OrgTJWater/pipeline-lifetime
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update prediction results workflow

Browse Source
This commit is contained in:
jiang
2026-07-20 13:47:39 +08:00
parent 8d347cfbc2
commit a20339d35e
10 changed files with 239 additions and 207 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app/prediction.py
+143 -116
View File
@@ -61,6 +61,16 @@ PIPE_AGE_COLUMN = "管龄（年）"
LEGACY_PIPE_AGE_COLUMN = "管龄"
STATUS_COLUMN = "状态"
EVENT_AGE_COLUMN = "事件/观察管龄（年）"
COLUMN_ALIASES = {
"管径（mm）": "管径",
"管径（mm)": "管径",
"流速（m/s）": "流速",
"压力（MPa)": "压力",
"压力（MPa）": "压力",
"温度（℃）": "温度",
"年均降雨量（mm）": "降雨量",
"降雨量（mm）": "降雨量",
}
INPUT_COLUMNS = [
ID_COLUMN,
PIPE_AGE_COLUMN,
@@ -68,6 +78,7 @@ INPUT_COLUMNS = [
EVENT_AGE_COLUMN,
*FEATURES,
]
CHART_DISPLAY_LIMIT = 10
MATERIAL_COLUMN = "管材"
MATERIAL_CODE_OPTIONS = [
(1, "镀锌"),
@@ -103,6 +114,7 @@ MATERIAL_ALIAS_TO_CODE.update(
)
SUPPORTED_EXTENSIONS = {".csv", ".xls", ".xlsx"}
CHINESE_FONT_PROP = None
DEFECT_GRADE_VALUES = {"无": 0.0, "轻度": 1.0, "中度": 3.0, "严重": 5.0}
class PredictionError(Exception):
@@ -119,12 +131,35 @@ class PredictionArtifacts:
excel_path: Path
image_path: Path
image_filename: str
importance_filename: str | None
sample_count: int
summary_rows: list[dict[str, Any]]
analysis_text: str
class ModelBundleAdapter:
"""Expose a bundled preprocessor and survival model as one predictor."""
def __init__(self, preprocessor, model) -> None:
self.preprocessor = preprocessor
self.model = model
def _transform(self, frame: pd.DataFrame):
values = frame[FEATURES].copy()
values[MATERIAL_COLUMN] = values[MATERIAL_COLUMN].map(
lambda value: f"M{int(value)}" if pd.notna(value) else value
)
values["位置"] = values["位置"].map(
lambda value: f"L{int(value)}" if pd.notna(value) else value
)
return self.preprocessor.transform(values.to_numpy(dtype=object)).astype(np.float32)
def predict_survival_function(self, frame: pd.DataFrame):
return self.model.predict_survival_function(self._transform(frame))
def predict(self, frame: pd.DataFrame):
return self.model.predict(self._transform(frame))
def configure_matplotlib_fonts():
for font_path in CHINESE_FONT_FILES:
path = Path(font_path)
@@ -159,6 +194,10 @@ def load_model(model_path: str):
if not os.path.exists(model_path):
raise FileNotFoundError(f"未找到模型文件: {model_path}")
model = joblib.load(model_path)
if isinstance(model, dict):
if not {"preprocessor", "model"}.issubset(model):
raise ValueError("模型包缺少 preprocessor 或 model。")
model = ModelBundleAdapter(model["preprocessor"], model["model"])
feature_names = getattr(model, "feature_names_in_", None)
if feature_names is not None and list(feature_names) != FEATURES:
raise ValueError(
@@ -195,17 +234,25 @@ def read_input_file(path: Path, suffix: str) -> pd.DataFrame:
if suffix == ".csv":
df = pd.read_csv(path, dtype={ID_COLUMN: "string"})
else:
df = pd.read_excel(path, dtype={ID_COLUMN: "string"})
try:
df = pd.read_excel(path, sheet_name="Template", dtype={ID_COLUMN: "string"})
except ValueError:
df = pd.read_excel(path, dtype={ID_COLUMN: "string"})
df["_excel_row"] = df.index + 2
except Exception as exc:
logging.exception("文件解析失败: %s", exc)
raise PredictionError("文件解析失败，请检查编码或表格格式。")
return normalize_input_columns(df)
df = normalize_input_columns(df)
if suffix in {".xls", ".xlsx"}:
df = fill_defects_from_detail_sheet(path, df)
return df
def normalize_input_columns(df: pd.DataFrame) -> pd.DataFrame:
