Automatic feeding system for industrial aquaculture

Abstract

The invention discloses an automatic feeding system for industrial aquaculture, and relates to the technical field of feed putting, and the automatic feeding system is characterized in that a data collection module collects aquaculture data of an aquaculture area, sends the aquaculture data to a feed proportion prediction module through a main control module for feed proportion prediction, and sends a prediction result to an integrated feed tower; each sub-feed tower is provided with a weight sensor, feed output of different sub-feed towers is carried out according to a prediction result, the feed distribution module carries out feed mixing synchronously, and the mixed feed is conveyed to the terminal feeding module through the pipeline module; the terminal feeding module is provided with a gravity sensor; the terminal feeding and air conveying module receives a feeding instruction, the feeding instruction is blown to the corresponding breeding unit through the air conveying module, and feeding is stopped after the preset feeding weight is reached. According to the invention, component proportioning and standardized feeding of the feed are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of feed delivery, and more particularly to an automatic feeding system for factory farming. Background Art

[0002] my country is currently the world's largest aquaculture country, with aquaculture accounting for about 70% of the world's total aquaculture. Aquatic feed is a feed specially provided for aquatic animal farming. According to the feeding species, aquatic feed can be divided into fish feed, shrimp feed and crab feed; according to the characteristics of the feed, it can be divided into compound feed, concentrated feed and premixed feed. The raw materials for aquatic feed production are mainly composed of fish meal, grain raw materials and oils, and fish meal and grain raw materials often account for more than 50% of the feed cost.

[0003] Fish need to eat fish food to grow up. Among the various production inputs of aquaculture, feed input costs are the largest. Surveys show that the feed input costs of general aquaculture products account for more than 70% of the total cost of aquaculture. Generally, fish food is thrown into the fish pond manually for fish to eat. When fish food is thrown into the fish pond manually, it is time-consuming and labor-intensive, and the feeding efficiency is low. In addition, the breeding techniques of different breeders are different. It is difficult to achieve standardized feeding based on sensory experience alone, which will have an adverse effect on fish growth, thereby reducing the yield per unit area. At the same time, high-density intensive farming is an important feature of current aquaculture. High-density farming increases the amount of feed, increases the amount of breeding garbage in the water body, and thus deteriorates the water quality. It is manifested as a decrease in dissolved oxygen in the water, and excessive ammonia nitrogen and nitrite content. These factors will cause oxidative damage to fish, thereby reducing the digestion and absorption of feed by fish, reducing the growth and immune function of fish, and increasing the morbidity and mortality of fish. How can these hazards be minimized while maintaining high-density farming conditions?

[0004] Therefore, how to proportion the ingredients of feed and standardize the feed delivery is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] In view of this, the present invention provides an automatic feeding system for factory farming, which can predict the composition ratio of feed in the time dimension and data dimension, provide accurate feed nutritional ratio for aquaculture, and standardize the feed delivery through air delivery.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] An automatic feeding system for factory farming includes: a data collection module, a main control module, an integrated feed tower, a weight sensor, a feed ratio prediction module, a wind delivery module, a pipeline module, a material distribution module, and a terminal feeding module. The data collection module collects farming data of the farming area and sends it to the feed ratio prediction module through the main control module. The feed ratio prediction module performs feed ratio prediction and sends the prediction result to the integrated feed tower through the main control module. The integrated feed tower is provided with a plurality of sub-feed towers, each of which is equipped with different feeds. The prediction results are sent to the sub-feed towers respectively. The sub-feed towers are all provided with weight sensors. The prediction results are made by using different feed ratios in the prediction results. The output weight of feeds in different sub-towers is controlled by the weight sensor; the material distribution module synchronously receives the prediction result, performs feed mixing operation after receiving the prediction result, and transports the mixed feed to the terminal feeding module through the pipeline module; the terminal feeding module is provided with a gravity sensor; when the main control module receives a feeding instruction, it is synchronously sent to the terminal feeding module and the air delivery module, and the air delivery module blows the mixed feed pre-stored in the terminal feeding module to the corresponding breeding unit, and the weight of the mixed feed is sent to the main control module through the gravity sensor provided in the terminal feeding module, and the feeding is stopped after the preset feeding weight is reached.

