Method for establishing intelligent model for dynamic deposition rule of chicken body components of Snow Mountain

Abstract

The invention discloses a method for establishing an intelligent model of a dynamic deposition rule of body components of Xueshan hens, and relates to the technical field of animal breeding. By adopting the method, the feed efficiency can be improved, and the accurate growth model can clearly present the rules of weight increase and body composition deposition of the snow mountain hens in different growth stages. Through the prediction of the growth model, the feed nutrition supply can be planned in advance, and the nutrition demand amounts at different feeding stages are refined, so that more accurate formula preparation is guided. Through accurate growth model prediction, feed waste or shortage can be avoided, each chicken can obtain sufficient nutrition in a proper stage, the growth speed is increased, the breeding period is shortened, the slaughtering amount in unit time is increased, the breeding efficiency is remarkably improved, and the breeding cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of animal breeding technology, and more specifically to a method for establishing an intelligent model of the dynamic deposition law of body components of snow mountain chickens. Background Art

[0002] Yellow-feathered broilers are a local breed in my country, with an annual output of about 4 billion chickens, accounting for about 40% of the total output of broilers. There are many varieties of yellow-feathered broilers, and the growth rate and age of different varieties vary greatly. Snow Mountain Chicken is a slow-growing yellow-feathered broiler system independently bred by Jiangsu Lihua Food Group Co., Ltd. It has the characteristics of excellent meat quality, high uniformity, and suitable for slaughter and cold fresh. However, the growth curve and nutrient deposition law of Snow Mountain Chicken are still unclear, which seriously restricts the precise feed formulation and large-scale breeding of this variety.

[0003] The growth and development law of broilers can be represented by an "S" curve, which has three stages: uniform growth period, accelerated growth period and stable period. The growth curve of poultry weight or body composition can be fitted by mathematical formula or artificial intelligence (LSTM model) to construct a potential growth model for poultry. The growth model can predict the growth of poultry weight and body composition, and dynamically predict the daily nutrient requirements of poultry. Currently, the commonly used growth curve models include Logistic, Gompertz and VonBertalanffy. In recent years, reports on poultry growth models have mainly focused on fast-growing white-feathered broilers and slow-growing Qingyuan Ma chickens and Wenchang chickens. There are no reports on the growth and development law model of slow-growing snow mountain chickens based on Logistic, Gompertz, Von Bertalanffy and artificial intelligence fitting (LSTM model). Summary of the invention

[0004] In view of this, the present invention provides a method for establishing an intelligent model of the dynamic deposition law of the body components of Snow Mountain Chicken.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for establishing an intelligent model of dynamic deposition law of body components of snow mountain chicken, comprising:

[0007] (1) Under the normal floor rearing mode, 720 one-day-old Snow Mountain grass cocks and hens with similar body weights were selected and randomly divided into 6 replicate pens, with 120 in each pen. Each chicken was equipped with a wing tag, and the chickens were allowed to feed and drink freely. The experimental measurement period was from 1 to 98 days old (d). Among them, at 1, 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84, 91, and 98 d, each chicken was weighed; at 1, 14, 28, 42, 56, 70, 84, and 98 d, one chicken was randomly selected from each replicate pen for slaughter sampling and was euthanized by asphyxiation.

[0008] (2) Throughout the experiment, a corn-soybean meal-based basal diet was fed, and the experimental management measures followed the regulations for slow-growing chickens.

[0009] (3) The conventional nutritional indexes, production performance, and nutrient deposition indexes of the diet were detected.

[0010] (4) The experimental data were statistically analyzed using R language 4.2.0 software. Two-way analysis of variance (gender and age) and Tukey multiple comparisons were performed. The statistical significance level was P < 0.05, and the experimental data of each group were expressed as means ± SEM; the NLS nonlinear model was used to fit the models for the age and body weight, energy deposition, protein deposition, fat deposition, ash deposition, calcium deposition, and phosphorus deposition of Snow Mountain grass chickens respectively. The Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model were used for nonlinear fitting, and the fitting effects of various models were compared.

