新建
EIS 自动化拟合等效电路模型 ECM
JiaweiMiao
推荐镜像 :Basic Image:bohrium-notebook:2023-03-26
推荐机型 :c8_m31_1 * NVIDIA T4
赞
目录
数据集
EIS-ECM(v1)
EIS 自动化拟合等效电路模型 ECM
代码
文本
Abstract
电化学系统的电化学阻抗谱 (EIS) 数据分析通常包括使用专家知识定义等效电路模型 (ECM)，然后优化模型参数以解卷积各种电阻、电容、电感或扩散响应。对于小型数据集，此过程可以手动进行；但是，对于具有广泛 EIS 响应的大量数据集，手动定义适当的 ECM 是不可行的。自动识别 ECM 将大大加快大量 EIS 数据的分析。我们展示了机器学习方法，用于对 QuantumScape 为 BatteryDEV 黑客马拉松提供的 9,300 个阻抗谱的 ECM 进行分类。
本文分为两个模型的实现，第一个是 RF 模型，第二个是 CNN 模型
关于阻抗 EIS 预测 ECM 的benchmark文章参考：Machine Learning Benchmarks for the Classification of Equivalent Circuit Models from Electrochemical Impedance Spectra
代码
文本
Introduction
关于EIS以及等效电路模型有以下背景需要了解：
电化学阻抗谱（EIS）是一种常用于表征电池的非破坏性方法，它揭示了LiB内部的动态电极动力学过程。基于不同电化学过程表现出的时间常数，可以通过将阻抗谱的变化与固体电解质界面（SEI）生长、电荷转移和锂离子扩散等内部机制联系起来来表征电池性能。辅助研究工具包括等效电路模型（ECM）、弛豫时间分布（DRT）、阻抗机制模型等。 可以区分和量化多时间尺度特征，以更好地了解电池状态。已有工作显示2，借助机器学习方法，在电芯筛选和老化监测这两种情况下，根据阻抗可以可靠地量化关键的物理化学参数。 结果表明，电芯筛选时可以EIS准确估算电池电极的弯曲度、孔隙率和活性物质含量，误差小于2%；在老化监测方面，SEI电阻和电荷转移电阻可以准确估计，误差小于5% 。 上述应用表明EIS提供了对电池更全面的理解。
EIS 可以用于多种电话学系统的评估，如例如锂电池、燃料电池、超级电容器、腐蚀 或生物系统。对于电池，利用 EIS 的特定研究领域包括等效电路模型 (ECM) 表征、阻断电极实验以研究纯离子或电子行为、扩散过程建模、多孔电极表征、通过传输线建模进行电极表征以及电池性能监测。然而，为了分析这些光谱并指定特定的机制（如电子电阻、电荷转移、质量传输等），电化学家通常使用 ECM 来表示电池中的不同物理化学过程，通过电感、电阻、电容或它们的组合等电路元件对其进行参数化。定义 ECM 的结构通常需要专家判断，这意味着对大量 EIS 测量的评估是一个难以自动化的过程。
机器学习方法在电化学领域日益流行，特别是在EIS数据分析中，已成功应用于电池寿命预测和各种光谱数据分析。近期发展包括基于贝叶斯推理的EIS数据模型选择。尽管已有一些开源工具和数据集，但仍有巨大发展空间。本文通过分享由QuantumScape提供的大型合成EIS数据集和未标记数据集，为开源电池数据和软件做出贡献，重点使用ML方法准确分类等效电路模型(ECM)。这些数据集用于BatteryDEV黑客马拉松，旨在寻找ECM识别问题的跨学科解决方案。
代码
文本
加载工具包
代码
文本
已隐藏单元格
代码
文本
已隐藏单元格
已隐藏输出
代码
文本
设置随机数种子
代码
文本
[3]
np.random.seed(42)
plt.ion()
<matplotlib.pyplot._IonContext at 0x7f07ff4df4c0>
代码
文本
RF model
构建估计函数
代码
文本
[4]
def tune_nb_estimator(
X_train_scaled: np.ndarray,
y_train: np.ndarray,
folds: int = 5,
params_dict: dict = {},
) -> Tuple[int, int]:
n_estimators = [10, 50, 100, 150, 200, 500, 800, 1000]
f1 = []
f1_std = []
# remove nb_estimators if they are already in params_dict
if "n_estimators" in params_dict.keys():
params_dict.pop("n_estimators")
for nb in n_estimators:
print(f"CV with number of estimators: {nb}")
clf = RandomForestClassifier(
class_weight="balanced_subsample",
n_estimators=nb,
n_jobs=-1,
random_state=42,
**params_dict,
)
# Cross-validation: Predict the test samples based on a predictor that was trained with the
# remaining data. Repeat until prediction of each sample is obtained.
# (Only one prediction per sample is allowed)
# Only these two cv methods work. Reson: Each sample can only belong to EXACTLY one test set.
# Other methods of cross validation might violate this constraint
# For more information see:
# https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.cross_val_predict.html
scores = cross_val_score(
clf, X_train_scaled, y_train, cv=folds, n_jobs=-1, scoring="f1_macro"
)
f1.append(scores.mean())
f1_std.append(scores.std())
# Pick the best number of estimators based on the one standard deviation rule
best_nb = n_estimators[np.argmax(f1)]
print("Best number of estimators: {}".format(best_nb))
# If you desire a smaller model, uncomment the following lines and use best_nb_std
f1_max_loc = np.argmax(f1)
filtered_lst = [
(i, element)
for i, element in enumerate(f1)
if element > f1[f1_max_loc] - (1 * f1_std[f1_max_loc])
]
f1_std_max_loc, _ = min(filtered_lst)
best_nb_std = n_estimators[f1_std_max_loc]
print("Best number of estimators (1std): {}".format(best_nb_std))
return best_nb, best_nb_std
代码
文本
构建模型
代码
文本
[5]
