historical feed intake data stored in the database.

[0072] The input data for constructing a dairy cow feed intake prediction model is based on historical feed intake influencing factors, and the historical feed intake is used as the expected output data of the dairy cow feed intake prediction model.

[0073] Based on the input data and the expected output data, the dairy cow feed intake prediction model is trained to obtain the trained dairy cow feed intake prediction model.

[0074] Optionally, an intelligent optimization algorithm can be used to train the weight parameters and bias parameters of the dairy cow feed intake prediction model to obtain the trained dairy cow feed intake prediction model.

[0075] In this embodiment, the dairy cow feed intake prediction model is trained based on the input data and the expected output data to obtain the trained dairy cow feed intake prediction model, including:

[0076] Set the iteration counter t=1, randomly initialize multiple particles, the value of each particle is within a predetermined range, and each particle includes D-dimensional parameters, where D represents the number of parameters to be trained in the dairy cow feed intake prediction model.

[0077] Optionally, upper and lower limits can be set for each dimension parameter. Each dimension parameter can be randomly generated between the upper and lower limits, or other initial parameter generation strategies can be used to generate multiple particles to make the solution more evenly distributed in the solution space, which can help to speed up the convergence speed and improve the optimization accuracy.

[0078] Determine the fitness of each particle, and take the position of each particle as its individual optimal position. Take the position of the optimal individual with the highest fitness among all particles as the global optimal position.

[0079] In addition to evaluating each particle using its fitness, an error function can also be used to evaluate each particle, thereby determining the quality of each particle's position.

[0080] Each particle is updated based on its individual optimal position and the global optimal position, while simultaneously handling out-of-bounds errors, resulting in the updated particle.

[0081] Optionally, the out-of-bounds handling may include: when a particle's parameter in a certain dimension exceeds its upper limit, transforming it to the upper limit or randomly generating the parameter in that dimension between the lower and upper limits; when a particle's parameter in a certain dimension is below its lower limit, transforming it to the lower limit or randomly generating the parameter in that dimension between the lower and upper limits.

[0082] Obtain the fitness of the updated particles and determine whether the fitness of each updated particle is greater than the fitness of its individual optimal position. If so, update the individual optimal position; otherwise, proceed to the update step of the global optimal position.

[0083] Determine whether the fitness of each individual's optimal position is greater than the fitness of the global optimal position. If so, replace the global optimal position with the optimal position of that individual; otherwise, proceed to the determination step of the iteration counter.

[0084] Determine if the count value of the iteration counter is greater than the maximum number of updates. If so, output the global optimal position and use this global optimal position as the network parameter of the cow feed intake prediction model to obtain the trained cow feed intake prediction model. Otherwise, return to the step of updating particles.

[0085] Optionally, to enhance the training effect, a simulated annealing algorithm can be introduced. Combining the simulated annealing algorithm with the algorithm provided in this embodiment can accelerate the training speed and improve the training effect.

[0086] In this embodiment, the fitness of the particle is obtained through a fitness function, which is:

[0087]

[0088] Where f represents fitness, d nj This represents the actual output data of the j-th neuron in the output layer of the dairy cow feed intake prediction model under the n-th input, where n = 1, 2, ..., N, N represents the batch size of the data used to train the dairy cow feed intake prediction model, and j = 1, 2, ..., J, J represents the total number of neurons in the output layer. nj d nj The corresponding expected output data.

[0089] It is worth noting that, in addition to the fitness function described in this embodiment, other functions that can evaluate the quality of the network can also be used as the fitness function.

[0090] In this embodiment, each particle is updated based on its individual optimal position and the global optimal position, while simultaneously handling out-of-bounds errors, resulting in updated particles, including:

[0091] Based on the individual optimal position and the global optimal position, each particle is updated as follows:

[0092]

[0093] in, This represents the d-th dimension parameter of the i-th particle in the t-th update. Indicates the updated ω t Let c1 represent the inertia weight in the t-th update, c1 represent the first update coefficient, r1 represent the first random number between (0,1), and p represent the first random number between (0,1). ad c1 represents the mean of the optimal positions of all individuals, c2 represents the second update coefficient, r2 represents the second random number between (0,1), and p g This represents the global optimal position; when a particle goes out of bounds, an updated particle is obtained.

[0094] Optionally, this embodiment provides transformation formulas for the first update coefficient c1 and the second update coefficient c2, specifically as follows:

[0095]

[0096] Among them, c 1int c represents the initial value of the first update coefficient. 1fin represents the final value of the first update coefficient, t represents the current update count, and T represents the maximum update count. Once the maximum update count is reached, the update can end, and the globally optimal position will be output. 2int c represents the initial value of the second update coefficient. 2fin This represents the final value of the second update coefficient.

