云环境下基于AI知识分析的负载均衡方法

Load Balancing Based on AI Knowledge Analysis in Cloud Environment

摘要: 针对云计算环境下传统负载均衡方法难以对蕴含在历史数据中的流量信息进行知识分析与智能利用问题，提出一种基于AI知识分析的云平台负载智能均衡方法LSTM-TD3 (Long Short Term Memory and Twin Delayed Deep Deterministic policy gradient algorithm)。LSTM-TD3以资源利用率、服务器连接数、响应时间、请求历史数据等为知识参数输入，首先通过LSTM进行任务数预测建模；紧接着以初始化预测数据强化学习模型TD3，形成基于AI分析与评估的优化负载均衡计算模型；最后通过实际训练以及仿真实验对LSTM-TD3的负载均衡效果进行验证测试。实验结果表明，相比传统的无负载均衡、轮询算法、Q-learning和TD3算法等云环境负载均衡方法，LSTM-TD3负载均衡性能分别提高25.4%，6.41%，3.56%和2.85%，能达到更好的资源负载均衡效果，资源利用率更高。

Abstract: In view of the difficulty of traditional load balancing methods in intelligent utilization of traffic information contained in historical data, a cloud load balancing method LSTM-TD3 (Long Short Term Memory and Twin Delayed Deep Deterministic policy gradient algorithm) based on AI knowledge analysis is proposed. LSTM-TD3 takes resource utilization, number of server connections, response time, request history data, etc. as knowledge parameters input. First, it uses LSTM to predict and model the number of tasks; then, the initialization prediction data is used to strengthen the learning model TD3 to form an optimized load balancing calculation model based on AI analysis and evaluation; finally, the load balancing effect of lSTM-TD3 is verified and tested through actual training and simulation experiments. Experimental results show that compared with traditional cloud environment load balancing methods such as no load balancing, polling algorithm, Q-learning and TD3 algorithm, the performance of LSTM-TD3 load balancing is improved by 25.4%, 6.41%, 3.56% and 2.85% respectively, which can achieve better resource load balancing effect and higher resource utilization.