[0011] Logistic model,

[0012] Gompertz model,

[0013] Von Bertalanffy model, y(t) = A(1 - Be -kt ) 3 ;

[0014] Polynomial model, y(t) = a0 + a1t + a2t 2 +... + a n t n ;

[0015] LSTM model, f t = σ(W f ·[h t-1 , x t + b f )

[0016] it = σ(W i · [h t-1 , x t + b i );

[0017]

[0018] o t = σ(W o · [h t-1 , x t + b o );

[0019] h t = o t ⊙ tanh(C t );

[0020] Where: y is the nutrient deposition of broilers when the age reaches x; a is the asymptote of the curve (the maximum deposition at maturity); b is the maximum relative growth rate; c is the inflection point age when the growth rate reaches the maximum; t is the age; k is the growth rate parameter;

[0021] h t : The output prediction at time point t, which can represent the predicted values of growth indicators such as animal weight and body fat percentage;

[0022] f t : The forget gate, which determines how much information of the previous growth state should be retained, for example, determines the influence degree of the past growth trend on the current prediction;

[0023] i t : The input gate, which determines how much new information should be accepted, for example, the influence intensity of current environmental conditions or changes in feeding management;

[0024] o t : The output gate, which controls which currently calculated growth information should be output as the prediction result;

[0025] : The candidate growth state, which contains new information that may affect the current growth;

[0026] C t : The cell state, which represents the long-term memory of animal growth and stores the cumulative information of the entire growth cycle;

[0027] C t-1 : The growth state at the previous time step, which represents the previous growth history;

[0028] W f ,W i ,W C ,W o: The weight matrix of each gating unit determines the influence degree of different input features on growth prediction;

[0029] b f , b i , b C , b o : The bias term is used to adjust the baseline prediction value of the model;

[0030] (5) After the model is fitted, load the R language modelr package and use indicators such as the coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE), Akaike information criterion (AIC), and Bayesian information criterion (BIC) to comprehensively evaluate the fitting of the Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model to the data of the dynamic deposition laws of body energy, body protein, body fat, body ash, and calcium and phosphorus.

[0031] In the present invention, R 2 (coefficient of determination): It represents the proportion of the total variation in the observations explained by the model, and the calculation formula (RSS is the residual sum of squares, and TSS is the total sum of squares). R 2 ranges from 0 to 1, and the closer it is to 1, the better the model fits the data. R 2 emphasizes the explanatory power for the observed data, but does not penalize the model complexity - adding more parameters can almost always improve R 2 , even if these parameters are only fitting noise. Therefore, R 2 is more used to describe the fitting degree rather than a comprehensive comparison of the model quality.

[0032] RMSE (root mean square error) and MAE (mean absolute error): These two indicators directly measure the magnitude of the deviation between the predicted value and the observed value. RMSE is the square root of the mean of the squared residuals, and the formula is

[0033] MAE is the average of the absolute values of the residuals RMSE is more sensitive to large errors (because the square term amplifies large deviations), while MAE equally weights and averages each error. The smaller the two, the better, and the unit is the same as the original data, which is convenient for explaining the average error level of the model. Like R 2 , RMSE and MAE only focus on the fitting residuals and do not directly consider the model complexity. Therefore, a more complex model usually does not have a higher RMSE / MAE than a simple model (unless overfitting occurs and the prediction becomes discrete). On the training set, increasing the parameters usually can reduce RMSE / MAE.

[0034] AIC (Akaike Information Criterion): An index for evaluating models from the perspective of information theory, and its formula is AIC = 2k - 2ln(L), where k is the number of model parameters and L is the maximum log-likelihood of the model. Intuitively, -2ln(L) measures the degree to which the model does not fit the data (similar to RSS), and 2k is the penalty for the model complexity. The smaller the AIC value, the better the model performs after balancing the goodness of fit and complexity. AIC focuses on the estimation of predictive performance - it attempts to select the model that performs best on new data. Therefore, simply put, AIC rewards and punishes simultaneously: an improvement in the goodness of fit of the model will reduce -2ln(L), making AIC decrease, but increasing the number of parameters will increase 2k, making AIC increase. The best model is usually the one that can explain the data without being overly complex. It should be noted that AIC is applicable to comparisons with a fixed sample, and AIC values between different data sets cannot be directly compared.

[0035] BIC (Bayesian Information Criterion): Similar to AIC, it is also a model selection criterion, and its formula is BIC = -2ln(L) + kln(n). The penalty coefficient in front of k in BIC is ln(n). When the sample size n is large, ln(n) is larger than 2, so BIC punishes the number of parameters more strongly than AIC. This reflects the Bayesian idea of BIC: a larger data set provides more accurate model discrimination, so more complex models need to be punished more strictly. The smaller the BIC value, the better. Different from AIC, BIC will tend to select the true model (if the true model is among the candidates) in the large-sample limit, while AIC tends to select the model with the smallest prediction error, even if it has redundant parameters. Generally speaking, BIC emphasizes model simplicity more, so it often "punishes" more severely than AIC when there are more parameters.

