Improve PID controller through reinforcement learning

Reinforcement learning is a mathematical framework for agents to interact intelligently with their environment. Unlike supervised learning, where a system learns with the help of labeled data, reinforcement learning agents learn how to act by trial and error only receiving a reward signal from their environments. A field where reinforcement learning has been prominently successful is robotics [3]. However, real-world control problems are also particularly challenging because of the noise and high- dimensionality of input data (e.g., visual input). In recent years, in the field of supervised learning, deep neural networks have been successfully used to extract meaning from this kind of data. Building on these advances, deep reinforcement learning was used to solve complex problems like Atari games and Go. Mnih et al. [1] built a system with fixed hyper parameters able to learn to play 49 different Atari games only from raw pixel inputs. However, in order to apply the same methods to real-world control problems, deep reinforcement learning has to be able to deal with continuous action spaces. Discretizing continuous action spaces would scale poorly, since the number of discrete actions grows exponentially with the dimensionality of the action. Furthermore, having a parametrized policy can be advantageous because it can generalize in the action space. Therefore with this thesis we study state-of-the-art deep reinforcement learning algorithm, Deep Deterministic Policy Gradients. We provide a theoretical comparison to other popular methods, an evaluation of its performance, identify its limitations and investigate future directions of research. The remainder of the thesis is organized as follows. We start by introducing the field of interest, machine learning, focusing our attention of deep learning and reinforcement learning. We continue by describing in details the two main algorithms, core of this study, namely Deep Q-Network (DQN) and Deep Deterministic Policy Gradients (DDPG). We then provide implementatory details of DDPG and our test environment, followed by a description of benchmark test cases. Finally, we discuss the results of our evaluation, identifying limitations of the current approach and proposing future avenues of research.

Using Deep Reinforcement Learning (DRL) algorithm to deal with autonomous driving tasks usually have unsatisfied performance due to lack of robustness and means to escape local optimum. In this article, we designs a Survival-Oriented Reinforcement Learning (SORL) model that tackle these problems by setting survival rather than maximize total reward as first priority. In SORL model, we model autonomous driving task as Constrained Markov Decision Process (CMDP) and introduce Negative-Avoidance Function to learn from previous failure. The SORL model greatly speed up the training process and improve the robustness of normal Deep Reinforcement Learning algorithm.

<span>The utilization of games in training the reinforcement learning (RL) agent is to describe the complex and high-dimensional real-world data. By utilizing games, RL researchers will be able to evade high experimental costs in training an agent to do intelligence tasks. The objective of this research is to generate intelligent agent behaviors in multi-agent game artificial intelligence (AI) using deep reinforcement learning (DRL) algorithm. A basic RL algorithm called deep Q network is chosen to be implemented. The agent is trained by the environment's raw pixel images and the action list information. The experiments conducted by using this algorithm show the agent’s decision-making ability in choosing a favorable action. In the default setting for the algorithm, the training is set into 1 epoch and 0.0025 learning rate. The number of training iterations is set to one as the training function will be repeatedly called for every 4-timestep. However, the author also experimented with two different scenarios in training the agent and compared the results. The experimental findings demonstrate that our agents learn correctly and successfully while actively participating in the game in real time. Additionally, our agent can quickly adjust against a different enemy on a varied map because of the observed knowledge from prior training.</span>

Much human and computational effort has aimed to improve how deep reinforcement learning (DRL) algorithms perform on benchmarks such as the Atari Learning Environment. Comparatively less effort has focused on understanding what has been learned by such methods, and investigating and comparing the representations learned by different families of DRL algorithms. Sources of friction include the onerous computational requirements, and general logistical and architectural complications for running DRL algorithms at scale. We lessen this friction, by (1) training several algorithms at scale and releasing trained models, (2) integrating with a previous DRL model release, and (3) releasing code that makes it easy for anyone to load, visualize, and analyze such models. This paper introduces the Atari Zoo framework, which contains models trained across benchmark Atari games, in an easy-to-use format, as well as code that implements common modes of analysis and connects such models to a popular neural network visualization library. Further, to demonstrate the potential of this dataset and software package, we show initial quantitative and qualitative comparisons between the performance and representations of several DRL algorithms, highlighting interesting and previously unknown distinctions between them.

Multi-agent reinforcement learning dealing with hybrid action spaces: A case study for off-grid oriented renewable building energy system

Reinforcement Learning (RL) has become one of the most powerful methods in Artificial Intelligence (AI) which allows a system to learn as it is engaged in an interaction with its environment and to maximize its actions by performing a trial and error process. Although classical RL algorithms can give us a fundamental idea on RL, like the Q-Learning, its usefulness wanes in high-dimensional or continuous-state spaces. To overcome these shortcomings, Deep Reinforcement Learning (DRL) algorithms including Deep Q-Networks (DQN) build on neural networks to learn action-value functions, allowing horizontal scalability and more efficient learning. This paper reports on the experimental comparison of Q-Learning and DQN based on the environment CartPole-v1. Both of the algorithms were trained in the controlled environment and tested under various metrics that include cumulative rewards, success rate, convergence speed, safety (failure rate), and path efficiency. Findings indicate that Q-Learning moves towards moderate stability with slower convergence but DQN takes shorter learning time, is more reliable and performs better. Such results indicate the ground-breaking potential of the DRL in the progress of AI toward transcending classical RL restrictions. Future research topics encompass implementation of more challenging experimental settings, a test of other DRL algorithms (e.g., PPO, SAC), and generalization, scalability and safety in real-world environments.

