五一数学建模论文与代码
A mining-safety modelling project that predicts rockburst risk from electromagnetic radiation and acoustic-emission signals.
97.99%
EMR test accuracy
99.40%
AE test accuracy
31p
sanitized paper
01 / Overview
The May Day competition project predicts rockburst risk from electromagnetic radiation (EMR) and acoustic-emission (AE) signals. It combines time-domain features, FFT features, wavelet denoising and calibrated classifiers.
Objectives
Competition narrative
The page foregrounds the modelling decision and result before implementation details.
Stage 01
EMR and AE measurements over time.
Stage 02
Wavelet cleanup and rolling statistics.
Stage 03
Mean, variance, peak, frequency and spectral energy.
Stage 04
Random Forest, Naive Bayes and Platt-scaled probabilities.
Stage 05
Precursor intervals and highest-risk time windows.
02 / Package highlights
03 / Model
The workflow extracted seven descriptive features including mean, variance, peak factor, skewness, dominant frequency and spectral energy density. FFT features and rolling statistics made the time-series variation visible to tabular classifiers.
A Gaussian Naive Bayes classifier was wrapped in CalibratedClassifierCV using Platt scaling. This turns raw classifier outputs into probability estimates that can be compared across windows and used to rank risk.
The paper reports 97.99% classification accuracy for EMR precursor signals and 99.40% for AE signals on the test set. The output includes precursor intervals and a highest-probability time window for the prediction data.
04 / Key code
Only verified source excerpts and pseudocode taken from the project papers are shown.
Original Python excerpt for signal preprocessing and spectral features.
def cal_fft_features(signal):
signal = np.asarray(signal).squeeze()
fft_abs = np.abs(np.fft.fft(signal))
return np.mean(fft_abs), np.std(fft_abs), np.max(fft_abs)
signal_denoised = wavelet_denoising(
data['声波强度 (AE)'].values
)
data['声波强度 (AE)_denoised'] = signal_denoised
fft_features = data['声波强度 (AE)_denoised'].apply(
cal_fft_features
)
data['fft_mean'], data['fft_std'], data['fft_max'] = zip(
*fft_features
)
data['d1'] = data['声波强度 (AE)_denoised'].diff(1).fillna(0)
data['rolling_mean'] = (
data['声波强度 (AE)_denoised'].rolling(10).mean().bfill()
)
data['rolling_std'] = (
data['声波强度 (AE)_denoised'].rolling(10).std().bfill()
)Original Python excerpt for probability calibration and risk ranking.
model = GaussianNB()
model.fit(X_train, y_train)
calibrated_model = CalibratedClassifierCV(
model, method='sigmoid', cv=5
)
calibrated_model.fit(X_train, y_train)
probabilities = calibrated_model.predict_proba(X_test)
predictions = calibrated_model.predict(X_test)
predict_proba = calibrated_model.predict_proba(
predict_data[features_list]
)
highest_index = np.argmax(predict_proba[:, 1])
highest_time = predict_data.iloc[highest_index]['time']
highest_probability = predict_proba[highest_index, 1]Reproducibility notes
Limitations / next steps
05 / Downloads
Open the original report, presentation or source archive. PDF cover pages with private details were removed.
31-page paper with the personal-information cover removed.