Study Of The Key Techniques Of Intelligent Support System For Sow Healthy Culture

With the rapid development of pig breeding industry,strengthening the key technology research of intelligent system has become one of the key contents of swine production.As the source of production,sow health is directly related to the safety of breeding.Harsh environment and improper management have adverse effects on sow health.To realize the precise assessment,prediction and regulation of swine house environment and precise feeding,environmental comfort and intelligent feeding methods should be studied,which is significant for the swine production.This paper takes sow as the research object and builds an intelligent breeding support system.The methods of environmental control,ammonia concentration prediction and modeling,environmental comfort evaluation and prediction and the expert system,the development and application of intelligent management system are mainly studied.The main research contents and results are as follows:（1）In view of uneven distribution of environmental parameters in sow house,three-dimensional measurement model is established.Temporal and spatial distribution characteristics of temperature,relative humidity,concentrations of NH₃ and CO₂ are analyzed.In order to control the sow house environment in different seasons,variable universe fuzzy controllers of temperature and ventilation are designed.In addition,temperature is predicted by a combination of grey model and weights and the predictive value is employed as a feedback value of the temperature controller.The contradiction between temperature and ventilation control is solved by heating compensation based on thermal balance equation and the control of sow house environment is optimized.（2）In order to improve the prediction accuracy of current prediction model of NH₃concentration and a novel prediction model is proposed.The key variables are selected with random forest algorithm.Abnormal data are eliminated with Grubbs criterion.Improved self-adaptive weighted fusion algorithm fuses the data sources.ACFOA algorithm is employed to find optimal parameters,and finally the NH₃ concentration prediction model based on ACFOA-LSSVR is established by combining different kernel functions.The results demonstrate that the model has the highest fitting degree and the smallest errors comparing with other models,such as single kernel function ACFOA-LSSVR,GS-LSSVR,GA-LSSVR and PSO-LSSVR.Furthermore,the GS-LSSVR,GA-LSSVR and PSO-LSSVR models and this model are compared under different time scales,which proves this model is more suitable for long-term prediction.（3）The evaluation and prediction method of comfortable degree of lactating sow is studied.Herein,the coupling effects of the environmental factors are considered.The evaluation index system of environmental comfort of lactating sow is established and the construction method of membership function is proposed.The IAHP-FCE model is established to assess the comfortable level of lactating sow.Furthermore,environmental comfort prediction model based on MSCCS-LSSVR is developed by mutative scale chaos cuckoo search algorithm.The combination of gamma and sigma is optimized by this algorithm.The results show that the IAHP-FCE model has higher correlation coefficients of 0.722 and 0.662 comparing with the single factor evaluation（SFE）model under different weather conditions and it is more effective to deal with uncertainty and ambiguity in comfort evaluation.In addition,the MSCCS-LSSVR model has the highest prediction accuracy comparing with the models optimized by GS,GA and PSO algorithms.The MAE,RMSE and R~2 are 0.0639,0.1787 and 0.9086,respectively.The model has better prediction accuracy and generalization ability.（4）To guide and manage the sow breeding process practically,an expert system is studied.Production knowledge representation is employed to establish the knowledge base.Daily feeding amount is decided by using an uncertainty reasoning method according to different combinations of conditions in the rule set.An improved Rete algorithm based on reused degree model is studied to construct a network with high reused degree.The results show that,the reused performance of nodes of the RDM-Rete model increases by 25%comparing with the standard Rete model.The average inference time of the Hash-RDM-Rete model decreases by 42.57%and 32.35%,respectively,comparing with the standard Rete model and RDM-Rete model.The deviation ratio of the inference results is zero.It is demonstrated that the Hash-RDM-Rete model can significantly reduce the complexity of the constructed network and speed up inference.（5）An intelligent breeding management system is developed to obtain real-time environmental information,which has the functions of intelligent evaluation and prediction,control and feeding.Early warning of environmental variations is realized with the evaluation and prediction model.The key environmental parameters are controlled within the comfortable range and the adverse impact of harmful gases on sow health is eliminated.In addition,the relationship between environmental comfort and feeding data is studied,which provides guidance to select conditions when establishing a decision model.Furthermore,intelligent feeders based on the expert system and internet of things（Io T）are developed to feed pregnant and lactating sows.The results show that,enough feed intake is obtained in different stages,and the piglets weaned sow year（PSY）of pregnant sow could increase by 1.51,which improves the economic efficiency of healthy culture.