Problem definition: The beer game is widely used in supply chain management classes to demonstrate the bullwhip effect and the importance of supply chain coordination. The game is a decentralized, multiagent, cooperative problem that can be modeled as a serial supply chain network in which agents choose order quantities while cooperatively attempting to minimize the network’s total cost, although each agent only observes local information. Academic/practical relevance: Under some conditions, a base-stock replenishment policy is optimal. However, in a decentralized supply chain in which some agents act irrationally, there is no known optimal policy for an agent wishing to act optimally. Methodology: We propose a deep reinforcement learning (RL) algorithm to play the beer game. Our algorithm makes no assumptions about costs or other settings. As with any deep RL algorithm, training is computationally intensive, but once trained, the algorithm executes in real time. We propose a transfer-learning approach so that training performed for one agent can be adapted quickly for other agents and settings. Results: When playing with teammates who follow a base-stock policy, our algorithm obtains near-optimal order quantities. More important, it performs significantly better than a base-stock policy when other agents use a more realistic model of human ordering behavior. We observe similar results using a real-world data set. Sensitivity analysis shows that a trained model is robust to changes in the cost coefficients. Finally, applying transfer learning reduces the training time by one order of magnitude. Managerial implications: This paper shows how artificial intelligence can be applied to inventory optimization. Our approach can be extended to other supply chain optimization problems, especially those in which supply chain partners act in irrational or unpredictable ways. Our RL agent has been integrated into a new online beer game, which has been played more than 17,000 times by more than 4,000 people.

The online 3D bin packing problem(3D-BPP) is widely used in the logistics industry and is of great practical significance for promoting the intelligent transformation of the industry. The heuristic algorithm relies too much on manual experience to formulate more perfect packing rules. In recent years, many scholars solve 3D-BPP via deep reinforcement learning(DRL) algorithms. However, they ignore many skills used in manual packing, one of the most important skill is workers put the item aside if the item is packed improperly. Inspired by this skill, we propose a DRL algorithm with a buffer zone. Firstly, we define the wasted space and the buffer zone. And then, we integrate them into the DRL algorithm framework. Importantly, we compare the bin utilization with di erent thresholds of wasted space and di erent buffer zone sizes. Experimental results show that our algorithm outperforms existing heuristic algorithms and DRL algorithms.

As vehicular technology advances, intelligent vehicles generate numerous computation‐intensive tasks, challenging the computational resources of both the vehicles and the Internet of Vehicles (IoV). Traditional IoV struggles with fixed network structures and limited scalability, unable to meet the growing computational demands and next‐generation mobile communication technologies. In congested areas, near‐end Mobile Edge Computing (MEC) resources are often overtaxed, while far‐end MEC servers are underused, resulting in poor service quality. A novel network framework utilizing sixth‐generation mobile communication (6G) and digital twin technologies, combined with task migration, promises to alleviate these inefficiencies. To address these challenges, a task migration and re‐offloading model based on task attribute classification is introduced, employing a hybrid deep reinforcement learning (DRL) algorithm—Dueling Double Q Network DDPG (QDPG). This algorithm merges the strengths of the Deep Deterministic Policy Gradient (DDPG) and the Dueling Double Deep Q‐Network (D3QN), effectively handling continuous and discrete action domains to optimize task migration and re‐offloading in IoV. The inclusion of the Mini Batch K‐Means algorithm enhances learning efficiency and optimization in the DRL algorithm. Experimental results show that QDPG significantly boosts task efficiency and computational performance, providing a robust solution for resource allocation in IoV.

Path planning refers to that the mobile robot can obtain the surrounding environment information and its own state information through the sensor carried by itself, which can avoid obstacles and move towards the target point. Deep reinforcement learning consists of two parts: reinforcement learning and deep learning, mainly used to deal with perception and decision-making problems, has become an important research branch in the field of artificial intelligence. This paper first introduces the basic knowledge of deep learning and reinforcement learning. Then, the research status of deep reinforcement learning algorithm based on value function and strategy gradient in path planning is described, and the application research of deep reinforcement learning in computer game, video game and autonomous navigation is described. Finally, I made a brief summary and outlook on the algorithms and applications of deep reinforcement learning.

