ABSTRACT
In this paper, four feeding treatments including continuous feeding (SR00), recycling of 2 days starvation and 2 days refeeding (SR22), recycling of 7 days starvation and 2 days refeeding (SR72), and recycling of 7 days starvation and 7 days refeeding (SR77) were designed, and the feeding treatments were quantified as two treatment factors, i. e., starvation stress (SS) and starvation frequency (CF). Combining these two factors with the factors dry matter feed intake (FI), body weight (BW), water temperature (TE), water salinity (SA), water pH (PH) and growth time (GT), three BP artificial neural networks were constructed to predict the weight gain (WG), specific growth rate (SGR), and feed conversion ratio (FCR) of Lateolabrax japonicus, respectively. The results showed that the WG, SGR and FCR of L. japonicus were significantly affected by different feeding treatments. Throughout a 8-week trial, the WG and SGR of starved fish couldn't catch up to those of control fish. Except for SR72 group whose FCR was markedly higher than that of control group, no differences in FCR were observed between control group and experimental groups SR22 and SR77. The study also indicated that artificial neural network could well predict WG and SGR, but was unavailable for FCR. Among the eight factors, FI, SS, CF and GT had significant contributions to both WG and SGR. Furthermore, WG and SGR were predominantly dependent on FI and SS, respectively. Based on 4999 randomizations, the contribution rate of the treatment factors (including related FI) to WG and SGR was 64.9% and 79.7% , respectively.
