基于过程神经网络的热平衡温度预测研究

为缩短航天器热平衡试验周期，以降低航天器研制成本，提出了一种基于过程神经网络的热平衡温度预测模型。为简化该模型的学习过程，提出了一种基于正交基函数展开的基本学习算法，利用基函数的正交性不仅可以简化模型中的时间累积运算过程，而且能提高模型对解决实际问题的适应性。同时，为增强模型的外推预测能力，在基本学习算法的基础上给出了一种基于新增样本的学习算法，使模型既能对新增样本进行快速学习又不损失对原有样本的记忆。实际应用表明，该预测模型能够利用某型号卫星热平衡试验中某监测点进入稳定工况后40小时内的试验数据提前42．5—68小时获得该监测点的极限热平衡温度。

Thermal Equilibrium Temperature Prediction Based on Process Neural Network

In order to shorten the duration of the thermal balance test of the spacecraft to reduce the development cost of the spacecraft, a thermal equilibrium temperature prediction model based on the process neural network is proposed. To simplify the learning procedure of the proposed prediction model, a basic learning algorithm based on the expansion of the orthogonal basis functions is given. With the orthogonality of the orthogonal basis functions, the time aggregation operation in the proposed model can be simplified and the adaptability of the proposed model to the practical problem resolving can be raised. Furthermore, in order to reinforce the extrapolation capability of the proposed prediction model, a learning algorithm for new pattern based on the basic learning algorithm is developed, which can learn the new patterns quickly without degrading the recall of the old patterns. The application test results indicate that the proposed prediction model can utilize the first 40 hours stable test data in the thermal test on some monitoring point of some type satellite to obtain the ultimate thermal equilibrium temperature of the monitoring point about 42.5 to 68 hours in advance.