[0097] In this embodiment, the inertia weight ω in the t-th update t for:

[0098]

[0099]

[0100] Among them, w max w represents the maximum value of the inertia weight. max Set to 0.9; w min w represents the minimum inertia weight. min Set to 0.4; b represents the damping factor, k t f represents the population evolution factor in the t-th update. adt f represents the average fitness of all particles in the t-th update. ad m This represents the average fitness of all particles in the m-th update.

[0101] It can be observed that the evolutionary process causes particles to gradually approach the global optimum. In the early stages of evolution, a relatively dispersed population is needed to find the approximate location of the global optimum, so the inertia weight decreases relatively smoothly and uniformly. In the later stages of evolution, as particles continuously approach the global optimum, the inertia weight oscillates more significantly, effectively enhancing the particle's local search ability. Therefore, the dynamically adaptive inertia weight ensures the algorithm's ability to balance local and global search.

[0102] Although the optimization algorithm in this embodiment can take into account both global and local optimization, the accuracy of local optimization is not as good as that of gradient descent. Therefore, in order to ensure training quality, gradient descent can be used for retraining, thereby satisfying the training accuracy and avoiding getting trapped in local optima.

[0103] In this embodiment, based on the trained dairy cow feed intake prediction model, the next feed intake is predicted to obtain the predicted feed intake, including:

[0104] Obtain the target feed intake influencing factors for dairy cows, use these factors as input to the trained dairy cow feed intake prediction model, and predict the next feed intake to obtain the predicted feed intake.

[0105] In this embodiment, after feed is given, acceleration information transmitted by an accelerometer worn by the cow is collected, and based on the transmitted acceleration information, the cow's current actual feed intake is identified, including:

[0106] After the feed is fed, the acceleration information transmitted by the accelerometer worn on the cow is collected. The acceleration information is then input into a pre-trained neural network for recognizing feeding behavior categories to obtain the cow's feeding state, which includes the cow's rolling feeding state, chewing state, and non-feeding state.

[0107] The target acceleration information of dairy cows in the feeding states of curling and chewing is determined, and the target acceleration information is input into a pre-trained feeding rate estimation neural network to obtain the feeding rate prediction value.

[0108] Optionally, the feeding rate estimation neural network can also be constructed from an input layer, a hidden layer, and an output layer, and trained using the neural network training method described above, thereby achieving the prediction of the feeding rate.

[0109] The target acceleration information can be the triaxial acceleration information sensed by the triaxial accelerometer.

[0110] After each feeding, a cow may chew multiple times. Therefore, by adding up the time for each feeding and the corresponding chewing time, we can obtain the feeding time for a single feeding session. Then, by adding up all the feeding times for each single session, we can obtain the cow's total feeding time.

[0111] Based on the predicted feeding rate and the corresponding duration, the actual current feed intake of the dairy cow is obtained.

[0112] Based on the predicted feeding rate and the corresponding duration, the actual feed intake of the dairy cow is obtained, including:

[0113] Based on the single-time rolling behavior, the time for each rolling and chewing after rolling is determined. Multiplying the single-time eating time by the predicted feeding rate gives the actual amount of food consumed per feeding.

[0114] By accumulating multiple actual feed intakes, the current actual feed intake of a dairy cow can be obtained. It is worth noting that the current actual feed intake of a dairy cow refers to the actual feed intake collected during the cow's feeding period.

[0115] In this embodiment, the predicted feed intake and actual feed intake of dairy cows are correlated and stored in the cloud to complete the monitoring of dairy cow feeding behavior, including:

[0116] The smart collar can acquire the identity information of the cow and, based on this information, link the cow's predicted feed intake with its actual feed intake. This linked data is then stored in the cloud to monitor the cow's feeding behavior.