[0008] Preferably, the data collection module includes: a water body data collection unit responsible for collecting water body data; an aquatic product data collection unit responsible for collecting aquatic product data; and the breeding data includes aquatic product data and water body data.

[0009] Preferably, the aquatic data collection unit comprises:

[0010] The sampling unit is responsible for collecting aquatic samples at fixed intervals;

[0011] A sample calculation unit is used to obtain the mean, standard deviation and empirical cumulative distribution function of the aquatic sample; an iterative method is used to obtain an estimated value of a normal distribution parameter, and the normal distribution parameter is compared with a preset threshold value, and a sample with a minimum normal distribution parameter among samples that meet the preset threshold value and are greater than the normal distribution parameter is obtained as the best sample;

[0012] The growth assessment unit uses a pre-trained growth assessment model to perform assessment based on the best sample in each fixed time period to obtain a growth rate assessment value.

[0013] Preferably, the sample calculation unit specifically includes: obtaining the mean, standard deviation and empirical cumulative distribution function of the aquatic samples; obtaining estimated values ​​of normal distribution parameters using an iterative method based on the mean, standard deviation and distribution function; comparing the normal distribution parameters with a preset threshold, and obtaining the sample with the minimum normal distribution parameter among the samples that meet the preset threshold and are greater than the normal distribution parameter as the best sample.

[0014] Preferably, the feed ratio prediction module is trained by historical data, and the historical data is historical breeding data containing feed ratio variable parameters, specifically including:

[0015] In the deep learning module, the historical water body data is the variable parameters directly obtained by the detection of each sensor at the same time. The historical water body data is used as input, and the feed ratio variable parameters are used as output. The modules are input into the deep convolutional neural network for processing, and the relationship between different variable parameters at the same time is learned, and the relationship between the historical feed ratio variable parameters and the historical breeding data is obtained;

[0016] The gated learning module sets the time step, performs sliding window sampling on the historical aquatic data according to the time step, inputs the sampled data into the deep gated recurrent neural network for single-step prediction, learns the relationship between the historical feed ratio variable parameters at different times, and obtains the relationship between the historical feed ratio variable parameters at multiple times and the current time data;

[0017] The multi-feature fusion module inputs the historical output prediction data of the deep convolutional neural network before the last fully connected layer and the historical output prediction data of the deep gated recurrent neural network before the last fully connected layer into the multi-feature fusion layer for fusion, and then inputs the fusion features output by the multi-feature fusion layer into the last fully connected layer, trains the deep convolutional neural network and the deep gated recurrent network using the back propagation algorithm, adjusts the weight parameters of the deep convolutional neural network and the deep gated recurrent network, and optimizes the trained deep convolutional neural network and the deep gated recurrent network;

[0018] The comprehensive prediction module uses the trained deep convolutional neural network and deep gated recurrent network combined with a multi-feature fusion layer to form a trained multi-feature fusion deep neural network. The breeding data to be tested is input into the trained multi-feature fusion deep neural network and then the output prediction is used to obtain the feed ratio variable parameters.

[0019] Preferably, the deep convolutional neural network in the deep learning module captures data dimension feature data of multivariate time series data, and the historical output prediction data of the deep convolutional neural network before the last fully connected layer is the data dimension feature data; the deep gated recurrent network captures time dimension feature data of the target data, and the historical output prediction data of the deep gated recurrent neural network before the last fully connected layer is the time dimension feature data.

[0020] Preferably, the deep gated recurrent neural network in the gated learning module is composed of three gated recurrent layers connected in sequence to form a three-layer gated recurrent network, including three layers of processing modules: the first processing module includes a first gated recurrent layer, which is sequentially transmitted to the second processing module via a reset gate and an update gate; the second processing module includes a second gated recurrent layer and a second Dropout layer, which is sequentially transmitted to the third processing module via a reset gate and an update gate; the third processing module includes a third gated recurrent layer, which is sequentially transmitted to the mapping layer via a reset gate and an update gate; the first gated recurrent layer, the second gated recurrent layer and the third gated recurrent layer all include gated recurrent units.