aliases = dict(COLUMN_ALIASES)
if PIPE_AGE_COLUMN not in df.columns and LEGACY_PIPE_AGE_COLUMN in df.columns:
return df.rename(columns={LEGACY_PIPE_AGE_COLUMN: PIPE_AGE_COLUMN})
return df
aliases[LEGACY_PIPE_AGE_COLUMN] = PIPE_AGE_COLUMN
return df.rename(columns={name: aliases[name] for name in df.columns if name in aliases})
def normalize_id_value(value: Any, fallback: str) -> str:
@@ -216,6 +263,79 @@ def normalize_id_value(value: Any, fallback: str) -> str:
return str(value)
def defect_grade_value(value: Any) -> float | None:
if pd.isna(value):
return None
text = str(value).strip()
if text == "":
return None
if text in DEFECT_GRADE_VALUES:
return DEFECT_GRADE_VALUES[text]
try:
return float(text)
except ValueError:
return None
def weighted_defect_score(values: list[Any], weights: list[float]) -> float | None:
scores = [defect_grade_value(value) for value in values]
if any(score is None for score in scores):
return None
return round(sum(score * weight for score, weight in zip(scores, weights)), 3)
def numeric_or_none(value: Any) -> float | None:
try:
if pd.isna(value):
return None
return float(value)
except (TypeError, ValueError):
return None
def fill_defects_from_detail_sheet(path: Path, df: pd.DataFrame) -> pd.DataFrame:
if "结构缺陷" not in df.columns or "功能缺陷" not in df.columns or "_excel_row" not in df.columns:
return df
structure = pd.to_numeric(df["结构缺陷"], errors="coerce")
function = pd.to_numeric(df["功能缺陷"], errors="coerce")
needs_fill = structure.isna() | function.isna()
df["结构缺陷"] = structure
df["功能缺陷"] = function
if not needs_fill.any():
return df
try:
from openpyxl import load_workbook
workbook = load_workbook(path, data_only=False, read_only=True)
defect_sheet = workbook["缺陷计算"]
except Exception as exc:
logging.info("未能读取缺陷计算工作表，跳过缺陷自动计算: %s", exc)
return df
for index in df.index[needs_fill]:
excel_row = int(df.at[index, "_excel_row"])
if pd.isna(df.at[index, "结构缺陷"]):
value = numeric_or_none(defect_sheet.cell(excel_row, 5).value)
if value is None:
value = weighted_defect_score(
[defect_sheet.cell(excel_row, column).value for column in (2, 3, 4)],
[0.5, 0.4, 0.1],
)
df.at[index, "结构缺陷"] = value
if pd.isna(df.at[index, "功能缺陷"]):
value = numeric_or_none(defect_sheet.cell(excel_row, 10).value)
if value is None:
value = weighted_defect_score(
[defect_sheet.cell(excel_row, column).value for column in (6, 7, 8, 9)],
[0.5, 0.2, 0.2, 0.1],
)
df.at[index, "功能缺陷"] = value
return df
def validate_input_frame(df: pd.DataFrame) -> None:
if df.empty:
raise PredictionError("上传文件没有可预测的数据。")
@@ -288,99 +408,6 @@ def make_analysis_text(summary_rows: list[dict[str, Any]]) -> str:
)
def compute_feature_importance(model, x_test: pd.DataFrame) -> np.ndarray | None:
try:
importances = model.feature_importances_
except Exception:
importances = None
if importances is not None and len(importances) == len(FEATURES):
vals = np.asarray(importances, dtype=float)
else:
try:
baseline = np.asarray(model.predict(x_test), dtype=float)
except Exception as exc:
logging.exception("特征重要性基线预测失败: %s", exc)
return None
rng = np.random.default_rng(42)
vals = np.zeros(len(FEATURES), dtype=float)
for j, feat in enumerate(FEATURES):
diffs = []
for _ in range(5):
x_perm = x_test.copy()
x_perm[feat] = rng.permutation(x_perm[feat].to_numpy())
try:
perm_pred = np.asarray(model.predict(x_perm), dtype=float)
except Exception:
perm_pred = baseline
diffs.append(float(np.mean(np.abs(perm_pred - baseline))))
vals[j] = float(np.mean(diffs)) if diffs else 0.0
vals = np.clip(vals, a_min=0.0, a_max=None)
total = float(vals.sum())
if total > 0:
vals = vals / total
return vals
def render_importance_chart(values: np.ndarray, save_path: Path) -> None:
from matplotlib.colors import LinearSegmentedColormap
keep = values >= 5e-4
if not bool(np.any(keep)):
keep = np.ones_like(values, dtype=bool)
kept_feats = [FEATURES[i] for i in range(len(FEATURES)) if keep[i]]
kept_vals = values[keep]
order = np.argsort(kept_vals)
sorted_feats = [kept_feats[k] for k in order]
sorted_vals = kept_vals[order]
n = len(sorted_vals)
fig, ax = plt.subplots(figsize=(9, max(3.0, 0.62 * n + 1.6)))
cmap = LinearSegmentedColormap.from_list("brand", ["#7fb2e6", "#005EB8", "#0c4188"])
colors = cmap(np.linspace(0.15, 1.0, n)) if n else None
y_pos = np.arange(n)
bars = ax.barh(y_pos, sorted_vals, color=colors, height=0.66, edgecolor="white", linewidth=0.8, zorder=3)
font_kwargs = chinese_font_kwargs()
ax.set_xlabel("相对重要性", fontsize=11, color="#475569", **font_kwargs)
ax.set_title("模型输入因素重要性排序", fontsize=15, fontweight="bold", color="#0f172a", pad=14, **font_kwargs)
ax.set_yticks(y_pos)
ax.set_yticklabels(sorted_feats, **font_kwargs)
ax.grid(axis="x", color="#e2e8f0", linewidth=1, zorder=0)
ax.set_axisbelow(True)
for spine in ("top", "right", "left"):
ax.spines[spine].set_visible(False)
ax.spines["bottom"].set_color("#cbd5e1")
ax.tick_params(axis="y", length=0, labelsize=11)
ax.tick_params(axis="x", colors="#94a3b8", labelsize=9)
if CHINESE_FONT_PROP:
for label in [*ax.get_yticklabels(), *ax.get_xticklabels()]:
label.set_fontproperties(CHINESE_FONT_PROP)
max_val = float(sorted_vals.max()) if n else 0.0
for bar, value in zip(bars, sorted_vals):
ax.text(
bar.get_width() + max_val * 0.012,
bar.get_y() + bar.get_height() / 2,
f"{value:.1%}",
va="center",
ha="left",
fontsize=10,
fontweight="bold",
color="#1e293b",
**font_kwargs,
)
if max_val > 0:
ax.set_xlim(0, max_val * 1.18)
fig.tight_layout()
fig.savefig(save_path, dpi=160, bbox_inches="tight")
plt.close(fig)
def run_prediction(uploaded: FileStorage, user_id: int, model) -> PredictionArtifacts:
original_filename = uploaded.filename or ""
if not original_filename:
@@ -409,15 +436,6 @@ def run_prediction(uploaded: FileStorage, user_id: int, model) -> PredictionArti
image_path = IMAGE_DIR / image_filename
summary_rows, summary_sheet_rows = render_survival_chart(df, curves, image_path)
importance_filename = None
try:
importance_values = compute_feature_importance(model, x_test)
if importance_values is not None:
importance_filename = f"importance_{user_id}_{run_id}.png"
render_importance_chart(importance_values, IMAGE_DIR / importance_filename)
except Exception as exc:
logging.exception("生成特征重要性图失败: %s", exc)
safe_stem = Path(saved_filename).stem
excel_path = user_dir / f"{safe_stem}_pre.xlsx"
write_prediction_workbook(excel_path, curves, summary_rows, summary_sheet_rows)
@@ -427,7 +445,6 @@ def run_prediction(uploaded: FileStorage, user_id: int, model) -> PredictionArti
excel_path=excel_path,
image_path=image_path,
image_filename=image_filename,
importance_filename=importance_filename,
sample_count=len(summary_rows),
summary_rows=summary_rows,
analysis_text=make_analysis_text(summary_rows),
@@ -441,13 +458,17 @@ def render_survival_chart(df: pd.DataFrame, curves, image_path: Path) -> tuple[l
