A precision feeding system suitable for dairy farming

Abstract

This invention discloses a precision feeding system suitable for dairy farming, belonging to the field of dairy farming technology. It is equipped with a first pressure sensor and a second pressure sensor, which can measure the weight of feed in the feed trough and the weight of dairy cows in the feeding cage in real time, and record historical data of each dairy cow. Based on the historical data, the weight of feed in the feed trough and the weight of dairy cows in the feeding cage, and by introducing a deep learning method, the feed amount can be obtained, thereby achieving precision feeding in dairy farming.

Description

Technical Field

[0001] This invention belongs to the field of dairy farming technology, specifically relating to a precision feeding system suitable for dairy farming. Background Technology

[0002] With the rapid development of animal husbandry, small-scale, scattered farming has gradually shifted to large-scale dairy farming, leading to the rapid development of the dairy product industry. Milk has become a staple food in people's daily lives, and the health of dairy cows and the quality of milk directly affect people's lives. However, traditional large-scale farming mainly relies on manual feeding, with staff regularly inspecting the condition of dairy cows and using mechanized milking. The milk goes directly to subsequent processing stages, making it impossible to monitor each dairy cow's life process from feeding to milk production in real time, and thus unable to control the health of dairy cows and the quality of milk in a timely manner.

[0003] In current dairy farming practices, automatic feeders are often used to reduce the tediousness and inefficiency of manual feeding, thus enabling timed feeding. However, automatic feeders can only achieve timed feeding and cannot provide precise, intensive feeding for dairy cows. Summary of the Invention

[0004] This invention provides a precision feeding system suitable for dairy farming, in order to solve the problems existing in the prior art.

[0005] A precision feeding system suitable for dairy farming includes an automatic feeder, a feed trough located below the outlet of the automatic feeder, a first pressure sensor located below the feed trough, a feeding cage for carrying dairy cows, a second pressure sensor located below the feeding cage, and an intelligent controller.

[0006] The automatic feeder is used to carry feed and receive control from the intelligent controller to feed the feed into the trough below the outlet of the automatic feeder.

[0007] The feed trough is installed in the feeding cage that carries the dairy cows, and the feed trough is used to receive feed from the automatic feeder.

[0008] The first pressure sensor is used to sense the first total weight corresponding to the feed trough carrying the feed in real time, and to feed back the first total weight to the intelligent controller;

[0009] The feeding cage is used to hold the feed trough and the cows, and only one cow is allowed to enter at a time, thus dividing the cows into sections for feeding.

[0010] The second pressure sensor is used to sense the second total weight of the feed trough and the cow in real time, and to feed the second total weight back to the intelligent controller;

[0011] The intelligent controller is used to sense the RFID ear tags worn by the cows, determine the unique RFID information corresponding to the cows, store the historical feeding data of each cow based on the unique RFID information of the cows, and obtain the feed input amount based on the historical feeding data, the first total weight and the second total weight.

[0012] The intelligent controller is also used to control the automatic feeder to feed the corresponding weight of feed into the feed trough according to the feed feeding amount.

[0013] Furthermore, the method for obtaining the weight of the feed in the trough is as follows:

[0014] When no feed is added to the trough, the first sensing value of the first pressure sensor is obtained.

[0015] When feed is added to the trough, the second sensing value of the first pressure sensor is obtained, and the weight of the feed is obtained by subtracting the first sensing value from the second sensing value.

[0016] Furthermore, the method for obtaining the weight of the dairy cows in the feeding cage is as follows:

[0017] When there are no cows in the feeding cage, acquire the third sensing value of the second pressure sensor;

[0018] When there are cows in the feeding cage, the fourth sensor value of the second pressure sensor is obtained, and the weight of the cow is obtained by subtracting the third sensor value from the fourth sensor value.