[0021] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses an automatic feeding system for factory farming. Through data collection and analysis, the system can predict and optimize the feed ratio in real time to maximize the farming effect and improve the farming efficiency; the application of deep learning and data fusion technology improves the intelligent analysis ability of farming data and can better adapt to changes in the farming environment; combined with the growth evaluation of aquatic and water data, farming decisions are more scientific to achieve better farming effects and benefits; the feed ratio is predicted in the time dimension and the data dimension, providing accurate feed nutrient ratios for aquaculture, and at the same time, standardized feed delivery is carried out through air delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0023] Figure 1 A schematic diagram of the system structure provided by the present invention;

[0024] Figure 2 This is a schematic diagram of the structure of the feed ratio prediction module provided by the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The embodiment of the present invention discloses an automatic feeding system for factory farming, such as Figure 1 As shown, it includes: a data collection module, a main control module, an integrated material tower, a weight sensor, a feed ratio prediction module, a wind conveying module, a pipeline module, a material distribution module, and a terminal feeding module. The data collection module collects the breeding data of the breeding area and sends it to the feed ratio prediction module through the main control module. The feed ratio prediction module performs feed ratio prediction and sends the prediction result to the integrated material tower through the main control module. The integrated material tower is provided with a plurality of sub-material towers, each of which is equipped with different feeds. The prediction results are sent to the sub-material towers respectively. The sub-material towers are all provided with weight sensors. Different feed ratios in the prediction results are predicted. The output weight of the feed in the sub-tower is controlled by a weight sensor; the feeding module receives the prediction results synchronously, performs feed mixing operations after receiving the prediction results, and transports the mixed feed to the terminal feeding module through the pipeline module; the terminal feeding module is provided with a gravity sensor; when the main control module receives the feeding instruction, it is sent to the terminal feeding module and the air delivery module synchronously, and the air delivery module blows the mixed feed pre-stored in the terminal feeding module to the corresponding breeding unit, and the weight of the mixed feed is sent to the main control module through the gravity sensor set in the terminal feeding module, and the feeding is stopped after reaching the preset feeding weight.

[0027] In a specific embodiment, the data collection module includes: a water body data collection unit responsible for collecting water body data; an aquatic product data collection unit responsible for collecting aquatic product data; and aquaculture data includes aquatic product data and water body data.

[0028] In a specific embodiment, water body data include water temperature, dissolved oxygen, pH value, ammonia nitrogen, nitrite, etc., aquatic product data include growth rate, mortality rate, disease rate, etc.; feed ingredients include fish meal, soy protein concentrate, casein, shrimp shell meal, flour, cassava flour, soybean oil, fish oil, squid paste, vitamins, minerals, choline chloride, calcium dihydrogen phosphate, VC phosphate, etc.

[0029] In a specific embodiment, the aquatic data collection unit includes:

[0030] The sampling unit is responsible for collecting aquatic samples at fixed intervals;

[0031] Sample calculation unit, to obtain the mean μ0, standard deviation σ0 and empirical cumulative distribution function G(x) of aquatic samples;

[0032]

[0033] Among them, x i is the data set of the sample, i = 1, 2, ..., n, x1 ≤ x2 ≤ ... ≤ x n .

[0034] Iterative method to obtain normal distribution parameter estimates and Compare the normal distribution parameter with a preset threshold, and obtain the sample with the minimum normal distribution parameter among the samples that meet the preset threshold and are greater than the normal distribution parameter as the best sample;

[0035] The growth assessment unit uses a pre-trained growth assessment model to perform assessment based on the best sample in each fixed time period to obtain a growth rate assessment value.

[0036] In a specific embodiment, a scattered signal image is obtained by sampling through forward-looking sonar detection, and the length of the best sample is calculated based on the area pixel data of the scattered signal image to obtain the aquatic body length data, which specifically includes:

[0037] Feature extraction part: It contains four dilated convolution modules of different depths. The step size of the convolution module is 1, the convolution kernel size is 3, the dilated convolution coefficient is 2, and lReLU is used as the activation function. The result of convolution at each step size is shown in the following formula, where l, m, u are the filter sizes, v and o are the length and width of the input image respectively, W is the shared weight, and bias is the bias;

[0038]

[0039] O is the Leaky-ReLU activation function:

[0040] Among them, v∈[0,H in ),o∈[0,W in ),u∈[0,C in );

