Artificial intelligence-based metabolic energy prediction model for animal feed proportioning optimization

Abstract With the progress of science and technology, Artificial Intelligence (AI) technology has become one of the mainstream technologies in the current society, providing an important driving force for human development. Thereby, in order to improve the effect of animal feeding, using AI technology to improve animal feed has become a necessary measure. Based on this, this work designs to use Long Short-Term Memory (LSTM) technology to build an intelligent prediction model of metabolic energy, which provides a reference for animal feed proportioning design. This work also explores the comprehensive performance of the LSTM model through simulation evaluation. The model is evaluated with different nodes as the main indicators. The results show that compared with the models with 5 and 20 nodes, the model with 10 nodes has better performance, and the highest data calculation accuracy of the model is about 90%. Meanwhile, the highest fitting degree of the model designed is 98.2%, and the lowest is 96.2%. It suggests that the model designed can better predict metabolic energy. This work provides technical support for expanding the application scope of AI technology and contributes to the intelligence of animal feeding.


Introduction
With the continuous progress of Artificial Intelligence (AI) technology and society, intelligent production has become the main way of social development, so it has become necessary to achieve intelligence in various industries (Mashamba-Thompson and Crayton 2020).Based on this, in order to realise intelligent animal feeding programs to improve the effect of animal feeding, it has become a mainstream measure to predict the metabolic energy through AI technology, and then provide a reference for animal feed proportioning design.Animal feed proportioning is a crucial basis for raising animals.By improving the formulation method of feed proportioning, the effect and accuracy of feed proportioning can be improved, and the speed and intelligence of feed proportioning can be enhanced (Gasco et al. 2020).Moreover, many current studies have also provided important help for this approach.Morais et al. (2020) pointed out that feed is animals' food source and becomes an important material basis for the long-term development of animal husbandry and the animal production industry.As a crucial part of animal nutrition and feed science, animal feed preparation has become a relatively complete research system after a hundred years of development.Therefore, optimising the animal feed formula system and improving the feed proportioning have critical practical significance for promoting the development of animal husbandry, increasing the proportion of the agricultural economy and industry, and promoting the overall development of China's economy (Morais et al. 2020).Georganas et al. (2020) held that feed proportion is a significant basis for the scientific feeding of animals.Economical and reasonable feed formula must be designed according to the nutrient requirements specified in the feeding proportion (Georganas et al. 2020).Based on the selected feeding standards, appropriate adjustments can be made according to animal growth or production performance in the feeding practice (Alshelmani et al. 2021;Alshelmani et al. 2021;Kareem et al. 2018;Kairalla et al. 2022).Marton et al. (2021) measured the variation of dry matter digestibility and metabolic energy of commonly used duck feed materials (corn, soybean meal, cottonseed meal and wheat bran) by using a bionic digestive system.Besides, they discussed the repeatability and precision of the system to determine the metabolic energy of duck feed raw materials, providing a reference for establishing a biological titre method for the quantitative determination of feed nutrients based on the bionic digestive system (Marton et al. 2021).Kazak and Cohen (2020) designed an artificial small intestinal juice similar to the digestive enzyme activity in duck small intestinal juice according to the variation range of digestive enzyme zymogram in duck small intestinal juice under normal conditions.They established an in vitro evaluation method for estimating duck feed metabolic energy using the pepsin -artificial duck small intestinal juice two-step enzymatic hydrolysis method.This method provided a modern measurement method for constructing effective energy parameters of waterfowl breeding standards (Kazak and Cohen 2020).Chimmula and Zhang (2020) pointed out that biological manufacturing plays a decisive role in society's sustainable economic development model with the advantages of low-carbon, circular, green and clean and its development is the general trend in the future.With technological progress, biological manufacturing has formed massive data in the regulation of microbial cell metabolism and online data of biological processing processes.However, the imperfection of traditional data analysis theory greatly limits the mining of massive data.Making use of AI's advantages in data processing and deeply mining the knowledge contained in the data through machine learning is a crucial direction to realise the intelligent path of biological manufacturing (Chimmula and Zhang 2020).To sum up, the current utilisation of metabolic energy and the design of animal feed proportioning has become a practical technology, but the current application of AI technology is not mature in feed proportioning optimisation.Therefore, in order to highlight the role of AI technology and improve the intelligence level of feed proportioning optimisation, this work designs to use AI technology to build a metabolic energy prediction model, and then provide support for optimising animal feed proportioning.
To sum up, this work first expounds on the role and importance of animal feed proportioning, which provides a theoretical basis for optimising animal feed proportioning.Then, the importance of metabolic energy prediction in the optimisation of animal feed proportioning is discussed to highlight the importance of this work.Finally, an AI metabolic energy prediction model is constructed based on Long Short-Term Memory (LSTM), and its comprehensive performance is evaluated through simulation experiments.This work supports the optimisation of animal feed proportioning and contributes to the further development of AI technology.