[0036] Focus of index evaluation: R 2 , RMSE, and MAE mainly focus on the fitting accuracy of the model to existing data and do not directly consider the model complexity; AIC and BIC, on the other hand, attempt to balance the goodness of fit and simplicity and focus on the generalization ability of the model (i.e., the prediction reliability for new data). These two types of indices often conflict: the higher the goodness of fit of a model (high R 2 and low RMSE), usually means that it uses more degrees of freedom, which may lead to a higher (worse) AIC / BIC. Therefore, when selecting a model, it is necessary to consider comprehensively - only pursuing the highest R 2 may result in selecting an overfitting model, while strictly following the minimum BIC may sacrifice some fitting accuracy in exchange for simplicity. This trade-off is essentially a bias-variance trade-off: complex models have high variance (fit the training data extremely well but are unstable for new data), and simple models have high bias (fit the training data poorly but are more robust).

[0037] Preferably, the production performance is detected by the following method:

[0038] Weigh each chicken individually and weigh them repeatedly in the morning (they have been fasting for more than 8 hours, and the fasting time can be appropriately shortened for younger birds). Record the feed consumption, number of dead birds, and weight of dead birds at each stage. Calculate the average daily weight gain, average feed intake, and feed-to-weight ratio for each stage and the entire period.

[0039] Preferably, nutrient deposition is detected by the following method:

[0040] The chickens were weighed and then asphyxiated to death, the contents of the digestive tract were removed, and the fasting weight was recorded. The chickens were weighed and then asphyxiated to death, the contents of the digestive tract were removed, and the fasting weight was recorded. The experimental chickens were initially crushed with a poultry grinder (Model 160, customized by Yusheng Machinery Manufacturing Factory, Xingtai, Hebei, China), and then weighed after a second crushing with a bone mud grinder (Model GN-200, customized by Yusheng Machinery Manufacturing Factory, Xingtai, Hebei, China). Then, the flattened samples were sterilized in an oven at 105°C for 15 minutes, and then weighed after baking at 65°C for 48 hours. The samples that were dried into hard blocks were then finely crushed with a wall-breaking grinder. Finally, air-dried samples were obtained after 24 hours of moisture resorption. Two independent samples were taken by the two quartering methods and stored in a -20°C low-temperature refrigerator. Moisture, ash, calcium, phosphorus, crude protein, crude fat and gross energy were to be tested;

[0041] Nutrient deposition = body weight × carcass nutrient content.

[0042] It can be seen from the above technical solution that compared with the prior art, the present invention has the following beneficial effects:

[0043] 1) It can improve feed efficiency. The accurate growth model can clearly present the laws of weight gain and body composition deposition of Snow Mountain Chicken at different growth stages. Through the prediction of the growth model, the feed nutrition supply can be planned in advance, and the nutritional requirements at different feeding stages can be refined to guide more accurate formula preparation. At present, the nutritional requirements in meat poultry farming are mostly evaluated on a monthly basis. The nutritional requirements are only an average value, which can easily lead to overfeeding or malnutrition, affecting the biological efficiency and economic benefits of farming. Through accurate growth model prediction, feed waste or shortage can be avoided, so that each chicken can obtain sufficient nutrition at the appropriate stage, accelerate growth, shorten the breeding cycle, increase the number of chickens slaughtered per unit time, significantly improve breeding efficiency, and reduce breeding costs.

[0044] 2) The model can refine feeding management and provide a scientific basis for breeders. For example, based on the prediction of the inflection point of the growth curve, breeders can timely adjust the breeding environment parameters, such as temperature, humidity, and light, to fully explore the genetic growth potential from the perspective of environmental control. During the critical growth period of Snow Mountain Chicken, by precisely controlling the environment and meeting its specific needs for temperature and light, the incidence of diseases can be effectively reduced, the overall health level of the flock can be improved, and the stability of breeding benefits can be ensured.

[0045] 3) For the industrial development, the present invention plays an important promoting role. On the one hand, the model establishment method can be extended to all links of the snow mountain partridge chicken industrial chain, from seedling cultivation to adult chicken sales. Production decisions are made based on a unified scientific model in each link, which is conducive to realizing industrial standardization and enhancing the market competitiveness of snow mountain partridge chicken products. On the other hand, the data accumulated by the model can provide reference for breeding work, help breeding experts screen out snow mountain partridge chicken varieties with more excellent traits, promote the optimization and update of varieties, and drive the entire snow mountain partridge chicken industry to develop in a high-quality and sustainable direction.

