Comprehensive insights into the application strategy of kitchen waste derived hydrochar: Random forest-based modelling

Abstract

Due to the complex and variable composition of kitchen waste (KW), it is difficult to master the product properties after hydrothermal carbonization treatment. In this study, random forest method was adopted to firstly predict the application performance of hydrochar from the perspectives of fuel, soil amendment and adsorbent. A total of 9 correlation models were established, with organic components and reaction parameters taken as input variables. The training set was supported by 132 literature data and 153 experimental data, with 18 experimental data as testing set. Results showed that the correlation coefficients of 9 models were up to 0.89–0.99. The relative significance of input variables was also acquired by mean decrease impurity method. It was found that the effects of reaction temperature/time on hydrochar properties were relatively negligible. Thus, hydrochar properties can be generally predicted based on the composition of KW feedstocks. The high heating value of hydrochar was mainly in the range of 20–30 MJ/kg, which was comparable to that of coal. To obtain hydrochar with good fuel performance, KW with high lipid (>50%) and low carbohydrate (<25%) is suitable. The pH value and cation exchange capacity of hydrochar was 3.23–5.69 and 1.08–42.37 cmol+/kg, respectively, which can be used for the improvement of saline-alkali land. KW of high protein (>35%), high lipid (>25%) and low carbohydrate (<35%) was potential to be converted into excellent adsorbents with specific surface area over 2000 m2/g.