[0019] Furthermore, based on the unique RFID information of each dairy cow, historical feeding data for each cow is stored, and the feed input amount is obtained based on the historical feeding data, the first total weight, and the second total weight, including:

[0020] Based on the unique RFID information of each dairy cow, the historical feeding time period of each dairy cow and the corresponding feed amount for each historical feeding time period are stored on a daily basis to obtain the historical feeding data of the dairy cow;

[0021] The system senses the RFID ear tag of the milk in the feeding cage and obtains the target RFID information corresponding to the target cow that is currently feeding, as well as the current time period.

[0022] Based on the second total weight, the target weight corresponding to the target cow that is currently feeding is determined, and the weight range in which the target weight is located is determined, thereby determining the feed weight range corresponding to the weight range. The weight range is preset data, and each weight range is preset with a feed weight range.

[0023] Based on the current time period, determine the feed amount of the target cow in the historical feeding data corresponding to the current time period. Use the feed amount and historical weight of the historical time periods in N consecutive days as input data, and use the feed amount on the N+1th day as the expected output data to obtain multiple sets of training data for the feeding prediction model. Among them, the historical weight is collected along with the cow's historical feeding data.

[0024] Based on multiple sets of training data, the feeding prediction model is trained to obtain the trained feeding prediction model.

[0025] Based on the current time period, obtain the feed intake and historical weight of the previous N consecutive days, and use them as input data for the trained feed prediction model to obtain the predicted feed intake of the target dairy cow.

[0026] Determine whether the predicted feed intake of the target dairy cow is within the corresponding feed input weight range. If so, obtain the feed input based on the predicted feed intake of the target dairy cow; otherwise, adjust it to the nearest boundary value of the feed input weight range.

[0027] Based on the current total weight, obtain the weight of the remaining feed in the trough. Subtract the weight of the remaining feed in the trough from the predicted feed intake to obtain the feed input amount.

[0028] Furthermore, the feeding prediction model is set as a classification neural network that takes vector data or matrix data as input. Before the input data is input into the feeding prediction model, it is converted into the data type corresponding to the feeding prediction model.

[0029] Furthermore, based on multiple sets of training data, the feeding prediction model is trained to obtain a fully trained feeding prediction model, including:

[0030] The network parameters of the feeding prediction model are initialized using a chaotic sequence strategy to obtain individual network parameters; wherein, depending on the data type corresponding to the feeding prediction model, the individual network parameters are either vector parameters or matrix parameters.

[0031] Repeatedly acquire multiple individual network parameters;

[0032] The food intake and historical weight of N consecutive days in the training data are used as the input data for the food prediction model, and the food intake on the N+1th day is used as the expected output data to obtain the fitness of individual network parameters.

[0033] The network parameter individual with the highest fitness is taken as the global optimum, and a global search is performed on all network parameter individuals based on the global optimum to obtain the first updated value of the network parameter individual.

[0034] Based on the first updated value of the individual network parameter, a further local search is performed to obtain the second updated value of the individual network parameter;

[0035] Based on the first and second update values ​​of the individual network parameters, the individual network parameters are updated to obtain the updated individual network parameters.

[0036] Determine if there exists an individual with updated network parameters whose fitness is greater than a set threshold or whose current update count t is greater than the maximum update count T. If so, use the individual with the highest fitness as the final network parameter of the feeding prediction model to obtain the trained feeding prediction model; otherwise, proceed to the next training iteration.

[0037] Furthermore, the fitness is obtained through a fitness function, which is the negative of the error function of the feeding prediction model.

[0038] Furthermore, based on the global optimum, a global search is performed on all individual network parameters to obtain the first updated value of each individual network parameter:

[0039] f i =f min +(f max -f min )*β

[0040]

[0041] ω=t / T

[0042]

[0043] Among them, f i f represents the update frequency of the i-th individual network parameter. max f represents the upper limit of the update frequency. min This represents the lower bound of the update frequency, and β represents a random number between (0,1); Let represent the update step size of the i-th network parameter individual during the t-th training iteration, and Update after handling out-of-bounds errors Let ω represent the update step size of the i-th network parameter individual during the (t-1)-th training iteration, and let ω represent the inertial weights. This represents the i-th network parameter during the (t-1)-th training iteration. Let represent the global optimum value at the (t-1)th training iteration, and T represent the pre-set maximum number of updates. Indicates the updated That is, the first updated value of the individual network parameter.