[0046] In summary, the method for establishing the growth and development law model of the snow mountain partridge chicken of the present invention has an inestimable value for improving feed efficiency, realizing refined feeding management, and promoting industrial progress, and will bring a positive impact on the snow mountain partridge chicken breeding industry. Brief Description of the Drawings

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0048] Figure 1 For the Logistic model to fit the change curve of the live weight of each rooster and hen;

[0049] Figure 2 For the Gompertz model to fit the change curve of the live weight of each rooster and hen;

[0050] Figure 3 For the polynomial model to fit the change curve of the live weight of each rooster and hen;

[0051] Figure 4 For the Von Bertalanffy model to fit the change curve of the live weight of each rooster and hen;

[0052] Figure 5 For the LSTM model to fit the change curve of the live weight of each rooster and hen;

[0053] Figure 6 For the Logistic model to fit the change curve of the body protein content of roosters and hens;

[0054] Figure 7 For the Gompertz model to fit the change curve of the body protein content of roosters and hens;

[0055] Figure 9To fit the change curves of the body protein content of roosters and hens with the Von Bertalanffy model;

[0056] Figure 8 To fit the change curves of the body protein content of roosters and hens with the polynomial model;

[0057] Figure 10 To fit the change curves of the body protein content of roosters and hens with the LSTM model;

[0058] Figure 11 To fit the change curves of the body fat content of roosters and hens with the Logistic model;

[0059] Figure 12 To fit the change curves of the body fat content of roosters and hens with the Gompertz model;

[0060] Figure 13 To fit the change curves of the body fat content of roosters and hens with the polynomial model;

[0061] Figure 14 To fit the change curves of the body fat content of roosters and hens with the Von Bertalanffy model;

[0062] Figure 15 To fit the change curves of the body fat content of roosters and hens with the LSTM model;

[0063] Figure 16 To fit the change curves of the energy of roosters and hens with the Logistic model;

[0064] Figure 17 To fit the change curves of the energy of roosters and hens with the Gompertz model;

[0065] Figure 18 To fit the change curves of the energy of roosters and hens with the polynomial model;

[0066] Figure 19 To fit the change curves of the energy of roosters and hens with the Von Bertalanffy model;

[0067] Figure 20 To fit the change curves of the energy of roosters and hens with the LSTM model;

[0068] Figure 21 To fit the change curves of the crude ash content of roosters and hens with the Logistic model;

[0069] Figure 22 To fit the change curves of the crude ash content of roosters and hens with the Gompertz model;

[0070] Figure 23To fit the change curves of crude ash content in roosters and hens with a polynomial model;

[0071] Figure 24 To fit the change curves of crude ash content in roosters and hens with the Von Bertalanffy model;

[0072] Figure 25 To fit the change curves of crude ash content in roosters and hens with the LSTM model;

[0073] Figure 26 To fit the change curves of calcium content in roosters and hens with the Logistic model;

[0074] Figure 27 To fit the change curves of calcium content in roosters and hens with the Gompertz model;

[0075] Figure 28 To fit the change curves of calcium content in roosters and hens with a polynomial model;

[0076] Figure 29 To fit the change curves of calcium content in roosters and hens with the Von Bertalanffy model;

[0077] Figure 30 To fit the change curves of calcium content in roosters and hens with the LSTM model;

[0078] Figure 31 To fit the change curves of phosphorus content in roosters and hens with the Logistic model;

[0079] Figure 32 To fit the change curves of phosphorus content in roosters and hens with the Gompertz model;

[0080] Figure 33 To fit the change curves of phosphorus content in roosters and hens with a polynomial model;

[0081] Figure 34 To fit the change curves of phosphorus content in roosters and hens with the Von Bertalanffy model;

[0082] Figure 35 To fit the change curves of phosphorus content in roosters and hens with the LSTM model;

[0083] Figure 36 To fit the change curves of body protein in roosters and hens with the Logistic model;

[0084] Figure 37 To fit the change curves of body protein in roosters and hens with the Gompertz model;

[0085] Figure 38 To fit the change curves of body protein in roosters and hens with a polynomial model;

[0086] Figure 39 Fitting the change curves of body protein in roosters and hens with the Von Bertalanffy model;

[0087] Figure 40 Fitting the change curves of body protein in roosters and hens with the LSTM model;

[0088] Figure 41 Fitting the change curves of body fat in roosters and hens with the Logistic model;

[0089] Figure 42 Fitting the change curves of body fat in roosters and hens with the Gompertz model;

[0090] Figure 43 Fitting the change curves of body fat in roosters and hens with the polynomial model;

[0091] Figure 44 Fitting the change curves of body fat in roosters and hens with the Von Bertalanffy model;

