diff --git a/Beijing_air_quality_prediction/README.md b/Beijing_air_quality_prediction/README.md
new file mode 100644
index 0000000..1cb805a
--- /dev/null
+++ b/Beijing_air_quality_prediction/README.md
@@ -0,0 +1,13 @@
+### 3. SARIMA时间序列预测模型
+
+#### 建模思路
+1. **数据特性分析**  
+   - 通过ADF检验验证序列平稳性（p值=0.03，通过5%显著性水平）
+   - 季节分解显示存在周周期性（s=7）和长期趋势
+   - ACF/PACF分析确定初始参数范围
+
+2. **参数选择**  
+   采用网格搜索确定最优参数组合：
+   ```python
+   final_order = (1, 1, 1)          # 非季节性参数
+   seasonal_order = (1, 1, 1, 7)    # 周周期季节性
diff --git a/air_quality_prediction.ipynb b/air_quality_prediction.ipynb
index 9571e6f..29a6b88 100644
--- a/air_quality_prediction.ipynb
+++ b/air_quality_prediction.ipynb
@@ -532,8 +532,8 @@
   {
    "metadata": {
     "ExecuteTime": {
-     "end_time": "2025-03-24T07:17:24.611373Z",
-     "start_time": "2025-03-24T07:17:22.170632Z"
+     "end_time": "2025-03-24T07:23:26.872703Z",
+     "start_time": "2025-03-24T07:23:24.709736Z"
     }
    },
    "cell_type": "code",
@@ -572,6 +572,7 @@
     "ax.set_xlabel('日期', fontsize=12)\n",
     "ax.set_ylabel('AQI数值', fontsize=12)\n",
     "plt.legend()\n",
+    "plt.savefig('./images/AQI.png', dpi=200, bbox_inches='tight')\n",
     "plt.show()\n",
     "\n",
     "# 计算拟合度指标\n",
@@ -594,13 +595,6 @@
    ],
    "id": "24996a0c06820cdc",
    "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "<class 'pandas.core.indexes.datetimes.DatetimeIndex'>\n"
-     ]
-    },
     {
      "data": {
       "text/plain": [
@@ -622,7 +616,7 @@
      ]
     }
    ],
-   "execution_count": 12
+   "execution_count": 13
   },
   {
    "metadata": {},
diff --git a/images/AQI.png b/images/AQI.png
new file mode 100644
index 0000000..62b3118
Binary files /dev/null and b/images/AQI.png differ
diff --git a/readme.md b/readme.md
index 9cacd4f..21daa94 100644
--- a/readme.md
+++ b/readme.md
@@ -62,6 +62,8 @@
 
 ## 题目3
 
+### SARIMA模型解读
+
 ### XGBOOST模型解读
 1. 该模型使用历史AQI数据，并进行周期性编码和滞后特征构建（3小时粒度的滞后特征（最多7天）），作为特征工程。
 2. 每次预测均采用该时间点以前的真实数据，即每次预测均为单步预测。