Design and experiments of the fertilization monitoring system based on the PSO-BP for sugarcane

Abstract: Sugarcane is mainly planted in hilly areas, such as the province of Guangxi and Yunnan, China. Time-varying and nonlinear working parameters can often be found in the sugarcane horizontal planters, due to the relatively complex and changeable operating conditions. A high failure rate of blockage can often occur in the fertilization mechanism in this case. Moreover, it is difficult to maintain the damage to the chain and drive shaft after the blockage. The performance of fertilization can also be reduced to destroy the transmission mechanism, because the wet and agglomerated fertilizer can be concurrently blocked in the fertilization mechanism of the sugarcane horizontal planter. Moreover, it is still lacking in the automatic control of clearing and blocking in the fertilization mechanism of mechanical transmission type. In this study, a fertilization monitoring system was proposed to carry out the electro-hydraulic transmission and control transformation of the fertilization mechanism. A set of fertilization and anti-blocking control system was constructed using Particle Swarm Optimization (PSO) - Back Propagation (BP) neural network prediction. The input parameters were taken as the pressure and speed of the fertilizing motor, as well as the amount of fertilizer in the fertilizer tank, whereas, the output was the working state (no load, normal, heavy load, and blocked) of the fertilizing mechanism. The BP neural network was used to establish the mapping relationship between the input and the output. The PSO was used to optimize the weights and thresholds of the BP. After that, the prediction accuracy increased from 97% to 99%, and the determination coefficient R2 increased from 0.977 5 to 0.982 9. The results showed that the PSO-optimized BP neural network presented a better prediction effect. The BP neural network optimized by the PSO was used to identify the fertilization state with higher accuracy. The PSO-optimized BP neural network was selected as the network model to predict the working state of fertilization. The control program of the single-chip microcomputer was written into the trained prediction model. The control system of the fertilizer application mechanism was designed, where the pressure transmitter was to collect the pressure value of the hydraulic motor, the Hall proximity switch was to collect the speed value of the screw shaft, and the photoelectric sensor was to monitor the fertilizer status in the fertilizer box in real time. The workshop test was carried out, where the test indicators were the accuracy rate to identify the response of the fertilization mechanism working state, and the probability of preventing blockage under heavy load. The results showed that: the accuracy rate of working state response recognition was 89% under the heavy load state. The control system was used to control the forward and reverse rotation of the fertilization motor. The probability was 87.5% for the removal of blockages. Therefore, the monitoring system with a neural network can be used to accurately identify the various working states of the fertilization mechanism during the field experiment. The heavy-load state of the fertilization mechanism can be accurately predicted by the monitoring system. The anti-blocking control command was executed without blockage failure. Anyway, the fertilization anti-clogging monitoring system can fully meet the working condition prediction and anti-clogging control requirements of the fertilization mechanism under complex and changeable working conditions. Consequently, the working condition monitoring and anti-blocking control system of the fertilization mechanism in the sugarcane planters can be expected to promote the high quality and efficiency of fertilization operations, in order to effectively reduce the blockage failure rate and the time of downtime for troubleshooting. This finding can also provide a new reference for the automation transformation of fertilization.