[0092] Figure 45 Fitting the change curves of body fat in roosters and hens with the LSTM model;

[0093] Figure 46 Fitting the change curves of energy in roosters and hens with the Logistic model;

[0094] Figure 47 Fitting the change curves of energy in roosters and hens with the Gompertz model;

[0095] Figure 48 Fitting the change curves of energy in roosters and hens with the polynomial model;

[0096] Figure 49 Fitting the change curves of energy in roosters and hens with the Von Bertalanffy model;

[0097] Figure 50 Fitting the change curves of energy in roosters and hens with the LSTM model;

[0098] Figure 51 Fitting the change curves of crude ash in roosters and hens with the Logistic model;

[0099] Figure 52 Fitting the change curves of crude ash in roosters and hens with the Gompertz model;

[0100] Figure 53 Fitting the change curves of crude ash in roosters and hens with the polynomial model;

[0101] Figure 54To fit the curves of crude ash content changes in roosters and hens with the Von Bertalanffy model;

[0102] Figure 55 To fit the curves of crude ash content changes in roosters and hens with the LSTM model;

[0103] Figure 56 To fit the curves of calcium content changes in roosters and hens with the Logistic model;

[0104] Figure 57 To fit the curves of calcium content changes in roosters and hens with the Gompertz model;

[0105] Figure 58 To fit the curves of calcium content changes in roosters and hens with the polynomial model;

[0106] Figure 59 To fit the curves of calcium content changes in roosters and hens with the Von Bertalanffy model;

[0107] Figure 60 To fit the curves of calcium content changes in roosters and hens with the LSTM model;

[0108] Figure 61 To fit the curves of phosphorus content changes in roosters and hens with the Logistic model;

[0109] Figure 62 To fit the curves of phosphorus content changes in roosters and hens with the Gompertz model;

[0110] Figure 63 The polynomial model fits the curves of phosphorus content changes in roosters and hens;

[0111] Figure 64 To fit the curves of phosphorus content changes in roosters and hens with the Von Bertalanffy model;

[0112] Figure 65 To fit the curves of phosphorus content changes in roosters and hens with the LSTM model. Detailed implementation manners

[0113] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0114] Embodiment 1

[0115] 1 Purpose and significance of the experimental study

[0116] It mainly focuses on the mechanism and application of energy conservation and emission reduction in the breeding of Soviet-style yellow-feathered broilers, and explores and clarifies the growth and development laws and nutrient deposition laws of high-quality yellow-feathered broilers represented by Lihua Company's Xueshan chickens through various experimental methods and data analysis.

[0117] 2 Materials and Methods

[0118] 2.1 Experimental Design

[0119] Under the normal floor rearing mode, 720 one-day-old Xueshan male and female chickens with similar weights were selected and randomly divided into 6 replicate pens, with 120 chickens in each pen, and each chicken was tagged with a wing tag. The chickens were allowed to feed and drink freely. The experimental measurement period was from 1 to 98 days old (d). Among them, at 1, 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84, 91, and 98 d, each chicken needed to be weighed; at 1, 14, 28, 42, 56, 70, 84, and 98 d, one chicken was randomly selected from each replicate pen for slaughter sampling (when the chickens were small, 3 - 4 chickens were sampled in a pool), and they were euthanized by asphyxiation.

[0120] 2.2 Diet Management

[0121] The chickens were fed normal commercial chicken feed (corn - soybean meal-based diet) throughout the experiment. The experimental management measures followed the regulations for slow-growing chickens.

[0122] 2.3 Detection Indexes

[0123] 2.3.1 Routine Diet Indexes

[0124] Detect the routine nutritional indexes of the diet.

[0125] 2.3.2 Production Performance

[0126] Each chicken was weighed on an empty stomach, and each replicate pen was weighed in the morning (fasted for more than 8 h, and the fasting time was appropriately reduced for smaller-day-old chickens), and the feed consumption, death and culling numbers, and death and culling weights at each stage were recorded. Calculate the average daily gain, average feed intake, and feed-to-weight ratio at each stage and throughout the experiment.

