Purpose

This paper aims to deal with application of artificial intelligence for solving real time control complication adhered with the controlled operation of a buck power converter. This type of converter finds application for power conversion at various levels for the direct current-direct current power industry to step down the input voltage.

Design/methodology/approach

Use of ANN-RL (Artificial Neural Networks- Reinforcement Learning)-based control algorithm to control buck power converter shows robustness against parameter and load variation. Because of non-linearity instigated by element used for switching, control of this converter becomes an arduous control predicament. All the classical control techniques are based on an approximate linear model of the step down converter and these techniques fail to handle actual non-linearity.

Findings

In this paper, a reinforcement learning-based algorithm has been used to handle and control buck power converter output voltage, without approximating the model of converter. The non-linearity instigated in converter is subjected to state of switch. Model of buck power converter is defined as a multi-step decision problem so that it can be solved using mathematical model of Markov decision process (MDP) and, in turn, reinforcement learning can be implemented. As MDP model is available for a discrete state system so model of converter has to be discretized and then value iteration is applied and output is analyzed. Load regulation and integral time absolute error analysis is done to show efficacy of this technique.

Originality/value

To mitigate the effect of discretization function approximation using neural network is applied. MATrix LABoratory has been used for implementation and result indicates an improvement in the overall response.

中文翻译：

---

基于ANN RL算法的电力行业降压电源转换器电路的开发

目的

本文旨在处理人工智能的应用，以解决与降压电源转换器的受控操作相关的实时控制复杂性。这种类型的转换器适用于直流-直流电源行业的各种级别的电源转换，以降低输入电压。

设计/方法/方法

使用基于 ANN-RL（人工神经网络 - 强化学习）的控制算法来控制降压电源转换器显示出对参数和负载变化的鲁棒性。由于用于开关的元件引起非线性，该转换器的控制成为一个艰巨的控制困境。所有经典的控制技术都基于降压转换器的近似线性模型，并且这些技术无法处理实际的非线性。

发现

在本文中，基于强化学习的算法已被用于处理和控制降压电源转换器输出电压，而无需近似转换器模型。转换器中引发的非线性受到开关状态的影响。降压电源转换器模型被定义为一个多步决策问题，因此可以使用马尔可夫决策过程（MDP）的数学模型来解决，进而可以实现强化学习。由于 MDP 模型可用于离散状态系统，因此必须对转换器模型进行离散化，然后应用值迭代并分析输出。进行负载调节和积分时间绝对误差分析以显示该技术的功效。

原创性/价值

为了减轻使用神经网络的离散化函数逼近的影响。MATrix LABoratory 已用于实施，结果表明整体响应有所改善。