Data Center Simulation with Two-Stage State Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for simulating data center operations consume excessive computing resources due to the need to evaluate various control strategies, which is challenging given the limited computing resources and restricted time.
Innovation Solution
A method and apparatus that utilize a data center simulation model comprising a first state prediction model with low precision and a second state prediction model with high precision, integrated with a reinforcement learning algorithm, to optimize the simulation process by quickly screening actions and reducing unnecessary high-precision simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-precision simulation models are used to evaluate control strategies, then the accuracy of simulation results is improved, but the consumption of computing resources increases
Solution Approach 1:
The simulation model is segmented into two distinct components: a low-precision model for rapid evaluation and a high-precision model for accurate simulation. The low-precision model filters out obviously suboptimal control strategies, while the high-precision model is applied only to promising candidates, thus reducing overall computing resource consumption while maintaining simulation accuracy where needed.
Solution Approach 2:
Different precision levels are applied to different parts of the simulation process. The low-precision model is used for the majority of control strategy evaluations, while the high-precision model is selectively applied only to specific promising strategies. This local differentiation of quality allows the system to maintain high accuracy where necessary while minimizing overall computational overhead.
2Reliability
If comprehensive evaluation of various control strategies is performed, then the quality of optimal control strategy identification is improved, but the simulation time increases
Solution Approach 1:
The low-precision model performs preliminary evaluation of control strategies before they are subjected to high-precision simulation. This preliminary action filters out obviously suboptimal strategies, allowing the system to focus computational time on evaluating only the most promising control strategies with the high-precision model, thus reducing overall simulation time while maintaining evaluation quality.
Solution Approach 2:
The two-stage evaluation process ensures continuous useful action by maintaining a pipeline where the low-precision model continuously filters strategies while the high-precision model simultaneously evaluates promising candidates. This continuous operation avoids idle time and ensures that computational resources are always productively engaged in meaningful evaluation tasks.
3Use of energy by moving object
If low-precision models are used for rapid simulation, then the computing resource consumption is reduced, but the accuracy of simulation results deteriorates
Solution Approach 1:
The low-precision model serves as an intermediary between the control strategy space and the high-precision model. It performs initial filtering and screening, acting as a gateway that identifies which strategies warrant further detailed evaluation. This intermediary role allows the system to reduce computing resource consumption while ensuring that no potentially optimal strategies are incorrectly discarded due to low-precision limitations.
Data Source
AI summary
A method and apparatus for optimizing operation simulation of a data center are disclosed, and the method includes: constructing a data center simulation model; acquiring a state data set and an action data set of the data center; inputting the state data set and the action data set into a first state prediction model to obtain a next state data set; judging whether the next state data set meets a state safe judgment condition; if so, inputting the state data set and the action data set into a second state prediction model to obtain a next state data set; optimizing network parameters of a reinforcement learning algorithm; and determining an action data set corresponding to a real-time state data set of the data center by using the trained reinforcement learning algorithm.