给deepseek提示词，让其通过风场数据矩阵识别台风

Question

既要调取deepseek接口，让deepseek能通过“风向场”和“风速场”的矩阵数据来识别台风，还让deepseek自己写了一段提示词，但识最后别的结果不好

提示词部分：

def build_analysis_prompt(self, region_data, point):
        """构建单个区域的提示词"""
        prompt = f"""台风特征分析指令

核心原则：
1. 严格识别北半球台风特有的逆时针气旋系统
2. 逆时针旋转是判定的绝对必要条件（优先级高于风速）
3. 任何无明显旋转特征的系统必须排除

数据说明：
- 矩阵方位规定：
    • 矩阵行方向：上→下 对应 北→南
    • 矩阵列方向：左→右 对应 西→东
- 风向矩阵中，角度表示风的去向(风向指向的方向)
    0°=北, 45°=东北, 90°=东, 135°=东南, 
    180°=南, 225°=西南, 270°=西, 315°=西北
- 北半球台风必须显示连续逆时针旋转模式
- 无效值: 风速NaN / 风向-999

核心分析要求
1. 风场连续旋转数学模型：
   - 设风眼中心为坐标原点O
   - 对任意位置点P(x,y)，计算：
        θ = arctan2(y,x)  // 位置方位角（正东为0°，逆时针增加）
        φ = 风向角（取自矩阵值）
   - 理想台风应满足：φ ≈ θ + 90°（严格切向逆时针旋转）
   - 允许偏差δ：|φ - (θ + 90°)| ≤ 30°（眼墙区≤15°，外围区≤30°）

2. 旋转连续性验证：
   a. 存在风向突变区（奇点）：
        • 寻找满足 max(Δφ)≥60° 的局部区域
        • 该区域应为中心低风速区（候选风眼）
   b. 围绕奇点的连续旋转：
        • 从奇点向外，方位角θ变化时，风向角φ应同步连续增加
        • 计算旋转一致性：∂φ/∂θ ≈ 1（理想值）
        • 可视化：想象从风眼发出的射线，沿线风向应平滑逆时针偏转

3. 风速分布要求：
   - 风向突变区对应最低风速区（V<5m/s）
   - 周围存在≥13.9m/s的风速带（允许不连续环状）
   - 风速梯度：从中心到外围≥5m/s

边界案例处理
1. 初生台风：
   - 接受∂φ/∂θ≈0.8-1.2的弱旋转
   - 低风速区可不连续
   - 最大风速≥13.9m/s即有效
   
2. 数据限制：
   - NaN区域不参与旋转计算
   - 有效数据点≥60%即可评估
   - 局部突变不影响整体模式

矩阵数据：
[风速矩阵(m/s)] (nan是无效值)
{np.array2string(region_data['speed'], threshold=np.inf, separator=', ', formatter={'float_kind': lambda x: "%.1f" % x})}

[风向矩阵(°)] (-999是无效值)
{np.array2string(region_data['dir'], threshold=np.inf, formatter={'int': lambda x: "%d" % x})}

必须严格遵循以下输出格式
判定: [是/否]
最大风速: [值]m/s
环形结构: [有/无]
风场旋转: [符合/不符合]
理由:
1. 风场数学模型分析: [φ与θ+90°的匹配度, 最大偏差位置]
2. 旋转连续性验证: [奇点识别, ∂φ/∂θ空间分布]
3. 风速-风向关联: [低风速区与旋转中心对应关系]
4. 综合结论: [100字以上判断依据]
"""
        return prompt