[0041] The output size of each layer is shown in the formula:

[0042]

[0043] Among them, H n , W n are the length and width of the feature map of the input nth layer, C n is the depth of the convolutional layer, Y n is the output of the nth convolutional layer, h is the number of padded pixels, k is the convolution kernel size, d is the dilated convolution coefficient, and s is the step size;

[0044] Attention mechanism part: mask and superimpose the feature map extracted by the convolution part through the formula to obtain a feature map with more obvious effective features;

[0045] Y O =Y+sigmoid(Y)*Y;

[0046]

[0047] Fully connected regression part: First, there is a fully connected layer with 4096 nodes. Secondly, the output of the layer containing 4096 feature vectors is evenly cut into 8 slices, and each slice is regressed for body length separately. Finally, the predicted 8 body lengths are weighted and superimposed to calculate the final body length, as shown in the formula;

[0048] Age=w1*age1+w2*age2+...+w8*age8;

[0049] Among them, Age is the final body length, (w1,w2,...,w8) is the weight coefficient, and (age1,age2...,age8) is the 8 slices obtained by averaging 4096 feature vectors.

[0050] The pre-trained growth assessment model adopts the ResNet-50 deep learning model framework and a training mechanism of training from scratch. It randomly initializes the ResNet-50 network weights and continuously updates the weights of all layers of the network in each iteration. The ResNet-50 network weights are randomly initialized by sampling from a normal distribution with a mean of 0 and a standard deviation of 0.02.

[0051] In a specific embodiment, the sample calculation unit includes:

[0052] A. Let y k =(μ k ,σ k ), the initial value of k is zero, that is, y0 = (μ0, σ0) is the initial point, a step size h>0 is selected, and the accuracy of the normal distribution parameter estimate of the sample ε>0;

[0053] B. Calculate the value of y k The normal cumulative distribution function F(x; y k ):

[0054]

[0055] Where φ represents the standard normal cumulative distribution function, LSL is the sample lower specification limit, and USL is the sample upper specification limit;

[0056] C. Calculate the normal cumulative distribution function F(x; y k) and the maximum difference between the empirical cumulative distribution function G(x) of the sample

[0057] D. Order j=1,2,3,4; calculate about The normal cumulative distribution function of

[0058]

[0059] E. Calculate the normal cumulative distribution function of the sample The maximum difference from the empirical cumulative distribution function G(x) of the sample

[0060] F. Calculation Minimum value of

[0061] G. The above obtained and For comparison: If Then k=k+1, And double the search step length h (return to D); if (Execute H);

[0062] H. Compare the search step size h with the accuracy ε of the parameter estimate: If h>ε, reduce the search step size h to the current step size. (Return D); If h≤ε, stop the iteration and get the mean estimate of the normal distribution of the batch product and standard deviation estimates

[0063] In a specific embodiment, the sample calculation unit specifically includes: obtaining the mean, standard deviation and empirical cumulative distribution function of the aquatic samples; obtaining the estimated value of the normal distribution parameter by using the iterative method according to the mean, standard deviation and distribution function; comparing the normal distribution parameter with a preset threshold, and obtaining the sample with the minimum normal distribution parameter among the samples that meet the preset threshold and are greater than the normal distribution parameter as the best sample.

[0064] In a specific embodiment, Figure 2 As shown, the feed ratio prediction module is trained through historical data, which is historical breeding data containing feed ratio variable parameters, including:

[0065] In the deep learning module, historical water body data is the variable parameters directly obtained by various sensors at the same time. The historical water body data is used as input and the feed ratio variable parameters are used as output. The modules are input into the deep convolutional neural network for processing, and the relationship between different variable parameters at the same time is learned, and the relationship between historical feed ratio variable parameters and historical breeding data is obtained.