plt.figure(figsize=(10, 5.6))
summary_rows: list[dict[str, Any]] = []
summary_sheet_rows: list[dict[str, Any]] = []
display_note = f"说明：图中仅展示前{CHART_DISPLAY_LIMIT}条管道的示例数据，不足{CHART_DISPLAY_LIMIT}条则全部展示；完整结果请下载电子表格。"
for i, curve in enumerate(curves):
times = [float(x) for x in list(curve.x)]
probs = [float(y) for y in list(curve.y)]
pipe_id = normalize_id_value(df.iloc[i][ID_COLUMN], f"Pipe_{i+1:03d}")
pipe_age = f"{df.iloc[i][PIPE_AGE_COLUMN]} 年" if PIPE_AGE_COLUMN in df.columns and pd.notna(df.iloc[i][PIPE_AGE_COLUMN]) else "-"
health_probability = interpolate_probability(times, probs, 10)
raw_pipe_age = df.iloc[i][PIPE_AGE_COLUMN] if PIPE_AGE_COLUMN in df.columns else None
pipe_age_value = float(raw_pipe_age) if pd.notna(raw_pipe_age) else 0.0
pipe_age = f"{raw_pipe_age} 年" if pd.notna(raw_pipe_age) else "-"
health_probability = interpolate_probability(times, probs, pipe_age_value)
health_risk = 1.0 - health_probability
remaining_life = estimate_remaining_life(times, probs)
grade_label, grade_desc, grade_class = grade_info(health_probability)
@@ -456,6 +477,7 @@ def render_survival_chart(df: pd.DataFrame, curves, image_path: Path) -> tuple[l
"pipe_id": pipe_id,
"pipe_age": pipe_age,
"health_probability": health_probability,
"health_risk": health_risk,
"remaining_life": remaining_life,
"grade_label": grade_label,
"grade_desc": grade_desc,
@@ -466,23 +488,25 @@ def render_survival_chart(df: pd.DataFrame, curves, image_path: Path) -> tuple[l
{
ID_COLUMN: pipe_id,
PIPE_AGE_COLUMN: pipe_age,
"健康等级": grade_label,
"健康风险值": health_risk,
}
)
plt.step(times, probs, where="post", linewidth=2, label=pipe_id)
if i < CHART_DISPLAY_LIMIT:
plt.step(times, probs, where="post", linewidth=2, label=pipe_id)
font_kwargs = chinese_font_kwargs()
plt.xlabel("管龄（年）", **font_kwargs)
plt.ylabel("健康风险", **font_kwargs)
plt.title("预测分析图", **font_kwargs)
plt.title("管道的剩余寿命分析图", **font_kwargs)
plt.figtext(0.5, 0.02, display_note, ha="center", fontsize=9, color="#475569", **font_kwargs)
plt.grid(alpha=0.18)
if len(summary_rows) <= 12:
if min(len(summary_rows), CHART_DISPLAY_LIMIT) <= 12:
plt.legend(loc="best", fontsize=8, prop=CHINESE_FONT_PROP)
ax = plt.gca()
if CHINESE_FONT_PROP:
for label in [*ax.get_xticklabels(), *ax.get_yticklabels()]:
label.set_fontproperties(CHINESE_FONT_PROP)
plt.tight_layout()
plt.tight_layout(rect=(0, 0.07, 1, 1))
plt.savefig(image_path, dpi=160, bbox_inches="tight")
plt.close()
return summary_rows, summary_sheet_rows
@@ -511,7 +535,7 @@ def write_prediction_workbook(
]
with pd.ExcelWriter(excel_path, engine="openpyxl") as writer:
summary_df = pd.DataFrame(summary_sheet_rows, columns=[ID_COLUMN, PIPE_AGE_COLUMN, "健康等级"])
summary_df = pd.DataFrame(summary_sheet_rows, columns=[ID_COLUMN, PIPE_AGE_COLUMN, "健康风险值"])
sample_df = pd.DataFrame(sample_data_rows, columns=sample_columns)
if ID_COLUMN in summary_df.columns:
summary_df[ID_COLUMN] = summary_df[ID_COLUMN].astype("string")
@@ -520,6 +544,9 @@ def write_prediction_workbook(
sample_df.to_excel(writer, sheet_name="样本数据", index=False)
sample_worksheet = writer.book["样本数据"]
summary_worksheet = writer.book["结果摘要"]
for cell in summary_worksheet["C"][1:]:
cell.number_format = "0.0%"
sample_worksheet.freeze_panes = "A2"
sample_worksheet.auto_filter.ref = sample_worksheet.dimensions
sample_worksheet.column_dimensions["A"].width = 12