Optimisation of animal feed proportioning
In a broad sense, feed is the general term for the food of animals raised by all people.In a narrow sense, feed mainly refers to the food of animals raised in agriculture or animal husbandry.The feed includes more than ten kinds of feed materials, such as soybean, soybean meal, corn, fish meal, amino acid, miscellaneous meal, whey powder, oil, meat and bone meal, cereals, and feed additives.Animal feed proportioning refers to the optimal combination of different feed materials to meet the nutritional needs of animals.For example, soybean meal, corn and other raw materials are mixed in a certain proportion and processed into feed to feed pigs to ensure normal growth of pigs (Nagarajan et al. 2021).Meanwhile, since human society is an economic society, it is not enough to ensure the normal growth of animals.There are also multiple other requirements to be met, such as fast growth, low cost and large income.In addition to the needs of energy and protein for animal growth, development and reproduction, humans also have requirements for them, such as more lean meat.In this way, humans must artificially change the nutritional needs of animals to achieve certain goals (Alagappan et al. 2022).Based on this, the design of animal feed proportioning plays a vital role in animal feeding.
The essence of feed proportioning is an operational research problem of optimal allocation of resources.It can be quantitatively described through appropriate linear or nonlinear decision-making models.The solution of these models can achieve the optimal allocation of resources, that is, to obtain the lowest cost of the formula or the maximum benefit of the formula.Linear decision models include Linear Programming (LP) and Multiple Goals Programming developed on this basis.With the development of other branches of applied mathematics and the needs of actual formula design, LP models derive Stochastic Nonlinear Programming, Fuzzy Linear Programming and Grey Linear Programming (Hong et al. 2020).The nonlinear decision model corresponds to the nonlinear programming mode.However, because it is more complex than the LP model, it is only gradually applied in recent years with the development of computer technology and animal nutrition science.
Therefore, it can be found that the current research method of animal feed proportioning is not intelligent enough, which leads to the inability to deepen and improve the intelligence level of animal feeding.Thereby, the current intelligent research on feed proportioning is imminent, and lots of research is urgently needed to provide support.

Intelligent metabolic energy prediction
Metabolic energy usually refers to the digestible energy minus the energy lost in urine and methane gas.Metabolic energy is the energy that can be transported and used by the animal.It considers the energy lost in the urine and the energy lost in methane gas produced in the digestion process of feed, which is more accurate than digestible energy.However, it is relatively difficult to measure metabolic energy in production practice because respiratory calorimetry is required for methane measurement.For ruminants, digestible energy Â 0.82 is often adopted in production to calculate the metabolic energy (He et al. 2020).Currently, the metabolic energy system is mainly used in poultry and some ruminants.
The energy that can be digested in the whole digestive tract is digestible energy, which is usually determined by measuring the combustion heat of the feed with an energy metre.Generally, faecal energy is the largest part of feed energy loss, and urinary energy is usually low, so digestible energy can represent most animals' energy needs.Moreover, compared with metabolic and net energy, determining digestible energy is relatively easy.Currently, the world's pig nutrition needs to use the digestible energy system (Lopaschuk et al. 2021).
The feed's net energy is the feed's metabolic energy minus the heat increment, which is used for the maintenance of the body and the energy of different production forms.The net energy used for maintenance is mainly used to do work in the body and dissipate in the form of heat.The net energy used for growth, fattening, lactation, egg production or wool production is either stored in the body or discharged in the form of chemical energy.The net energy used in this way is the energy retention of animals (Wiesner et al. 2021).The net energy system considers the loss of faecal energy, urine energy, gas energy, and body heat increment, which is more accurate than the digestible energy and metabolic energy.However, the net energy system is relatively complex, because the net energy value of any feed is different due to different animal production purposes.Moreover, for convenience, the different net energy of production is often converted into the same net energy.For example, the net energy used for maintenance and growth is converted into net energy of milk production.There is a large error in the conversion process.Besides, the measurement of net energy is also very difficult and time-consuming (Rabinowitz and Enerb€ ack 2020).At present, the energy needs of ruminants are mainly expressed in net energy system (Hargreaves and Spriet 2020).
Based on the above complex metabolic energy prediction process, intelligent metabolic energy prediction has become the main goal of the current evolution of metabolic energy prediction methods.Intelligent metabolic energy prediction is the prediction of animal metabolic energy through AI technology to support the animal feed proportioning process.