[0066] The gated learning module sets the time step, performs sliding window sampling on the historical aquatic data according to the time step, inputs the sampled data into the deep gated recurrent neural network for single-step prediction, learns the relationship between the historical feed ratio variable parameters at different times, and obtains the relationship between the historical feed ratio variable parameters at multiple times and the current time data;

[0067] The multi-feature fusion module inputs the historical output prediction data of the deep convolutional neural network before the last fully connected layer and the historical output prediction data of the deep gated recurrent neural network before the last fully connected layer into the multi-feature fusion layer for fusion, and then inputs the fusion features output by the multi-feature fusion layer into the last fully connected layer, trains the deep convolutional neural network and the deep gated recurrent network using the back propagation algorithm, adjusts the weight parameters of the deep convolutional neural network and the deep gated recurrent network, and optimizes the trained deep convolutional neural network and the deep gated recurrent network;

[0068] The comprehensive prediction module uses the trained deep convolutional neural network and deep gated recurrent network combined with a multi-feature fusion layer to form a trained multi-feature fusion deep neural network. The breeding data to be tested is input into the trained multi-feature fusion deep neural network and then the output prediction is used to obtain the feed ratio variable parameters.

[0069] In a specific embodiment, the deep convolutional neural network in the deep learning module captures the data dimension feature data of the multivariate time series data, and the historical output prediction data of the deep convolutional neural network before the last fully connected layer is the data dimension feature data; the deep gated recurrent network captures the time dimension feature data of the target data, and the historical output prediction data of the deep gated recurrent neural network before the last fully connected layer is the time dimension feature data.

[0070] In a specific embodiment, the deep convolutional neural network includes a plurality of convolutional blocks connected in sequence and a maximum pooling layer, each convolutional block is composed of a plurality of convolutional layers connected in sequence, the time series data is first normalized, and then processed by each convolutional block of the time domain convolution, and features are obtained using an activation function and mapped to a feature space, and the calculation formula is as follows:

[0071] P p =f(Q p *P p-1 +z p);

[0072] In the formula, Q p Represents the shared weight of the p-th convolutional block and the feature P of the p-1-th convolutional block p-1 Convolution, * is the convolution operator symbol, z p is the bias vector of the i-th convolutional block, f() represents the nonlinear activation function; the feature P of the p-th convolutional block is obtained through the nonlinear activation function relu p ;

[0073] After the time-domain convolution processing of each convolution block, the output of the final maximum pooling layer is the data dimension feature data of the target feature space.

[0074] In a specific embodiment, the deep gated recurrent neural network in the gated learning module is composed of three gated recurrent layers connected in sequence to form a three-layer gated recurrent network, including three layers of processing modules: the first processing module includes a first gated recurrent layer, which is sequentially transmitted to the second processing module via a reset gate and an update gate; the second processing module includes a second gated recurrent layer and a second Dropout layer, which is sequentially transmitted to the third processing module via a reset gate and an update gate; the third processing module includes a third gated recurrent layer, which is sequentially transmitted to the mapping layer via a reset gate and an update gate; the first gated recurrent layer, the second gated recurrent layer and the third gated recurrent layer all include gated recurrent units.

[0075] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0076] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic feeding system for factory farming, characterized in that: include: A data collection module, a main control module, an integrated material tower, a weight sensor, a feed ratio prediction module, a wind delivery module, a pipeline module, a material distribution module, and a terminal feeding module. The data collection module collects the breeding data of the breeding area, and sends it to the feed ratio prediction module through the main control module. The feed ratio prediction module performs feed ratio prediction, and the prediction result is sent to the integrated material tower through the main control module. The integrated material tower is provided with a plurality of sub-material towers, each of which is equipped with different feeds. The prediction results are sent to the sub-material towers respectively. The sub-material towers are all provided with weight sensors. The feed ratios of different sub-material towers are used to predict the feed ratios of the feeds. The weight sensor controls the output weight; the material distribution module synchronously receives the prediction result, performs feed mixing operation after receiving the prediction result, and transports the mixed feed to the terminal feeding module through the pipeline module; the terminal feeding module is provided with a gravity sensor; when the main control module receives the feeding instruction, it is synchronously sent to the terminal feeding module and the air delivery module, the air delivery module blows the mixed feed pre-stored in the terminal feeding module to the corresponding breeding unit, and the weight of the mixed feed is sent to the main control module through the gravity sensor provided in the terminal feeding module, and the feeding is stopped after the preset feeding weight is reached.

2. The automatic feeding system for factory farming according to claim 1 is characterized in that: The data collection module includes: a water body data collection unit, which is responsible for collecting water body data; an aquatic product data collection unit, which is responsible for collecting aquatic product data; and the breeding data includes aquatic product data and water body data.