In recent years, deep reinforcement learning (DRL) has made impressive achievements in many fields. However, existing DRL algorithms usually require a large amount of exploration to obtain a good action policy. In addition, in many complex situations, the reward function cannot be well designed to meet task requirements. These two problems will make it difficult for DRL to learn a good action policy within a relatively short period. The use of expert data can provide effective guidance and avoid unnecessary exploration. This study proposes a deep imitation reinforcement learning (DIRL) algorithm that uses a certain amount of expert demonstration data to speed up the training of DRL. In the proposed method, the learning agent imitates the expert's action policy by learning from demonstration data. After imitation learning, DRL is used to optimise the action policy in a self‐learning way. By experimental comparison on a video game called the Mario racing game, it is shown that the proposed DIRL algorithm with expert demonstration data can obtain much better performance than previous DRL algorithms without expert guidance.

The continuously changing nature of stock markets poses a non-trivial challenge to build automated trading agents that timely assist traders in their decision makings. Fixed sets of trading rules and offline pre-trained models are inefficient to adapt regarding the real-time fluctuations on the stock markets. Deep Reinforcement Learning (DRL) algorithms train autonomous agents that could potentially address such highly dynamic environments by integrating the generalization power of artificial neural networks with the online learning through experiences gained in the interactions with the environment. We investigate the capability of three DRL algorithms, namely Deep Deterministic Policy Gradient (DDPG), Twin Delayed DDPG (TD3), and Soft Actor-Critic (SAC), for tackling the automated stock trading problem. We employ two datasets, from the US and Vietnam markets, with different characteristics and market trends, on which the DRL-trained agents are compared against a random trading agent. Experimental results indicate the potentials of these DRL algorithms and also exhibit their pitfalls when being applied in datasets where the majority of the stocks are not up-trending. We propose a simple, but effective, technique to assist the agents to minimize their losses.

This paper addresses a dynamic job shop scheduling problem (DJSSP) based on a real stone paper factory, incorporating parallel machines, order rejection, tardiness costs, machine deterioration, and inventory costs. Orders arrive at variable times, with some potentially rejected due to capacity constraints. Also, raw materials must be purchased at appropriate times and quantities. The model optimizes three objectives: profit maximization, inventory cost variance minimization, and machine workload variance minimization. A heuristic solution method based on a double-network deep reinforcement learning (DRL) algorithm, complemented by an artificial neural network for inventory management, was developed. Evaluation against public benchmarks and other DRL algorithms highlights the impact of inventory cost variance and workload balancing. Real-case evaluation demonstrated that multi-objective optimization reduced inventory cost variance by 9.8%, machine workload variance by 21.87%, and raw material holdings by 21.7%, while slightly decreasing profit (3.2%) and slightly increasing purchase frequency (4.7%). Sensitivity analysis highlighted the individual effects of each objective on profit, workload, and makespan, confirming the practical benefits of the proposed approach.

In this paper, we are interested in symbiotic radio networks (SRNs), in which an Internet-of-Things (IoT) network parasitizes in a primary cellular network to achieve spectrum-, energy-, and infrastructure-efficient communications. Each IoT device transmits its own information by backscattering the signals from the primary network without using active radio-frequency (RF) transmitter chain. We consider the symbiosis between the cellular network and the IoT network and focus on the user association problem in SRN. Specifically, the base station (BS) in the primary network serves multiple cellular users using time division multiple access (TDMA) and each IoT device is associated with one cellular user for information transmission. The objective of user association is to link each IoT device to an appropriate cellular user by maximizing the sum rate of all IoT devices. However, the difficulty in obtaining the full real-time channel information makes it difficult to design an optimal policy for this problem. To overcome this issue, we propose two deep reinforcement learning (DRL) algorithms, both use historical information to infer the current information in order to make appropriate decisions. One algorithm, referred to as centralized DRL, makes decisions for all IoT devices at one time with globally available information. The other algorithm, referred to as distributed DRL, makes a decision only for one IoT device at one time using locally available information. Finally, simulation results show that the two proposed DRL algorithms achieve performance comparable to the optimal user association policy which requires perfect real-time information, and the distributed DRL algorithm has the advantage of scalability.

In a deregulated environment, the renewable energy producers will face the challenge of how to increase their revenues under uncertainties of power generation and time-varying electricity price. In traditional power network scheduling, prediction and optimization are two independent processes, which easily leads to information loss and modeling error. To deal with the uncertainty and realize an end-to-end controller, this article proposes an energy storage system control model (ESSCM) in the scene of the combined wind-solar storage system. The proposed ESSCM using deep reinforcement learning (DRL) algorithm is trained by interacting with the massive environment of a power grid without requiring the assumption on the uncertainties. It learns from scratch to realize the coordination operation of with wind power and photovoltaic power in a combined system, further maximize the benefits of the combined system in the electricity market. One state-of-the-art DRL algorithm, namely double deep Q-network, is used to formulate the proposed ESSCM. Numerical results illustrate that the proposed approach can effectively accommodate the uncertainty and bring high revenues to the combined system.