时空降水预测（python）

Question

我是Python新手，我正在尝试制作一个GUI，在地图上显示半小时时间范围内的预测降水量值。数据位于 txt 文件的文件夹中。每个文件包含每 15 分钟进行一次的测量，并包含经度、纬度和降水量值（每个文件中 270000 个位置）。为了对下一小时进行预测，我合并了最后 8 个文件，它们对应于两小时的时间范围，并添加了一个额外的列来记录测量时间。 这是数据集：

           LAT     LON    Prec
time
11:15:00  45.980  10.029    0.42  
11:15:00  45.981  10.071    0.47  
11:15:00  45.982  10.113    0.32  
11:15:00  45.984  10.155    0.30  
11:15:00  45.985  10.197    0.32  
...          ...     ...    ...    
13:00:00  29.299  33.978    0.00  
13:00:00  29.302  34.021    0.00  
13:00:00  29.304  34.064    0.00  
13:00:00  29.306  34.107    0.00  
13:00:00  29.309  34.150    0.00 

我使用过线性回归、svr 和 lstm，但我的结果到处都是。例如，当我使用线性回归时，它就像没有捕获现象的运动，它与最后一小时的数据基本相同。 这是我的代码：

import pandas as pd
import cartopy.crs as ccrs
import matplotlib.ticker as ticker
import cartopy.feature as cfeature
import matplotlib.pyplot as plt
import cartopy.crs as ccrs
import numpy as np
from sklearn.linear_model import LinearRegression
import warnings
warnings.filterwarnings("ignore", message="X does not have valid feature names")

# Load data
df_comb = pd.read_csv('combined_files.txt', delim_whitespace=' ', dtype='unicode')

df_comb = df_comb.astype({
    'LAT': float,
    'LON': float,
    'Prec': float,
    'time': str
})

df_comb['Prec'] = df_comb['Prec'].clip(lower=0)
df_comb['time'] = pd.to_datetime(df_comb['time'], format='%H%M').dt.time
df_comb = df_comb.set_index('time')

# Create lagged features for each of the 5 previous intervals
for i in range(1, 8):
    df_comb[f'Prec_lag{i}'] = df_comb['Prec'].shift(i)

# Drop the missing values introduced by lagging
df_comb = df_comb.dropna()

# Use the lagged features and location data for training
features = ['LAT', 'LON'] + [f'Prec_lag{i}' for i in range(1, 8)]
X_train = df_comb[features]
y_train = df_comb['Prec']

# Assign feature names to X_train
X_train.columns = features

# Create and fit the linear regression model
regression_model = LinearRegression()
regression_model.fit(X_train, y_train)

# List to store the predictions
predictions = []

# Iterate over the locations and make predictions
for location in df_comb.groupby(['LAT', 'LON']):
    lat, lon = location[0]
    location_data = location[1]
    X_pred = location_data[features].iloc[-1].values.reshape(1, -1)
    y_pred_next_hour = regression_model.predict(X_pred)
    predictions.append({'LAT': lat, 'LON': lon, 'Predicted_Precipitation': y_pred_next_hour[0]})

# Create a DataFrame from the predictions list
predictions_df = pd.DataFrame(predictions)
predictions_df['Predicted_Precipitation'] = predictions_df['Predicted_Precipitation'].clip(lower=0)

# Plotting
vmin = 0
vmax = 20

fig = plt.figure(figsize=(6, 6))
ax = plt.axes(projection=ccrs.PlateCarree())

cs = ax.tricontourf(predictions_df['LON'], predictions_df['LAT'], predictions_df['Predicted_Precipitation'], vmin=vmin, vmax=vmax, locator=ticker.MaxNLocator(150),
                    origin='lower',
                    transform=ccrs.PlateCarree(), cmap='jet', extend='neither')

ax.coastlines(resolution='10m')
ax.add_feature(cfeature.BORDERS, linestyle=':')
cbar_vmax = np.max(predictions_df['Predicted_Precipitation'])
plt.colorbar(cs, shrink=0.5, extend='neither', ticks=np.linspace(vmin, cbar_vmax, num=7), format='%.1f')
plt.tight_layout()
plt.show()

13:00真实数据：

13.30的真实数据，这是我预测后的预期：

“预测”结果：

如何解决这个问题？我知道数据集对于这种类型的预测是有限的。但这就是我被要求做的。

Answer 1

您好，可以分享一下您用LSTM模型进行上述操作的源代码吗，非常感谢您