3. The automatic feeding system for factory farming according to claim 2 is characterized in that: The aquatic data collection unit comprises: The sampling unit is responsible for collecting aquatic samples at fixed intervals; A sample calculation unit is used to obtain the mean, standard deviation and empirical cumulative distribution function of the aquatic sample; an iterative method is used to obtain an estimated value of a normal distribution parameter, and the normal distribution parameter is compared with a preset threshold value, and a sample with a minimum normal distribution parameter among samples that meet the preset threshold value and are greater than the normal distribution parameter is obtained as the best sample; The growth assessment unit uses a pre-trained growth assessment model to perform assessment based on the best sample in each fixed time period to obtain a growth rate assessment value.

4. The automatic feeding system for factory farming according to claim 3 is characterized in that: The sample calculation unit specifically includes: obtaining the mean, standard deviation and empirical cumulative distribution function of the aquatic sample; obtaining the estimated value of the normal distribution parameter by using the iterative method according to the mean, standard deviation and distribution function; comparing the normal distribution parameter with a preset threshold, and obtaining the sample with the minimum normal distribution parameter among the samples that meet the preset threshold and are greater than the normal distribution parameter as the best sample.

5. The automatic feeding system for factory farming according to claim 2 is characterized in that: The feed ratio prediction module is trained by historical data, which is historical breeding data containing feed ratio variable parameters, specifically including: In the deep learning module, the historical water body data is the variable parameters directly obtained by the detection of each sensor at the same time. The historical water body data is used as input, and the feed ratio variable parameters are used as output. The modules are input into the deep convolutional neural network for processing, and the relationship between different variable parameters at the same time is learned, and the relationship between the historical feed ratio variable parameters and the historical breeding data is obtained; The gated learning module sets the time step, performs sliding window sampling on the historical aquatic data according to the time step, inputs the sampled data into the deep gated recurrent neural network for single-step prediction, learns the relationship between the historical feed ratio variable parameters at different times, and obtains the relationship between the historical feed ratio variable parameters at multiple times and the current time data; The multi-feature fusion module inputs the historical output prediction data of the deep convolutional neural network before the last fully connected layer and the historical output prediction data of the deep gated recurrent neural network before the last fully connected layer into the multi-feature fusion layer for fusion, and then inputs the fusion features output by the multi-feature fusion layer into the last fully connected layer, trains the deep convolutional neural network and the deep gated recurrent network using the back propagation algorithm, adjusts the weight parameters of the deep convolutional neural network and the deep gated recurrent network, and optimizes the trained deep convolutional neural network and the deep gated recurrent network; The comprehensive prediction module uses the trained deep convolutional neural network and deep gated recurrent network combined with a multi-feature fusion layer to form a trained multi-feature fusion deep neural network. The breeding data to be tested is input into the trained multi-feature fusion deep neural network and then the output prediction is used to obtain the feed ratio variable parameters.

6. The automatic feeding system for factory farming according to claim 5, characterized in that: The deep convolutional neural network in the deep learning module captures the data dimension feature data of the multivariate time series data, and the historical output prediction data of the deep convolutional neural network before the last fully connected layer is the data dimension feature data; the deep gated recurrent network captures the time dimension feature data of the target data, and the historical output prediction data of the deep gated recurrent neural network before the last fully connected layer is the time dimension feature data.

7. The automatic feeding system for factory farming according to claim 5, characterized in that: The deep gated recurrent neural network in the gated learning module is composed of three gated recurrent layers connected in sequence to form a three-layer gated recurrent network, including three layers of processing modules: the first processing module includes a first gated recurrent layer, and the first gated recurrent layer is sequentially transmitted to the second processing module via a reset gate and an update gate; the second processing module includes a second gated recurrent layer and a second Dropout layer, and the second gated recurrent layer is sequentially transmitted to the third processing module via a reset gate and an update gate; the third processing module includes a third gated recurrent layer, and the third gated recurrent layer is sequentially transmitted to the mapping layer via a reset gate and an update gate; the first gated recurrent layer, the second gated recurrent layer and the third gated recurrent layer all include gated recurrent units.

Images