风向矩阵数据（-999为无效值）：

-999,-999,-999,240,242,247,260,270,277,280,277,277,277,280,282,282,282,285,285,287,290,290,290,292,292,295
-999,-999,-999,237,240,247,252,260,272,275,270,272,275,277,280,280,282,282,285,287,287,290,290,290,292,295
-999,-999,-999,252,247,247,-999,-999,-999,-999,267,270,275,275,277,277,282,282,285,287,287,287,287,290,292,292
-999,-999,-999,250,245,247,-999,-999,-999,-999,265,267,270,272,275,277,280,282,282,285,285,287,287,287,290,292
-999,-999,-999,247,245,252,-999,-999,-999,-999,262,265,270,272,275,277,277,280,282,282,285,285,287,287,287,290
-999,-999,-999,242,240,-999,-999,-999,-999,260,262,262,267,272,275,275,277,280,280,282,285,285,285,285,287,290
-999,-999,-999,237,242,-999,-999,-999,-999,-999,257,262,267,270,272,275,277,277,280,282,285,287,285,285,287,290
-999,-999,-999,240,237,-999,-999,-999,-999,250,255,262,267,270,272,277,277,277,280,285,285,287,287,287,287,290
-999,-999,-999,230,-999,-999,-999,-999,-999,245,252,260,267,270,275,277,277,280,282,282,285,287,290,290,290,290
-999,-999,-999,232,-999,-999,-999,-999,230,237,250,255,260,270,275,277,282,282,285,285,287,287,290,290,292,290
-999,-999,-999,225,215,-999,-999,-999,220,232,242,252,260,267,277,280,285,287,287,290,290,290,290,292,292,290
-999,-999,-999,207,200,202,210,210,207,222,235,247,257,272,280,285,290,287,290,290,292,295,292,292,292,292
-999,-999,-999,195,185,187,195,197,190,205,230,252,265,275,282,290,292,292,292,295,295,295,295,292,292,295
-999,-999,-999,185,157,172,180,180,177,185,210,255,277,292,302,305,297,297,297,295,297,297,297,295,292,295
-999,-999,-999,172,152,150,165,172,165,152,185,217,297,315,322,317,312,307,302,297,297,300,300,300,297,297
-999,-999,-999,160,152,137,142,155,155,137,127,115,27,352,340,332,325,317,315,310,300,305,305,305,302,302
-999,-999,-999,162,150,137,130,135,137,127,110,95,50,17,357,345,335,325,322,317,315,310,312,307,305,305
-999,-999,-999,162,147,135,125,122,122,117,105,90,67,40,17,0,342,332,325,322,322,317,320,315,307,305
-999,-999,-999,162,150,132,117,110,107,105,97,82,62,45,27,12,355,345,330,327,325,320,320,317,312,307
-999,-999,-999,142,140,130,117,105,97,90,82,72,60,45,30,15,7,350,335,330,327,325,325,317,315,312
-999,-999,-999,125,117,122,112,100,90,80,72,65,55,40,30,15,5,352,337,332,330,327,327,322,317,317
-999,-999,77,100,102,100,100,87,80,70,62,55,45,37,25,10,355,350,340,332,327,327,330,327,325,322
-999,-999,72,75,85,85,85,77,70,62,55,47,42,37,22,7,352,347,347,337,330,330,335,332,330,327
-999,-999,-999,65,65,67,67,70,65,57,52,47,40,32,20,2,350,345,345,340,335,332,332,332,335,332
-999,-999,-999,55,60,62,57,55,55,52,50,47,40,27,12,5,355,350,350,345,345,335,332,332,335,337
-999,-999,-999,57,55,57,55,52,50,50,47,42,32,15,7,10,2,355,350,345,347,337,335,335,335,337
-999,-999,-999,60,57,55,52,50,47,42,35,32,22,15,12,7,5,0,347,342,350,337,335,335,337,340
-999,-999,-999,57,57,52,50,47,45,45,42,40,30,20,10,10,10,0,345,345,347,337,332,332,340,342
-999,-999,-999,57,57,52,50,50,47,45,42,42,35,20,12,12,10,0,350,350,342,337,335,332,340,345
-999,-999,-999,55,52,52,55,55,55,50,47,42,32,22,17,15,10,5,357,352,342,340,337,337,342,347
-999,-999,-999,50,50,50,50,52,50,50,45,37,35,25,12,5,5,2,0,355,345,342,340,340,347,347
-999,-999,-999,50,50,50,50,50,47,45,37,37,35,27,12,357,5,5,5,357,350,345,342,345,350,350
-999,-999,-999,50,50,50,47,45,45,40,37,37,32,20,10,7,7,10,12,5,355,347,345,350,352,347
-999,-999,-999,47,47,42,42,40,37,30,30,27,25,20,20,20,22,20,15,7,357,350,347,350,350,345
-999,-999,-999,47,42,40,40,37,30,27,30,30,30,30,30,30,30,22,15,10,2,355,350,350,350,345
-999,-999,-999,35,37,37,37,32,27,32,37,37,32,32,32,27,22,20,15,10,5,2,352,350,347,342
-999,-999,-999,27,32,35,32,32,35,37,37,37,37,37,32,27,25,17,15,10,10,5,355,352,345,342
-999,-999,-999,30,35,35,35,35,35,37,37,40,40,40,35,27,27,22,17,12,12,7,357,355,350,345
-999,-999,-999,32,35,37,37,37,37,37,37,40,42,40,35,35,30,22,17,12,10,5,357,352,352,345
-999,-999,-999,35,35,37,37,37,37,35,40,42,42,40,37,32,27,25,17,12,10,2,355,350,352,342
-999,-999,-999,32,37,37,37,40,40,40,42,42,37,37,37,30,30,22,12,10,12,2,350,347,352,340

风速矩阵数据（nan为无效值）：

依靠风向矩阵数据转换的风向图

依靠风速矩阵数据转换的风速图

我该怎么更改提示词，让模型能识别台风

Answer 1

阿里嘎多学长整理AIGC生成，因移动端显示问题导致当前答案未能完全显示，请使用PC端查看更加详细的解答过程

解答

你想使用 DeepSeek 来识别台风，通过风场数据矩阵来实现。DeepSeek 提供了接口来调取风场数据矩阵，并使用机器学习算法来识别台风。

从你提供的代码中，可以看到你已经写了一段提示词 build_ana，但是最后的结果不太好。可能是因为你没有正确地使用 DeepSeek 的接口或没有选择合适的算法。

以下是一些建议：

1. 请确保你已经正确地安装了 DeepSeek 的 Python 库，并且已经 import 了相关的模块。
2. 请检查你的风场数据矩阵是否正确，是否包含足够的信息来识别台风。
3. 请尝试使用不同的算法或参数来调整识别结果。
4. 请检查你的提示词 build_ana 是否正确地使用了 DeepSeek 的接口和算法。

以下是一个简单的示例代码，使用 DeepSeek 来识别台风：

import deepseek

# 加载风场数据矩阵
data = pd.read_csv('wind_data.csv')

# 创建 DeepSeek 对象
ds = deepseek.DeepSeek()

# 设置算法和参数
ds.set_algorithm('tropical_cyclone')
ds.set_params({'threshold': 0.5})

# 使用 DeepSeek 来识别台风
results = ds.predict(data)

# 打印识别结果
print(results)

请注意，这只是一个简单的示例代码，你需要根据你的实际情况来调整代码和参数。

如果你需要更多帮助，请提供更多的代码和信息，我将尽力帮助你解决问题。