Design of intelligent metabolic energy prediction model
With the increasingly fierce competition in the product market, the advantages of product intelligence have been well used in practical operation and application.It is mainly manifested in: greatly improving the operator's working environment, reducing the work intensity, improving the work quality and efficiency, solving some dangerous situations or key construction applications, environmental protection and energy conservation, improving the automation and intelligence level of the machine, improving the reliability of the equipment, reducing the maintenance cost, and realising intelligent fault diagnosis (Lee and Yoon 2021).
It reveals that intelligent technology plays a vital role in the development of society as a whole, and also has great significance for the development of various industries in society (Vrontis et al. 2022).Based on this, in order to realise the intelligence and depth rationalisation of animal feed proportioning, this work uses AI technology to design a metabolic energy prediction model, thus providing a basic reference for animal feed proportioning design (Zhao et al. 2020).
AI technology is a new technology science to research and develop theories, methods, technologies and application systems for simulating, extending and expanding human intelligence (Mohanta et al. 2020).AI is a branch of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence.Research in this field includes robots, language recognition, image recognition, natural language processing and expert systems (Li et al. 2021).Due to the continuous development of society, human beings are increasingly pursuing the intelligence degree, so the generation of AI technology has also significantly impacted social development (Mekruksavanich and Jitpattanakul 2021).
Based on this, this work uses LSTM technology to design an intelligent metabolic energy prediction model to improve metabolic energy's prediction effect, and provide support for the development of intelligent animal feed proportioning technology in the future.Figure 1 displays the basic structural principle of the designed model.
Figure 1 suggests that LSTM is a deep learning model based on Recurrent Neural Network (RNN) technology.Hence, it also has the characteristics of RNN technology, and it is an optimisation model of RNN technology.Based on RNN technology, LSTM effectively overcomes the problem of gradient disappearance and realises long-term and short-term memory of information (Jalayer et al. 2021).The three-gate structure of the LSTM network can selectively remember the input timing information and the modified parameters of the weights of each neuron in the backpropagation process, and will not send its own behaviour to other neurons as input or output values (Lindemann et al. 2021).Moreover, only when the connection weight value of the forget gate is 1, can the corresponding data storage, reading, memory, transmission and other operations be performed.When the connection weight value is 0, the forget gate will not perform any operation (Park et al. 2020).The calculation equation of the LSTM memory unit is: represents the weight, and b represents the offset term.Meanwhile, there are hidden neurons, whose calculation equation is: The output result can be calculated according to the above equation.The calculation equation is: (5) The calculation of output results can show that LSTM needs to focus on data information to be memorised, that is, LSTM can decide to eliminate useless information through the calculation results to retain useful information (Shahid et al. 2020).LSTM and RNN technology also have forward output and reverse error transmission, so it is necessary to calculate its weight derivative.The calculation equation is: O represents the loss function and o represents the partial derivative.Based on this, after forward calculation, the calculation equation of the input gate is: The calculation equation of the forget gate is: The calculation equation of the memory unit is: The calculation equation of the output gate is: If reverse calculation is performed, the calculation equation of unit output is: The calculation equation of the output gate is: The calculation equation for the current status is: The calculation equation of the memory unit is: The calculation equation of the forget gate is: The calculation equation of the input gate is: The LSTM model has multiple outstanding advantages.First, the LSTM model improves the long-term dependence problem in the Convolutional Neural Network (CNN) model.Then, LSTM usually performs better than the time recursive neural network and the hidden Markov model.Finally, as a nonlinear model, LSTM can be used as a complex nonlinear element to construct a larger deep neural network.Therefore, this work constructs an AI metabolic energy prediction model based on LSTM, which provides a reference for the intelligent design of animal feed proportioning (Sherstinsky 2020).