[0127] 2.3.3 Nutrient Deposition

[0128] The chickens were weighed and then euthanized by asphyxiation. The contents of the digestive tract were removed, and the fasting body weight was recorded after weighing. The chickens were weighed and then euthanized by asphyxiation. The contents of the digestive tract were removed, and the fasting body weight was recorded after weighing. The experimental chickens were initially ground using a poultry grinder (Model 160, customized by Yusheng Machinery Factory, Xingtai, Hebei, China), and then ground a second time using a bone paste mill (Model GN-200, customized by Yusheng Machinery Factory, Xingtai, Hebei, China) and weighed. Then, the samples were spread out and sterilized in an oven at 105 °C for 15 minutes, and then dried at 65 °C for 48 hours and weighed. Next, the samples that had dried into hard blocks were finely ground using a cell wall breaker. Finally, after being conditioned for 24 hours, air-dried samples were obtained. Two independent samples were taken by the quartering method twice and stored in a -20 °C low-temperature refrigerator. Water, ash, calcium, phosphorus, crude protein, crude fat, and gross energy were to be measured;

[0129] The nutrient deposition amount = body weight × carcass nutrient content.

[0130] 3 Data processing and model analysis

[0131] 3.1 Analysis of variance

[0132] The experimental data were statistically analyzed using R language software version 4.2.0. Two-way analysis of variance (two-way ANOVA) and Tukey multiple comparisons were used, and the statistical significance level was P < 0.05. The experimental data for each group were expressed as means ± standard error of the mean (means ± SEM).

[0133] 3.2 Nonlinear regression fitting

[0134] The experimental data were statistically analyzed using R language software version 4.2.0. The NLS nonlinear model was used to fit the models for the age and body weight, energy deposition amount, protein deposition amount, fat deposition amount, ash deposition amount, calcium deposition amount, and phosphorus deposition amount of Xueshan partridge chickens respectively. The Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model were used for nonlinear fitting, and the fitting effects of various models were compared;

[0135] Logistic model,

[0136] Gompertz model,

[0137] Von Bertalanffy model, y(t) = A(1 - Be -kt ) 3 ;

[0138] Polynomial model, y(t) = a0 + a1t + a2t 2 +... + a n tn ;

[0139] LSTM model, f t = σ(W f · [h t-1 , x t + b f );

[0140] i t = σ(W i · [h t-1 , x t + b i );

[0141]

[0142] o t = σ(W o · [h t-1 , x t + b o );

[0143] h t = o t ⊙ tanh(C t );

[0144] In the formula: y is the nutrient deposition of broilers when the age reaches x; a is the asymptote of the curve (the maximum deposition at maturity); b is the maximum relative growth rate; c is the inflection point age when the growth rate reaches the maximum; t is the age; k is the growth rate parameter;

[0145] h t : The output prediction at time point t, which can represent the predicted values of growth indicators such as animal weight and body fat percentage;

[0146] f t : The forget gate, which determines how much information of the previous growth state should be retained, for example, determines the influence degree of the past growth trend on the current prediction;

[0147] i t : The input gate, which determines how much new information should be accepted, for example, the influence intensity of current environmental conditions or changes in feeding management;

[0148] o t : The output gate, which controls which currently calculated growth information should be output as the prediction result;

[0149] : The candidate growth state, which contains new information that may affect the current growth;

[0150] C t : The cell state, which represents the long-term memory of animal growth and stores the cumulative information of the entire growth cycle;

[0151] C t-1 : The growth state at the previous time step, representing the previous growth history;

[0152] W f , W i , W C , W o : The weight matrix of each gating unit, determining the influence degree of different input features on growth prediction;

[0153] b f , b i , b C , b o : The bias term, used to adjust the baseline prediction value of the model.

[0154] 3.3 Nonlinear regression model evaluation

[0155] After the model is fitted, load the modelr package in R language and use indicators such as the coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE), Akaike information criterion (AIC), and Bayesian information criterion (BIC) to comprehensively evaluate the fitting of the Logistic model and Gompertz model to the data of the dynamic deposition laws of body energy, body protein, body fat, body ash, and calcium and phosphorus.

[0156] 4 Results and analysis

[0157] 4.1 Changes in body weight and body composition weight of Snow Mountain partridge chickens at different ages

[0158] Table 1 Live body weight growth model of each chicken

[0159]

[0160] From Table 1 and Figure 1-5 it can be seen that:

[0161] The Gompertz model performs better overall, has good effects in terms of fitting degree and error control, and is relatively excellent in simulating the growth and development laws of roosters and hens. It is a relatively better choice among these models. However, the LSTM model also has good fitting ability for hens.

