The Study Of Artificial Intelligence-assisted Resource Treatment And Efficiency For Corn Straw

Corn straw is one of the main agricultural wastes in China,and its efficient resource utilization has been a challenge in this field.The corn straw has a dense structure,with lignin wrapped outside and cellulose and hemicellulose intertwined and distributed inside,so strengthening its pretreatment effect to remove lignin is a prerequisite for resource use technology for corn straw.Hemicellulose is mainly composed of heterogeneous polysaccharides including xylose,arabinose,mannose,and galactose,etc.The enzymatic conversion of macromolecular chain polysaccharides into oligomeric polysaccharides is an important way to utilize hemicellulose resourcefully.Polysaccharides have various biological activities such as anti-inflammatory,antioxidant,antitumor and low toxicity,and are potential alternatives for related drugs.Cellulose is made of large glucose glucans,which can be used as a substrate for bioethanol production through simultaneous enzymatic fermentation.However,simultaneous enzymatic fermentation for ethanol production system contains several biological processes,such as cellulose enzymatic digestion,Saccharomyces cerevisiae fermentation,and microflora interactions.Besides,the high price of enzymes prevents large-scale application.Therefore,developing efficient methods to optimize enzymatic digestion and fermentation is crucial for resource utilization of corn straw.In this study,we first investigated the optimum concentration of Na OH pretreatment for corn straw and carried out the detection for cellulose,hemicellulose,lignin content and straw weight loss.The lignin removal effect was visualized by SEM characterization analysis,and it was concluded that the Na OH solution with a concentration of 1.5% could combine better lignin removal,lower cellulose hemicellulose loss and suitable Na OH usage,which was the best Na OH solution concentration among all experimental groups.Therefore,this concentration of Na OH solution was used for uniform pretreatment of corn stover as a substrate for subsequent experiments.Yield prediction modeling,model interpretability analysis,and yield optimization are performed for two processes,hemicellulose to polysaccharide and cellulose fermentation to ethanol,respectively,using a variety of artificial intelligence approaches,including machine learning,deep learning,interpretability analysis,and optimization algorithm coupling.This study proposed a machine learning modeling framework to construct an enzymatic polysaccharide yield prediction model for xylan digestion corn straw production scenario,containing four machine learning models(LR,Tree,RF,XGB)and a self-designed deep learning model(DNN).Among them,XGB has the highest prediction accuracy of 95.6%,while RF and DNN models have slightly lower prediction accuracy with XGB,93.0% and 91.1%,respectively.Interpretability analysis quantified for the first time the contribution of each input variable to polysaccharide yield prediction(Enzyme Solution Volume: 43.7%;Time: 20.7%;Substrate Concentration: 15%;Temperature: 15%;p H: 5.6%).The optimization results of XGB-PY-GA,a polysaccharide yield optimization model constructed based on XGB,showed that the new optimization scheme significantly optimized the original polysaccharide production scheme by reducing the enzyme usage by 28%,decreasing the reaction temperature by 6°C and reducing the reaction time by 13 min while keeping the polysaccharide yield close.The analysis of the physicochemical properties for the polysaccharides showed that the surface of the obtained polysaccharide products was smooth,with the characteristic functional groups of polysaccharides and good thermal stability.The analysis of antioxidant activity showed good scavenging effect on superoxide radicals,hydroxyl radicals and ABTS radicals,but poor scavenging ability on DPPH radicals.This study achieved accurate prediction of ethanol yield through the construction of DNN-EY model and XGB-EY model.The results showed that the prediction accuracy of DNN-EY was 85%,which was higher than that of XGB-EY at 83.6%.Interpretability analysis showed the contribution of each input variable to the simultaneous enzymatic co-fermentation ethanol yield prediction(Enzyme Solution Volume: 61.7%;Time: 12.9%;Substrate Concentration: 10.4%;Temperature:7.7%;Inoculum volume: 7.3%).The optimization models DNN-EY-GA and XGB-EY-GA were optimized for the yield and efficacy of ethanol fermentation,demonstrating the efficiency and applicability of the method.From the high-throughput sequencing analysis of the fermentation broth before and after optimization,it was shown that the highest percentage of Bacilli microorganisms in the Bacillus phylum and the optimization significantly reduced the abundance of miscellaneous bacteria.In this study,highly accurate prediction models for enzymatic polysaccharide and ethanol fermentation were developed based on the combination of multiple artificial intelligence techniques.The efficiency of polysaccharide production and ethanol production is effectively improved by the assistance of model interpretation techniques and optimization algorithms to obtain a combination of lower cost production schemes for efficient utilization of corn straw,which provided an artificial intelligence reference method for multiple resource utilization directions of corn straw and has important reference significance for improving the utilization efficiency of agricultural waste.