Experimental data design
A total of 50 0 $ 3-week-old muscovy ducks were used as an example.Ten raw materials, including corn, soybean cake, rice bran cake, meat and bone meal, lysine, methionine, stone flour, calcium hydrogen phosphate, premix and salt, are proposed to participate in the formula design.Table 1 presents the nutrient content and its variation range of various raw materials.
Table 1 shows that the above raw materials are used as the basic elements of animal feed proportioning, and then the metabolic energy of animals is predicted through the LSTM model, thus providing a reference for designing appropriate animal feed proportioning to optimise the animal feeding process and improve the effectiveness of animal feeding.Table 2 displays the assurance probability of each nutrient index.
Tables 2, 3 show that the above specifications are adopted to predict and evaluate the metabolic energy of animals in the simulation experiment, thus providing a reference for the design of animal feed proportioning.

AI technical performance evaluation
LSTM model is an advanced AI technology, and its application scope is also extensive.Hence, this work uses LSTM to design a metabolic energy prediction model.On this basis, it first evaluates and analyzes the    Figures 3 and 4 suggest that the LSTM designed here has better performance and higher accuracy than CNN and RNN.

Training of metabolic energy prediction model
In order to improve the prediction effect of the model, the prediction effect of the model is first trained and through simulation experiments to detect     Fitting degree was calculated using the simple equation: (forecast 1/raw data) Â 100.

Detection of model metabolic energy prediction performance
As mentioned above, the metabolic energy prediction model is trained and evaluated first to provide support for improving the model's performance.Then, the final model is used as the experimental object for the detection and evaluation experiment of metabolic energy prediction.Figure 8 displays the results of this model's metabolic energy prediction performance test.
Figure 8 presents the effect of metabolic energy prediction of the model obtained by comparing the original data with the model prediction results.Table 4 shows the statistics of the accuracy results of the model prediction.

Conclusion
With the continuous evolution of animal feeding, intelligent feeding has become the main evolutionary goal of animal feeding in the current society.Therefore, this work aims to improve the intelligence level of animal feeding and improve the animal feed proportioning design.It designs to use AI technology to build a metabolic energy prediction model, and provide support for improving animal feed proportioning based on this technology.Although this work has designed a more advanced technical model and conducted a more comprehensive evaluation and analysis of the model, it has not explored the effect of the model in practical application, so it is unable to actually characterise the model's characteristics.Based on this, the research on the application of the model in real life will be strengthened in future research to further explore the characteristics and value of the model.

Figure 1 .
Figure 1.Design of Artificial Intelligence metabolic energy prediction model.

Figure 2 .
Figure 2. Performance evaluation results of the Long Short-Term memory (LSTM) model.

Figure 3 .
Figure 3.Comparison results of Long Short-Term memory (LSTM) and Convolutional Neural Network (CNN) models.

Figure 5 .
Figure 5. Training results of the metabolic energy prediction model in the Day time interval.

Figure 6 .
Figure 6.Training results of the metabolic energy prediction model in the week time interval.

Figure 7 .
Figure 7. Training results of the metabolic energy prediction model in the Month time interval.

Figure 8 .
Figure 8. Results of model metabolic energy prediction performance test.

Table 1 .
Nutrient content and variation range of various raw materials.

Table 2 .
Assurance probability, expansion and contraction and weight of each nutritional index.

Table 3 .
Constraints of nutrients and raw materials.

Table 4 .
Statistics of accuracy results of model prediction.
Table 4 displays the statistical results of the model's fitting degree.It reveals that the highest fitting degree of the model designed is 98.2%, and the lowest is 96.2%.It suggests that the model designed can better predict metabolic energy.