A Study Of Mixture Density Network In Adjusting Sub-reflector Surface Antennas

High precision sub-reflector surface adjusting mechanism, mainly used in the big caliber reflector antenna of high frequency band. In order to meet the sub-reflector surface adjusting large caliber antenna technology development trend, especially for antenna system in the process of operation, the sub-reflector surface need to track movement of the high precision of antenna system according to certain requirements at the same time. Sub-reflector antenna position of the precise adjustment is one of the key technologies to ensure the big caliber reflector antenna high pointing accuracy, with the in-depth study of parallel mechanism, the application of parallel mechanism is becoming more and more widely, so the object of this article is the six degree of freedom parallel mechanism using in the sub-reflector surface, to improve the pointing accuracy.Six degrees of freedom parallel mechanism has all kinds of error in operation process,so the operation data is multi-value. According to the characteristics of the six degree of freedom parallel mechanism, in order to meet the requirement for high precision operation of six degree of freedom parallel mechanism, the paper combines the mixture density network algorithm, the mixture density network algorithm is applied to the parallel mechanism, solve the problem of the multi-value effectively, improved the precision in operation process.The author’s major contributions are outlined as follows:1. Refer to a large number of relevant literature about six degrees of freedom parallel mechanism structure, understand the development and application of it, analyzed the cause of all kinds of error in the six degree of freedom parallel mechanism. Due to various errors identified in six degree of freedom parallel mechanism, the uncertainty will lead to multi-value problem in actual operation. It proves the rationality of the mixture density network applications on this object. Introduces forward and inverse kinematics solution algorithms, using them to produce a large number of platform position and leg quantity simulation data so that using in the simulation.2. Using the simulation data of leg flexible quantity and leg compensation quantity as the input and output of the network, training the traditional neural network and mixturedensity network respectively, compared the training results, the results indicate that mixture density network could be more close to the ideal output in the simulation data,thus prove that the advantage of mixture density network. Using the forward kinematics solution algorithms, convert the mixture density network output compensation quantity to the platform position, further reflect the application of mixture density network, the platform can reach more accurate position.3. Develop a measurement software which can adjust the parallel mechanism based on Delphi 7.0, this software is easy to use, it can not only be satisfied with the requirement of high precision, but also be friendly to human-computer interactions.