[0044] Furthermore, based on the first updated value of the individual network parameter, a local search is performed to obtain the second updated value of the individual network parameter:

[0045]

[0046] in, Indicates the updated That is, the second updated value of the individual network parameter; ε represents the update coefficient, and ε is a constant between [-1, 1], A t-1 The value represents the average loudness of all individual network parameters during this training, and rand represents a random number between (0,1). i t-1 This represents the pulse emission frequency corresponding to the i-th network parameter individual during the (t-1)-th training iteration.

[0047] Furthermore, based on the first and second update values ​​of the individual network parameters, it is determined that the individual network parameters should be updated, resulting in the updated individual network parameters as follows:

[0048]

[0049]

[0050]

[0051] Among them, W i t This represents the i-th updated network parameter. This represents the loudness of the i-th network parameter during the (t-1)-th training iteration. Represents individual network parameters fitness value, This represents the maximum fitness value corresponding to an individual in the network parameters. Indicates the updated The initial values ​​corresponding to the i-th network parameter individual The value is 0.95, where α represents the attenuation coefficient between (0,1), and r... i t Indicates the updated r i t-1 γ is greater than 0, where γ represents the increment coefficient, r i 0 Let r represent the initial pulse emission frequency of the i-th network parameter individual, and r i 0 =0.9.

[0052] This invention provides a precision feeding system suitable for dairy farming, which is equipped with a first pressure sensor and a second pressure sensor. It can measure the weight of feed in the feed trough and the weight of dairy cows in the feeding cage in real time, and record the historical data of each dairy cow. Based on the historical data, the weight of feed in the feed trough and the weight of dairy cows in the feeding cage, and by introducing a deep learning method, the feed amount can be obtained, thereby achieving precision feeding in dairy farming. Attached Figure Description

[0053] 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.

[0054] Figure 1 This is a schematic diagram of a precision feeding system for dairy farming provided in an embodiment of the present invention.

[0055] Among them, 1-automatic feeder, 2-feed trough, 3-first pressure sensor, 4-feeding cage, 5-second pressure sensor, and 6-intelligent controller.

[0056] 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

[0057] 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.

[0058] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0059] like Figure 1 As shown, a precision feeding system suitable for dairy farming includes an automatic feeder 1, a feed trough 2 located below the outlet of the automatic feeder 1, a first pressure sensor 3 located below the feed trough 2, a feeding cage 4 for carrying dairy cows, a second pressure sensor 5 located below the feeding cage 4, and an intelligent controller 6.

[0060] The automatic feeder 1 is used to carry feed and receive control from the intelligent controller 6 to feed the feed into the feed trough 2 below the outlet of the automatic feeder 1.

[0061] In the prior art, there are many automatic feeding machines controlled by host computers or other data processing devices, so this embodiment will not describe the automatic feeder in detail.

[0062] The feed trough 2 is installed in the feeding cage 4 that carries the dairy cows, and the feed trough 2 is used to receive the feed dispensed by the automatic feeder 1.

[0063] The feed trough 2 is used to feed dairy cows and is set below the discharge port of the automatic feeder 1 to achieve automatic feeding; and the feed trough 2 is set in the feeding cage 4 that carries the dairy cows, and the size of the feeding cage 4 is only allowed to allow one dairy cow to enter, thereby realizing the zoning of dairy cows.

[0064] The first pressure sensor 3 is used to sense the first total weight corresponding to the feed trough 2 carrying feed in real time, and to feed back the first total weight to the intelligent controller 6.

