Bivariate Fertilization Control Sequence Optimization Based on Relevance Vector Machine

Abstract:

The key technology to improve accuracy and uniformity of variable fertilization lies in reducing the response time and choosing the best control parameters. A bivariate control fertilization sequence decision-making method based on relevance vector machine was presented. The method took into account the interaction among the before-current-after, three adjacent fertilizing rates, the response time and working dead features of active roller length and speed control. An optimal control parameters decision-making was carried out using non-linear programming method. Finally, the problem of small fertilizing rate under big active roller length with small rotational speed was avoided. The impact of the fluted metering roller’s pulsation was reduced. In order to coordinate the limited computation power of the vehicle mounted computer and respond time of bivariate control, relevant vector machine was utilized to achieve a near-optimal real-time computing for fertilization control sequences. Laboratory test results showed that the average error was 4%, which was smaller than the original method, while this method was effective to avoid the extreme case that the motor could not drive the roller shaft with improper parameters. The field test showed this kind of applicator could response to the instruction in time, and it had good accuracy and consistency, which proved its practicality.