[0162] 4.2 Fitting analysis of the growth curve of Snow Mountain partridge chickens

[0163] The Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model were respectively used to fit the body protein content, body fat content, energy content, crude ash content, calcium content, and phosphorus content of Snow Mountain partridge chickens of different genders. The results are shown in Tables 2 - 7 andFigure 1-35 Perform fitting, and the results are shown in Table 2-7:

[0164] Table 2

[0165]

[0166] Table 3

[0167]

[0168] Table 4

[0169]

[0170]

[0171] Table 5

[0172]

[0173] Table 6

[0174]

[0175] Table 7

[0176]

[0177]

[0178] 4.3 Comparison of Nonlinear Growth Fitting Models for Xueshan Grass Chickens

[0179] Use the Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model to compare the data of nutrient (energy, body protein, body fat, crude ash, calcium, and phosphorus) deposition in Xueshan grass chickens of different genders with the change of days of age. The results are shown in Tables 8-13 and Figure 36-65 .

[0180] Table 8

[0181]

[0182] Table 9

[0183]

[0184]

[0185] Table 10

[0186]

[0187] Table 11

[0188]

[0189] Table 12

[0190]

[0191] Table 13

[0192]

[0193] 4.4 Fitting Analysis of Nutrient Deposition Rate of Xueshan Grass Chicken

[0194] 1. In the fitting analysis of the nutrient deposition rate of Xueshan grass chicken, different models showed different performances.

[0195] 2. In the data fitting of body protein deposition amount changing with age, in the data of hens, the polynomial model (0.961) and the LSTM model (0.967) R 2 were relatively optimal, and in the data of roosters, the LSTM model (0.952) R 2 was also relatively high. RMSE and MAE reflect the error situation, and the smaller the value, the more accurate the prediction. In the data of hens, the polynomial model (RMSE was 23.46, MAE was 17.04) and the LSTM model (RMSE was 22.2, MAE was 15.6) had smaller errors. Generally speaking, in the simulation of body protein deposition growth, the LSTM model and the polynomial model performed excellently in multiple key indicators and had better comprehensive performance.

[0196] 3. Fat deposition: In the data of hens, the polynomial model (0.917) and the LSTM model (0.934) R 2 were relatively high and had relatively good fitting effects; in the data of roosters, the differences among models were smaller. In terms of the RMSE and MAE indicators, in the data of hens, the polynomial model (RMSE was 37.83, MAE was 28.21) and the LSTM model (RMSE was 33.17, MAE was 22.84) had smaller errors and more accurate predictions; in the data of roosters, the values of each model were relatively close. Generally speaking, in the simulation of body fat deposition growth, the LSTM model, the polynomial model and the Gompertz model performed well in some indicators and had relatively leading comprehensive performance.

[0197] 4. Energy deposition: In the hen data, the Gompertz model (0.935), polynomial model (0.949), and LSTM model (0.957) are relatively high. The differences among the models in the rooster data are relatively small. In the hen data, the polynomial model (RMSE is 392.74, MAE is 257.22) and the LSTM model (RMSE is 360.05, MAE is 225.62) perform better. The values of each model in the rooster data are relatively close. Generally speaking, in the simulation of energy deposition growth, the overall performance of the LSTM model and the Gompertz model is relatively better.

[0198] 5. Crude ash deposition: The R of the LSTM model 2 is relatively high in both roosters and hens, and its fitting ability is outstanding. In terms of the RMSE and MAE indicators, the values of the LSTM model are generally small, and its prediction accuracy is better. Among the AIC and BIC indicators, some models perform well in the rooster or hen data. Generally speaking, the overall performance of the LSTM model is better, and it has obvious advantages in fitting and predicting the growth of crude ash deposition.

[0199] 6. Calcium deposition: The fitting effect of the LSTM model is the most prominent. The R 2 values of the fitting models for roosters and hens are the highest compared with other models, and it can well present the changes in calcium deposition.

[0200] 7. Phosphorus deposition: The polynomial model (5th degree) and the LSTM model are relatively good. The R 2 value of the polynomial model (5th degree) is 0.71 for roosters and 0.88 for hens, and the R 2 value of the LSTM model is 0.71 for roosters and 0.88 for hens, and it has a relatively high fitting degree among the models.

[0201] 5 Conclusions

[0202] Based on the above results, the Gompertz model generally has a good fitting effect on the growth and development laws of Xueshan partridge chickens. Whether it is the fitting of body weight or the deposition of various nutrients, it can achieve a good balance between fitting degree and error control. The LSTM model performs outstandingly in fitting the body weight of hens and the deposition of various nutrients, especially in the fitting of nutrients such as body fat and calcium. Generally speaking, the Gompertz model and the LSTM model have high application value in the study of the growth and development and nutrient deposition laws of Xueshan partridge chickens, and can provide strong support for further understanding the growth characteristics of Xueshan partridge chickens and optimizing breeding strategies.