[0065] Optionally, the first pressure sensor 3 can be replaced by a sensor matrix. The sensor matrix can measure more pressure values ​​to reduce measurement errors. For example, M first total weights can be measured, the maximum and minimum values ​​can be removed, and the average of the M-2 first total weights can be taken as the final first total weight.

[0066] The feeding cage 4 is used to hold the feed trough 2 and the dairy cows, and only one dairy cow is allowed to enter at a time, thereby dividing the dairy cows into sections for feeding.

[0067] The second pressure sensor 5 is used to sense the second total weight of the feed trough 2 and the cow in real time, and to feed back the second total weight to the intelligent controller 6.

[0068] Optionally, the second pressure sensor 5 can be replaced by a sensor matrix. The sensor matrix can measure more pressure values ​​to reduce measurement errors. For example, M second total weights can be measured, the maximum and minimum values ​​can be removed, and the average of the M-2 second total weights can be taken as the final second total weight.

[0069] The intelligent controller 6 is used to sense the RFID (Radio Frequency Identification) ear tags worn by the cows, determine the unique RFID information corresponding to the cows, store the historical feeding data of each cow based on the unique RFID information of the cows, and obtain the feed input amount based on the historical feeding data, the first total weight, and the second total weight.

[0070] To facilitate intelligent management of dairy cows, each cow needs to wear an RFID ear tag. When a cow enters feeding cage 4 to eat, the second pressure sensor 5 generates a pressure value. When the pressure value exceeds a certain threshold, it can be identified that a cow has entered, and the cow's unique RFID information can be read to retrieve its historical feeding data and store the current feeding data and the cow's corresponding weight. To avoid chaotic feeding, each cow can be equipped with its own feeding cage 4, thus achieving zoned and refined feeding.

[0071] The intelligent controller 6 is also used to control the automatic feeder 1 to feed the corresponding weight of feed into the feed trough 2 according to the feed feeding amount.

[0072] The intelligent controller 6 can be a computer, host computer, or other data processing device capable of performing the data processing process described in this embodiment. By using the weight data and historical feeding data of dairy cows, and with the aid of deep learning, it can predict feed intake and thus accurately deliver feed.

[0073] This invention provides a precision feeding system suitable for dairy farming, which is equipped with a first pressure sensor 3 and a second pressure sensor 5. It can measure the weight of feed in the feed trough 2 and the weight of dairy cows in the feeding cage 4 in real time, and record the historical data of each dairy cow. Based on the historical data, the weight of feed in the feed trough 2 and the weight of dairy cows in the feeding cage 4, and by introducing a deep learning method, the feed amount can be obtained, thereby achieving precision feeding in dairy farming.

[0074] In this embodiment, the method for obtaining the weight of the feed in the feed trough 2 is as follows:

[0075] When no feed is added to the feed trough 2, the first sensing value of the first pressure sensor 3 is obtained.

[0076] When feed is added to the feed trough 2, the second sensing value of the first pressure sensor 3 is obtained, and the weight of the feed is obtained by subtracting the first sensing value from the second sensing value.

[0077] After each feeding, some feed may remain uneaten. To conserve feed, the weight of the remaining feed can be recorded in real time. When feeding another cow, the weight of the remaining feed can be added to the feed, thus achieving the goal of saving feed.

[0078] In this embodiment, the method for obtaining the weight of the dairy cows in the feeding cage 4 is as follows:

[0079] When there are no cows in the feeding cage 4, the third sensing value of the second pressure sensor 5 is obtained.

[0080] When there are cows in the feeding cage 4, the fourth sensing value of the second pressure sensor 5 is obtained, and the weight of the cow is obtained by subtracting the third sensing value from the fourth sensing value.

[0081] To avoid false identification, a pressure threshold corresponding to the fourth sensor value can be set. When the fourth sensor value is less than the pressure threshold, it can be identified as a false trigger caused by a bird or other animal; when the fourth sensor value is greater than the pressure threshold, it can be identified as a cow being in a feeding cage, thus allowing subsequent operations to proceed.

