Inverse Model Optimization by Differential Evolution to improve Neural Predictive Control
Inverse Model Optimization by Differential Evolution to improve Neural Predictive Control
复制标题
DOI：
10.1109/scisisis50064.2020.9322702
复制
发表时间：
2020-12
期刊：
影响因子：
--
通讯作者：
Edgar Ademir Morales-Perez;H. Iba
中科院分区：
文献类型：
--
作者：
Edgar Ademir Morales-Perez;H. Iba
本文利用前馈控制方法中常见的逆模型控制信号增强了基于差分进化的启发式优化器，以改进基于神经网络的预测控制方案（NPC）。通过使用实验数据，我们训练LSTM神经网络作为NPC和逆逼近的模型预测器，这为优化器搜索最优控制序列提供了一个良好的起点。通过使用逆模型创建的初始种群初始化基于差分进化的优化器，我们实现了搜索空间缩减，从而更快地达到收敛。此外，我们的方法超越了传统技术，使用非线性控制基准进行了测试，证明了结果准确性的提高以及将预测控制与进化技术相结合的可行性。
In this paper a heuristic-optimizer based on Differential Evolution has been enhanced with Inverse-Model control signals, common in feedforward control methods, to improve a Neural Network-based Predictive control scheme (NPC). By using experimental data, we train LSTM Neural Networks as model predictors, for the NPC and for the Inverse approximation, that supply our optimizer with a good starting point for the search of optimal control sequences. We achieve a search space reduction where convergence is reached faster by initializing our differential evolution-based optimizer with an initial population created from the Inverse Model. Moreover, our method surpasses conventional techniques, tested using nonlinear control benchmarks, proving an increase in results accuracy and the feasibility of combining Predictive Control with Evolutionary techniques.