[0203] In the present specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the apparatuses disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple. For the relevant parts, reference can be made to the descriptions in the method section.

[0204] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those 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 invention. Therefore, the invention is not limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for establishing an intelligent model of the dynamic deposition law of the body composition of Xueshan partridge chickens, characterized in that, Including: (1) Under the normal floor rearing mode, 720 1-day-old Snow Mountain grass cocks and hens with similar body weights were selected and randomly divided into 6 replicated pens, with 120 chickens in each pen. Each chicken was fitted with a wing tag, and the chickens were allowed to feed and drink freely. The experimental measurement period was from 1 to 98 days old. Among them, at 1, 7, 14, 21, 28, 35, 42, 49, 56, 63, 70, 77, 84, 91, and 98 days, each chicken was weighed; at 1, 14, 28, 42, 56, 70, 84, and 98 days, one chicken was randomly selected from each replicated pen for slaughter sampling and was sacrificed by asphyxiation; (2) Throughout the experiment, a corn-soybean meal-based basal diet was fed, and the experimental management measures followed the slow-growing chicken management regulations; (3) The conventional nutritional indexes, production performance, and nutrient deposition indexes of the diet were detected; (4) The experimental data were statistically analyzed using R language 4.2.0 software. Two-way analysis of variance and Tukey multiple comparisons were used, and the statistical significance level was p < 0.

05. The experimental data of each group were expressed as the mean ± standard deviation; the NLS nonlinear model was used to fit the models of the age and body weight, energy deposition, protein deposition, fat deposition, ash deposition, calcium deposition, and phosphorus deposition of Snow Mountain grass chickens respectively. The Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model were used for nonlinear fitting, and the fitting effects of various models were compared; Logistic model, Gompertz model, Von Bertalanffy model, y(t) = A(1 - Be -kt ) 3 ; Polynomial model, y(t) = a0 + a1t + a2t 2 + … + a n t n ; LSTM model, f t = σ(W f · [h t-1 , x t + b f ) i t = σ(W i · [h t-1 , x t + b i ); o t = σ(W o · [h t-1 , x t + b o ); h t = o t ⊙tanh(C t ); Where: y is the nutrient deposition of broilers when the age reaches x; a is the asymptote of the curve; b is the maximum relative growth rate; c is the inflection point age when the growth rate reaches the maximum; t is the age; k is the growth rate parameter; h t is the output prediction at time point t; f t is the forget gate; i t is the input gate; o t is the output gate; is the candidate growth state; C t is the cell state; C t-1 is the growth state at the previous time step; W f ,W i ,W C ,W o is the weight matrix for each gating unit; b f ,b i ,b C ,b o is the bias term; (6) After model fitting, the R language modelr package was loaded, and the regression determination coefficient, root mean square error, mean absolute error, Akaike information criterion, and Bayesian information criterion were used to comprehensively evaluate the fitting of the Gompertz model, Logistic model, Von Bertalanffy model, polynomial model, and LSTM model to the data of the dynamic deposition laws of body energy, body protein, body fat, body ash, and calcium and phosphorus.

2. The method for establishing an intelligent model of the dynamic deposition law of body composition of Xueshan partridge chickens according to claim 1, wherein, The production performance was detected by the following method: Each chicken was weighed on an empty stomach, and each replicated pen was weighed in the morning. The feed consumption, death and culling numbers, and death and culling weights at each stage were recorded, and the average daily gain, average feed intake, and feed-to-weight ratio at each stage and throughout the period were calculated.

3. A method for establishing an intelligent model of the dynamic deposition law of the body composition of Xueshan partridge chickens according to claim 1, characterized in that, The nutrient deposition was detected by the following method: The chickens were weighed and then euthanized by asphyxiation. The contents of the digestive tract were removed, and the empty stomach weight was recorded after weighing. The chickens were weighed and then euthanized by asphyxiation. The contents of the digestive tract were removed, and the empty stomach weight was recorded after weighing; The experimental chickens were initially ground with a poultry grinder, and then weighed after being ground a second time with a bone paste grinder; Then, the sample was spread out and sterilized in an oven at 105 °C for 15 minutes, and then dried at 65 °C for 48 hours and weighed; Next, the sample dried into a hard block was finely ground with a cell wall breaker. Finally, an air-dried sample was obtained after rehydrating for 24 hours. Two independent samples were taken by the quartering method twice and stored in a -20 °C low-temperature refrigerator for determination of moisture, ash, calcium, phosphorus, crude protein, crude fat and gross energy; Nutrient deposition = body weight × carcass nutrient content.

Images