[0082] In this embodiment, based on the unique RFID information of each dairy cow, historical feeding data of each cow is stored, and the feed input amount is obtained based on the historical feeding data, the first total weight, and the second total weight, including:

[0083] Based on the unique RFID information of each dairy cow, the historical feeding time period of each dairy cow and the corresponding feed amount for each historical feeding time period are stored on a daily basis to obtain the historical feeding data of the dairy cow.

[0084] Multiple weight ranges can be preset, each with an upper and lower limit for feed intake. To ensure precise feeding, feed intake can be predicted, and feed can be administered between the upper and lower limits to make dairy cow feeding more scientific.

[0085] The system senses the RFID ear tag of the milk in the current feeding cage 4, and obtains the target RFID information corresponding to the target cow that is currently feeding and the current time period.

[0086] Based on the second total weight, the target weight corresponding to the target dairy cow currently feeding is determined, and the weight range within which the target weight falls is determined. This, in turn, determines the corresponding feed weight range for that weight range. The weight range is preset data, and each weight range has a preset feed weight range. It is worth noting that the feed weight range refers to the amount of feed given per feeding, meaning that each time period has a corresponding feed weight range.

[0087] Optionally, for more refined feeding, feed delivery times can be set. When dairy cows eat at the designated feed delivery time, the feed intake of the target cow in the corresponding historical feeding time period can be determined based on the current time period. The feed intake and historical weight of the target cow in the historical feeding time periods over N consecutive days are used as input data, and the feed intake on day N+1 is used as the expected output data, resulting in multiple training datasets for the feeding prediction model. The historical weight is collected along with the cow's historical feeding data.

[0088] For example, given 50 days of data, use historical data from days 1-30 as input and data from day 31 as output to obtain one set of training data. Then, use historical data from days 2-31 as input and data from day 32 as output to obtain another set of training data, and so on, to obtain multiple sets of training data.

[0089] The feeding prediction model is trained based on multiple sets of training data to obtain the trained feeding prediction model.

[0090] Optionally, the feeding prediction model can be set as a classification neural network, such as a convolutional neural network or a back propagation (BP) neural network. The feeding prediction model can be trained using intelligent optimization algorithms or gradient descent methods to obtain a trained feeding prediction model.

[0091] Based on the current time period, obtain the feed intake and historical body weight of the previous N consecutive days, and use them as input data for the trained feed prediction model to obtain the predicted feed intake of the target dairy cow.

[0092] Determine whether the predicted feed intake of the target dairy cow is within the corresponding feed input weight range. If so, obtain the feed input based on the predicted feed intake of the target dairy cow; otherwise, adjust it to the nearest boundary value of the feed input weight range.

[0093] The feed intake weight range can be pre-set data or an industry standard. When the predicted feed intake of the target dairy cow falls within the corresponding feed intake weight range, feed can be directly administered to achieve precise feeding. When the predicted feed intake of the target dairy cow does not fall within the corresponding feed intake weight range, it indicates that the cow's recent feed intake is not up to standard. Therefore, it is adjusted to the nearest boundary value of the feed intake weight range. That is, when the dairy cow's feed intake is low, more feed is given to promote its feeding; when the dairy cow's feed intake is high, less feed is given to control its feeding.

[0094] Based on the current total weight, obtain the weight of the remaining feed in trough 2, and subtract the weight of the remaining feed in trough 2 from the predicted feeding amount to obtain the feed input amount.

[0095] There may be uneaten feed in the feed trough 2. To avoid feed waste, the weight of the remaining feed can be subtracted from the predicted feed intake, thereby further realizing refined feeding.

[0096] In this embodiment, the feeding prediction model is set as a classification neural network that takes vector data or matrix data as input. Before the input data is input into the feeding prediction model, it is converted into the data type corresponding to the feeding prediction model.

[0097] For example, if the input data of a BP neural network is in vector form, then the input data needs to be converted into vector form; convolutional neural networks are mainly used for graph recognition, so the input data needs to be converted into matrix form.