[0117] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0118] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method of monitoring the feeding behavior of a dairy cow, characterized in that, include: A dairy cow feed intake prediction model is constructed, historical feed intake influencing factors and historical feed intake are obtained from a pre-stored database, and the model is trained based on these factors to obtain a trained model. This includes: Obtain historical feeding influencing factors pre-stored in the database. These factors include feeding time, lying time, number of lying times, steps, feed concentrate-to-roughage ratio, and body weight corresponding to historical time periods. Obtain historical feed intake pre-stored in the database. Construct input data for a dairy cow feed intake prediction model using historical feeding influencing factors, and use historical feed intake as the expected output data for the model. Train the dairy cow feed intake prediction model based on the input data and the expected output data to obtain the trained model. Based on the input data and the expected output data, the dairy cow feed intake prediction model is trained to obtain the trained dairy cow feed intake prediction model, including: Setting iteration counter t =1, randomly initializing a plurality of particles, each of the particles having a value within a predetermined range, and each of the particles including D dimensional parameters, D representing the number of parameters to be trained in the cow intake prediction model; Determine the fitness of each particle, and take the position of each particle as its individual optimal position. Take the position of the optimal individual with the highest fitness among all particles as the global optimal position. Each particle is updated based on its individual optimal position and global optimal position, while simultaneously handling out-of-bounds errors, resulting in the updated particle. Obtain the fitness of the updated particles and determine whether the fitness of each updated particle is greater than the fitness of its individual optimal position. If so, update the individual optimal position; otherwise, proceed to the update step of the global optimal position. Determine whether the fitness of each individual's optimal position is greater than the fitness of the global optimal position. If so, replace the global optimal position with the optimal position of that individual; otherwise, proceed to the determination step of the iteration counter. Determine if the count value of the iteration counter is greater than the maximum number of updates. If so, output the global optimal position and use this global optimal position as the network parameter of the cow feed intake prediction model to obtain the trained cow feed intake prediction model. Otherwise, return to the step of updating particles. Based on the trained dairy cow feed intake prediction model, the next feed intake is predicted, and the predicted feed intake is displayed to the staff to provide a reference for feed feeding. After receiving the instruction on the amount of feed to be fed by the staff through the terminal or human-computer interaction, the automatic feeder will feed the feed according to the feed feeding control. After the feed is fed, the acceleration information transmitted by the accelerometer worn by the cow is collected, and the actual feed intake of the cow is identified based on the transmitted acceleration information. The predicted feed intake and actual feed intake of dairy cows are correlated and stored in the cloud to monitor the feeding behavior of dairy cows.

2. The dairy cow feeding behavior monitoring method according to claim 1, characterized in that, Constructing a dairy cow feed intake prediction model includes: constructing a neural network model with an input layer, hidden layer, and output layer connected in sequence, and using this neural network model as the dairy cow feed intake prediction model.

3. The dairy cow feeding behavior monitoring method according to claim 1, characterized in that, The fitness of a particle is obtained through a fitness function, which is: wherein, represents fitness, represents actual output data of the i-th neuron in the output layer of the cow intake prediction model under the i-th input, n represents actual output data of the i-th neuron in the output layer of the cow intake prediction model under the i-th input, j = 1, 2,.., n , N , N represents batch size of data for training the cow intake prediction model, j = 1, 2,.., J , J represents total number of neurons in the output layer, represents corresponding expected output data.

4. The dairy cow feeding behavior monitoring method according to claim 1, characterized in that, According to the individual optimal position and the global optimal position, each particle is updated, and at the same time, the out-of-bound processing is performed, to obtain an updated particle, including: According to the individual optimal position and the global optimal position, each particle is updated, and at the same time, the out-of-bound processing is performed, to obtain an updated particle, including: wherein, denotes the t dimensional parameter of the i particle in the d th update, denotes the updated , t denotes the first update coefficient, denotes a first random number between (0,1), denotes the mean of all individual optimal positions, denotes the second update coefficient, denotes a second random number between (0,1), denotes the global optimal position; The particle is subjected to the out-of-bound processing, to obtain an updated particle.

5. The dairy cow feeding behavior monitoring method according to claim 4, characterized in that, No. t Inertia weights in the next update for: wherein, represents the maximum value of the inertia weight, is set to 0.9; represents the minimum value of the inertia weight, is set to 0.4; represents the damping factor, represents the population evolution factor in the t th update, represents the mean of the fitness of all particles in the t th update, represents the mean of the fitness of all particles in the m th update.

6. The dairy cow feeding behavior monitoring method according to claim 1, characterized in that, Based on the trained dairy cow feed intake prediction model, the next feed intake is predicted to obtain a predicted feed intake, including: Obtaining the target feed influencing factor corresponding to the dairy cow, taking the target feed influencing factor as the input of the trained dairy cow feed intake prediction model, predicting the next feed intake to obtain the predicted feed intake.

7. The dairy cow feeding behavior monitoring method according to claim 1, characterized in that, After the feed is fed, the acceleration information transmitted by the accelerometer worn on the dairy cow is collected, and based on the transmitted acceleration information, the current actual feed intake of the dairy cow is identified, including: After the feed is fed, the acceleration information transmitted by the accelerometer worn on the dairy cow is collected, and based on the transmitted acceleration information, the current actual feed intake of the dairy cow is identified, including: Determining the target acceleration information when the dairy cow feed state is the rolling state and the chewing state, and inputting the target acceleration information into the pre-trained feed rate estimation neural network to obtain a feed rate prediction value; According to the feed rate prediction value and the corresponding duration, the current actual feed intake of the dairy cow is obtained.

8. The dairy cow feeding behavior monitoring method according to claim 1, characterized in that, The predicted feed intake and the actual feed intake of the dairy cow are associated and stored in the cloud to complete the monitoring of the dairy cow feed behavior, including: Obtaining the identity information corresponding to the current dairy cow, and taking the identity information corresponding to the current dairy cow as the basis, associating the predicted feed intake and the actual feed intake of the current dairy cow to form associated data, and storing the associated data in the cloud to complete the monitoring of the dairy cow feed behavior.

Images