Simulation data and results for: Dynamic Satellite Communication Optimization via Multi-Satellite Cooperative SNR Prediction

Description

# Multi-Satellite Cooperative SNR Prediction — Dataset and Results # 多卫星协同 SNR 预测 —— 数据集与实验结果 This repository contains the simulation data and experimental results supporting the paper: **"Dynamic Satellite Communication Optimization via Multi-Satellite Cooperative SNR Prediction"** (submitted to *Computer Communications*, Elsevier). 本数据仓库存放支撑上述论文的仿真数据与实验结果。 --- ## 1. Directory Structure / 目录结构 ``` ├── dataset/ │ ├── predictions/ # 自适应运行各阶段的预测结果 (28 个 .npz) │ │ └── pred_stepwise_u10_steps2999_comboedf803fa.npz # ★ 论文主实验最终结果 │ │ (y_true, y_pred: 2999 步目标通信卫星 SNR 真值与预测值) │ └── adaptive_run_metrics/ # 自适应过程性能记录 (initialize + 7 个更新点) │ └── */metrics_summary.json # 各更新点的 overall loss/MAE/RMSE、所选卫星组合 │ ├── ablation/ # 消融实验（对应论文 Fig. 11 与 4.3 节表格） │ ├── ModelFusion/seed_0..4/ # 完整 CNN-LSTM-Attention 模型 × 5 个随机种子 │ ├── ModelNoAttn/seed_0..4/ # 去注意力机制 │ ├── ModelNoCNN/seed_0..4/ # 去 CNN 前端 │ ├── ModelNoLSTM/seed_0..4/ # 去 LSTM（自适应平均时序池化替代） │ │ ├── metrics.json # test_RMSE / test_MAE / test_R2 / 参数量 / 训练时间 │ │ ├── predictions.npz # 测试集预测值与真值 (preds, targets) │ │ ├── loss_history.npz # 训练/验证损失曲线 (train_loss, val_loss) │ │ └── best_model.pth # 最优模型权重 (PyTorch) │ └── README_消融实验说明.md # 消融结果汇总（Mean±Std，与论文数值一致） │ ├── communication/ # 通信策略对比实验（对应论文 Fig. 12–14 与 Table 2） │ ├── all_runs_metrics.json # 五种策略的 BER / 吞吐量 / 包错率 / 平均传输次数 │ ├── comm_summary.md # 结果汇总说明 │ └── seed_0..4/comm_metrics.json │ └── code_optional/ # 【可选】复现代码：公开前可自行决定是否保留 ├── src/ # 环境仿真、CNN-LSTM-Attention 模型、组合评分器、自适应控制器 ├── run_ablation.py # 消融实验入口 ├── run_comm_test.py # 通信策略对比入口 ├── summarize_ablation.py # 消融结果汇总脚本 └── requirements.txt # Python 依赖 ``` ## 2. Data Formats / 数据格式 - `.npz`：NumPy 压缩格式，`numpy.load()` 读取。 - `pred_*.npz`：`y_true`（真实 SNR）、`y_pred`（预测 SNR）、`meta_json`（运行元数据）。 - `predictions.npz`：`preds`、`targets`（测试集，float32）。 - `loss_history.npz`：`train_loss`、`val_loss`（按 epoch）。 - `.json`：UTF-8 文本，键名自解释。 - `.pth`：PyTorch 模型权重，需配合 `code_optional/src/models/` 中的模型定义加载。 ## 3. Correspondence with the Paper / 与论文的对应关系 | 论文位置 | 数据文件 | |---|---| | Fig. 9（动态预测全过程）| `dataset/predictions/pred_stepwise_u10_steps2999_comboedf803fa.npz` | | 4.3 节 自适应更新记录 | `dataset/adaptive_run_metrics/` | | Fig. 10–11 + 消融表格 | `ablation/`（5 种子 Mean±Std，见 README_消融实验说明.md） | | Fig. 12–14 + Table 2 | `communication/all_runs_metrics.json` | ## 4. Notes / 说明 - 仿真环境参数、特征构造与训练超参数见论文 4.1 节与 2.2 节。 - `code_optional/` 为可选复现代码：若不希望公开源码，上传 Mendeley Data 前删除该目录即可， 数据与结果文件不受影响。 - All data are generated by simulation; no real satellite measurements are involved. 全部数据由仿真生成，不涉及真实卫星测量数据。 Contact / 联系方式: qiliangang@hrbeu.edu.cn (Corresponding author: Liangang Qi, Harbin Engineering University)

Steps to reproduce

All data were generated by simulation. To reproduce the results: 1. Install dependencies: pip install -r code_optional/requirements.txt 2. Ablation study (Fig. 10-11 and Table in Section 4.3): python code_optional/run_ablation.py Per-seed metrics, predictions, loss curves and model weights are written to ablation/{ModelName}/seed_{0-4}/. 3. Communication strategy comparison (Fig. 12-14 and Table 2): python code_optional/run_comm_test.py Aggregated metrics are written to communication/all_runs_metrics.json. 4. The main 3000-step adaptive run used in the paper is provided as dataset/predictions/pred_stepwise_u10_steps2999_comboedf803fa.npz (y_true / y_pred), with per-update performance records in dataset/adaptive_run_metrics/. If the code_optional/ folder is not included in this dataset, the published .npz and .json files alone are sufficient to verify all reported figures and tables; see README.md for the file-to-figure correspondence.

Institutions

• Harbin Engineering University

  Heilongjiang, Harbin

Categories

Licence