[0098] In this embodiment, the feeding prediction model is trained based on multiple sets of training data to obtain a trained feeding prediction model, including:

[0099] The network parameters of the feeding prediction model are initialized using a chaotic sequence strategy, resulting in individual network parameters. Depending on the data type of the feeding prediction model, these individual network parameters can be either vector parameters or matrix parameters.

[0100] Repeatedly acquire multiple network parameter individuals.

[0101] Using the historical food intake and historical weight over N consecutive days in the training data as input data for the food intake prediction model, and the food intake on day N+1 as the expected output data, the fitness of individual network parameters is obtained.

[0102] In this embodiment, the fitness is obtained through a fitness function, which is the negative of the error function of the feeding prediction model.

[0103] For example, the fitness function can be:

[0104]

[0105] Among them, F i Let i = 1, 2, ..., I, where I represents the total number of individual parameters, p = 1, 2, ..., P, where P represents the total number of training data, and k = 1, 2, ..., K, where K represents the total number of outputs of the feeding prediction model, i.e., the total number of neurons in the output layer. y' represents the k-th actual output when the p-th input data is input. pk This represents the p-th expected output data.

[0106] The network parameter individual with the highest fitness is taken as the global optimum, and a global search is performed on all network parameter individuals based on the global optimum to obtain the first updated value of the network parameter individuals.

[0107] Based on the first updated value of the individual network parameter, a further local search is performed to obtain the second updated value of the individual network parameter.

[0108] Based on the first and second update values ​​of the individual network parameters, the individual network parameters are updated to obtain the updated individual network parameters.

[0109] Determine if there exists an individual with updated network parameters whose fitness is greater than a set threshold or whose current update count t is greater than the maximum update count T. If so, use the individual with the highest fitness as the final network parameter of the feeding prediction model to obtain the trained feeding prediction model; otherwise, proceed to the next training iteration.

[0110] In this embodiment, a global search is performed on all individual network parameters based on the global optimum, resulting in the first updated value for each individual network parameter:

[0111] f i =f min +(f max -f min )*β

[0112]

[0113] ω=t / T

[0114]

[0115] Among them, f i f represents the update frequency of the i-th individual network parameter. max f represents the upper limit of the update frequency. min This represents the lower limit of the update frequency, and β represents a random number between (0,1). Let represent the update step size of the i-th network parameter individual during the t-th training iteration, and Update after handling out-of-bounds errors Let ω represent the update step size of the i-th network parameter individual during the (t-1)-th training iteration, and let ω represent the inertial weights. This represents the i-th network parameter during the (t-1)-th training iteration. Let represent the global optimum value at the (t-1)th training iteration, and T represent the pre-set maximum number of updates. Indicates the updated That is, the first updated value of the individual network parameter.

[0116] By introducing inertial weights, the search area is updated, which enhances the overall algorithm's optimization ability. Furthermore, combining it with local search can further meet the search accuracy requirements.

[0117] In this embodiment, based on the first updated value of an individual network parameter, a further local search is performed to obtain the second updated value of that individual network parameter:

[0118]

[0119] in, Indicates the updated This refers to the second updated value of an individual network parameter. ε represents the update coefficient, and ε is a constant between [-1, 1]. A t-1 The value represents the average loudness of all individual network parameters during this training, and rand represents a random number between (0,1). i t-1 This represents the pulse emission frequency corresponding to the i-th network parameter individual during the (t-1)-th training iteration.

[0120] In this embodiment, based on the first and second update values ​​of the individual network parameters, it is determined that the individual network parameters will be updated, resulting in the updated individual network parameters as follows:

[0121]

[0122]

[0123]

[0124] Among them, W i t This represents the i-th updated network parameter. This represents the loudness of the i-th network parameter during the (t-1)-th training iteration. Represents individual network parameters fitness value, This represents the maximum fitness value corresponding to an individual in the network parameters. Indicates the updated The initial values ​​corresponding to the i-th network parameter individual The value is 0.95, where α represents the attenuation coefficient between (0,1), and r... i t Indicates the updated r i t-1 γ is greater than 0, where γ represents the increment coefficient, r i 0 Let r represent the initial pulse emission frequency of the i-th network parameter individual, and r i 0 =0.9.

