T/JSIA 0003-2020 Active Group standards

T/JSIA 0003-2020 Intelligent decision-making system standards based on reinforcement learning

Publish Date: 2020-02-28 Implement Date: 2020-02-28 For services related to genuine standard inquiry, procurement, translation, and other related services in China, please Contact Us

Price: $ 20.00

Out of stock

Basic Information

Standard Code: T/JSIA 0003-2020

Standard Type: Group standards

Standard Status: Active

is_force_gb: no

CCS Name: Security and Alarm System

ICS Name: Comprehensive Application of Information Technology

Publish Date: 2020-02-28

Implement Date: 2020-02-28

Scope

Main Technical Content: 1. Key Requirements 1.1 Environmental Requirements - Organize different application scenarios and provide a simulation environment with a universal interface design, allowing for the comparison of different reinforcement learning algorithms. Test the actual system through the simulation environment. Specifically, this includes: - How to use the environment - A unified custom interface specification for the environment 1.2 Observation - Reinforcement learning is typically implemented as a "agent-environment" loop. During interaction, the agent selects an action, and the environment returns an observation and reward value. 1.3 Space - Space is used to describe valid actions and observations, which are attributes of the environment. When interacting with the environment, the agent can directly interact based on the interface provided by the system. 1.4 Strategy - The input of the strategy is the state and action, and the output is the probability of taking the input action under the input state. 2. Run Design 2.1 Environment Configuration - Environment configuration mainly configures the basic environmental information of the initial state, distinguishes between different states for learning and exploration, and improves the experimental results of reinforcement learning algorithms. 2.2 Visualization Display - Real-time visualization of relevant environmental information of the current scene, which is convenient for developers to observe the effect and conduct corresponding optimization of the algorithm. 2.3 Interaction Norms - The interaction between the agent and the environment can be divided into two categories: human-computer interaction and automatic interaction. - a) Human-computer interaction: People interact with the system according to the corresponding interaction methods designed by the system, while the machine directly generates a corresponding strategy based on intelligent algorith