Adaptive Resource Scheduling for IoT Cloud Systems: An LSTM-Based Load Balancing Approach

In order to solve the problem of resource allocation imbalance caused by sudden workloadin IoT cloud platforms, this paper proposes a load balancing method that combines oneline Long Short-Term Memory (LSTM) network prediction and dynamic thresholding. This method constructs a time model by analyzing the historical load characterisitcs of node resource utilization. It introduces the Boltzmann strategy to convert the deviation between the predicted workeload and the target threshold into a probability distribution, achieving a balance mechanism between exploration and mining. This ensures continuous system resource optimization while maintaining service quality. The simulation results show that the proposed method is slightly superior to traditional polling algorithms in terms of resource utilization and reducing task failure rates, even in the case of fluctuating task arrival rates and sudden load peaks. Under sustained high load conditions, it reduced the average response time by about 0.65 units, lowered the failure rate by about 3.7%, and increased resource utilization by about 2%.