[0125] After training, gradient descent can be used to further optimize the parameter with the highest fitness value locally, effectively ensuring that the global optimum is reached. For example, when the current update count t is greater than the maximum update count T, gradient descent can be used for local optimization.

[0126] 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.

[0127] 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 precision feeding system for dairy cattle farming, characterized in that, The automatic feeder, the trough arranged below the outlet of the automatic feeder, the first pressure sensor arranged below the trough, the feeding cage for carrying the dairy cow, the second pressure sensor arranged below the feeding cage, and the intelligent controller are included. The automatic feeder is configured to carry the feed and receive the control of the intelligent controller, and to drop the carried feed into the trough below the outlet of the automatic feeder. The trough is arranged in the feeding cage for carrying the dairy cow, and the trough is configured to receive the feed dropped by the automatic feeder. The first pressure sensor is configured to sense the first total weight corresponding to the trough carrying the feed in real time, and to feed back the first total weight to the intelligent controller. The feeding cage is configured to carry the trough and the dairy cow, and to allow only one dairy cow to enter, so as to separate the dairy cows. The second pressure sensor is configured to sense the second total weight of the trough and the dairy cow in real time, and to feed back the second total weight to the intelligent controller. The intelligent controller is configured to sense the RFID ear tag worn by the dairy cow, to determine the unique RFID information corresponding to the dairy cow, to store the historical feeding data of each dairy cow based on the unique RFID information of the dairy cow, and to obtain the feed dropping amount based on the historical feeding data, the first total weight, and the second total weight. The intelligent controller is further configured to control the automatic feeder to drop the feed of the corresponding weight into the trough according to the feed dropping amount. The unique RFID information of the dairy cow is used as a basis to store the historical feeding data of each dairy cow, and the feed dropping amount is obtained based on the historical feeding data, the first total weight, and the second total weight, including: The unique RFID information of each dairy cow is used as a basis to store the historical feeding time period of each dairy cow and the corresponding feeding amount of each historical feeding time period on a daily basis, to obtain the historical feeding data of the dairy cow. The RFID ear tag of the dairy cow in the current feeding cage is sensed to obtain the target RFID information corresponding to the target dairy cow that is feeding and the current time period. The target weight corresponding to the target dairy cow that is feeding is determined based on the second total weight, and the target weight interval to which the target weight belongs is determined, so as to determine the feed dropping weight interval corresponding to the target weight interval, and the target weight interval is preset data, and each target weight interval is preset with a feed dropping weight interval. The feeding amount of the historical time period corresponding to the current time period in the historical feeding data of the target dairy cow is determined based on the current time period, and the feeding amount of the historical time period in N consecutive days and the historical weight are used as input data, and the feeding amount of the N+1 day is used as expected output data, to obtain multiple training data of the feeding prediction model; wherein the historical weight is collected together with the historical feeding data of the dairy cow. The feeding prediction model is trained based on the multiple training data to obtain the trained feeding prediction model. The feeding amount of the historical time period and the historical weight in the previous N consecutive days are obtained based on the current time period, and are used as input data of the trained feeding prediction model to obtain the feeding prediction amount of the target dairy cow. judging whether the eating prediction quantity of the target dairy cow is located in the corresponding feed delivery weight interval, if yes, obtaining the feed delivery quantity according to the eating prediction quantity of the target dairy cow, otherwise, correcting it to the nearest boundary value of the feed delivery weight interval; obtaining the weight of the remaining feed in the trough according to the current first total weight, and obtaining the feed delivery quantity by subtracting the weight of the remaining feed in the trough from the eating prediction quantity; training the eating prediction model based on multiple training data to obtain the trained eating prediction model, including: initializing the network parameters of the eating prediction model by using a chaotic sequence strategy to obtain a network parameter individual; wherein, according to the data type corresponding to the eating prediction model, the network parameter individual is a vector parameter or a matrix parameter; repeating the obtaining of multiple network parameter individuals; taking the eating quantity and the historical body weight of the historical time period in the N consecutive days in the training data as the input data corresponding to the eating prediction model, and taking the eating quantity of the N+1 day as the expected output data, to obtain the fitness of the network parameter individual; taking the network parameter individual with the maximum fitness as the global optimal value, and performing global search on all network parameter individuals based on the global optimal value to obtain the first updated value of the network parameter individual; further performing local search based on the first updated value of the network parameter individual to obtain the second updated value of the network parameter individual; determining to update the network parameter individual according to the first updated value and the second updated value of the network parameter individual to obtain the updated network parameter individual; judging whether the fitness of the updated network parameter individual is greater than a set threshold or the current update number t is greater than the maximum update number T, if yes, taking the network parameter individual with the maximum fitness as the final network parameter of the eating prediction model to obtain the trained eating prediction model, otherwise, entering the next training.

2. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, The method for obtaining the weight of the feed in the trough is: when the feed is not delivered in the trough, obtaining a first sensing value of a first pressure sensor; when the feed is delivered in the trough, obtaining a second sensing value of the first pressure sensor, and obtaining the weight of the delivered feed by subtracting the first sensing value from the second sensing value.

3. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, The method for obtaining the weight of the dairy cow in the feeding cage is: when the dairy cow does not exist in the feeding cage, obtaining a third sensing value of a second pressure sensor; when the dairy cow exists in the feeding cage, obtaining a fourth sensing value of the second pressure sensor, and obtaining the weight of the dairy cow by subtracting the third sensing value from the fourth sensing value.

4. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, The eating prediction model is set as a classification neural network with vector data or matrix data as input, and the input data is converted into the data type corresponding to the eating prediction model before being input into the eating prediction model.

5. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, The fitness is obtained by a fitness function, and the fitness function is the negative value of the error function of the eating prediction model.

6. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, The global search on all network parameter individuals based on the global optimal value to obtain the first updated value of the network parameter individual is: wherein, denotes the update frequency of the i-th network parameter individual, denotes the upper limit of the update frequency, denotes the lower limit of the update frequency, denotes a random number between (0, 1); denotes the update step size of the i-th network parameter individual at the t-th training, and after the boundary processing and then updating ; denotes the update step size of the i-th network parameter individual at the t-1-th training, denotes the inertia weight, denotes the i-th network parameter individual at the t-1-th training, denotes the global optimal value at the t-1-th training, denotes the maximum number of updates preset; denotes the updated , i.e. the first update value of the network parameter individual.

7. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, The further local search based on the first updated value of the network parameter individual to obtain the second updated value of the network parameter individual is: wherein, denotes the updated i.e. the second updated value of the network parameter individual; denotes an update coefficient, and is a constant between [-1, 1], denotes the average loudness of all network parameter individuals in this training, denotes a random number between (0, 1), denotes the pulse emission frequency corresponding to the i-th network parameter individual at the t-1 training.

8. The precision feeding system suitable for dairy farming as claimed in claim 1 wherein, According to the first update value and the second update value of the network parameter individual, it is determined that the network parameter individual is updated, and the updated network parameter individual is obtained as: wherein, denotes the i-th updated network parameter, denotes the loudness of the i-th network parameter individual at the time of the t-1 training, denotes the fitness value of the network parameter individual , denotes the maximum fitness value corresponding to the network parameter individual, denotes the updated , the initial of the i-th network parameter individual is 0.95, denotes the decay coefficient between (0, 1), denotes the updated , is greater than 0, denotes the increase coefficient, denotes the initial pulse emission frequency of the i-th network parameter individual, and .

Images