# Hyperopt for hyperparameter optimization did not work well for this dataset.
def baseline_model(
train_data_f: str,
test_data_file: str,
output_dir: str,
compare_models: bool = True,
cross_val: Optional[Literal["simple", "extended"]] = None,
filtered: bool = False,
save: bool = False,
save_model: bool = False,
) -> None:
"""Baseline model for predicting ECM from EIS data
Linear classifiers perform poorly on this dataset, so we use a RF classifier (non-linear)
Hyperparameter optimization with hpsklearn, keeping it simple for the baseline model
Parameters
----------
train_data_f : str
Path to training data file
test_data_file : str
Path to test data file
output_dir : str
Path to output directory
compare_models : bool
Compare different models cross validation accouracies
cross_val : bool
Perform cross validation
save : bool
Save figures
Returns
-------
acc_test : float
Test accuracy
acc_train : float
Train accuracy
"""
# Load data
train_data = np.loadtxt(train_data_f, delimiter=",")
test_data = np.loadtxt(test_data_file, delimiter=",")
# Preprocess data
X_train = train_data[:, 0:-1]
y_train = train_data[:, -1].ravel()
X_test = test_data[:, 0:-1]
y_test = test_data[:, -1].ravel()
# Standardize data using sklearn standard scaler
# This approach does not make sense, Zreal and Zimag should not be scaled independently
# scaler = StandardScaler()
# X_train_scaled = scaler.fit_transform(X_train)
# X_test_scaled = scaler.transform(X_test)
# The following ensures, that Zreal and Zimag are scaled together
for i in range(X_train.shape[0]):
divider = np.max(X_train[i, :30])
X_train[i, :] = X_train[i, :] / divider
X_train_scaled = X_train.copy()
for i in range(X_test.shape[0]):
divider = np.max(X_test[i, :30])
X_test[i, :] = X_test[i, :] / divider
X_test_scaled = X_test.copy()
folds = 5
if cross_val is not None:
if cross_val == "simple":
# Use sklearn 5 fold CV to determine the number of estimators
# Loop through all possibilities
best_nb, best_nb_std = tune_nb_estimator(X_train_scaled, y_train, folds)
simple_model = RandomForestClassifier(
class_weight="balanced_subsample",
n_estimators=best_nb,
n_jobs=-1,
random_state=42,
)
if cross_val == "extended":
folds = 5
parameters = {
"bootstrap": [True, False],
"max_depth": [10, 75, None],
"max_features": ["sqrt", None],
"min_samples_leaf": [1, 3, 5, 10],
"min_samples_split": [2, 3, 5, 8],
"n_estimators": [10, 100, 300, 600],
}
clf = RandomForestClassifier(
class_weight="balanced_subsample", max_depth=None, random_state=42
)
clf_gs = GridSearchCV(
clf, parameters, scoring="f1_macro", cv=folds, n_jobs=-1, verbose=5
)
clf_gs.fit(
X_train_scaled,
y_train,
)
# Print the best parameters
print("Best parameters set found on training set:")
print(clf_gs.best_params_)
clf = clf_gs.best_estimator_
with open(f"{output_dir}/best_hyperparameters.txt", "w") as f:
f.write(str(clf_gs.best_params_))
if compare_models:
# Models from previous runs
simple_model = RandomForestClassifier(
class_weight="balanced_subsample",
n_estimators=800,
n_jobs=-1,
random_state=42,
)
params_dict = {
"bootstrap": True,
"max_depth": 75,
"max_features": None,
"min_samples_leaf": 1,
"min_samples_split": 3,
"n_estimators": 600,
}
gs_model = RandomForestClassifier(
class_weight="balanced_subsample", n_jobs=-1, random_state=42, **params_dict
)
# Compare the performance of the model with only the number of estimators tuned and the larger grid search
gs_model_score = cross_val_score(
gs_model, X_train, y_train, cv=5, n_jobs=-1, scoring="f1_macro"
)
default_model_score = cross_val_score(
simple_model, X_train, y_train, cv=5, n_jobs=-1, scoring="f1_macro"
)
print("Grid search model score: ", gs_model_score.mean())
print("Simple model score: ", default_model_score.mean())
print("Difference: ", gs_model_score.mean() - default_model_score.mean())
# Standard deviations
print("Grid search model std: ", gs_model_score.std())
print("Simple model std: ", default_model_score.std())
print("Comparison done!")
# Results in the manuscript are based on the following parameters:
# Best parameters set found on filtered training set with grid search:
# (The results where the same fot the filtered and unfiltered data)
params_dict = {
"bootstrap": True,
"max_depth": 75,
"max_features": None,
"min_samples_leaf": 1,
"min_samples_split": 3,
"n_estimators": 300,
}
clf = RandomForestClassifier(
class_weight="balanced_subsample", n_jobs=-1, random_state=42, **params_dict
)
score = cross_val_score(clf, X_train, y_train, cv=5, n_jobs=-1, scoring="f1_macro")
print(score)
clf.fit(X_train_scaled, y_train)
y_test_pred = clf.predict(X_test_scaled)
y_train_pred = clf.predict(X_train_scaled)
with open("/bohr/eisecm-0e5z/v1/AutoECM/data/le_name_mapping.json", "r") as f:
mapping = json.load(f)
le = LabelEncoder()
mapping["classes"] = [mapping[str(int(i))] for i in range(9)]
le.classes_ = np.array(mapping["classes"])
plot_cm(y_test, y_test_pred, le, save=save, figname=f"{output_dir}/cm_rfb_test")
plot_cm(y_train, y_train_pred, le, save=save, figname=f"{output_dir}/cm_rfb_train")
plt.show()
# Save model
if save_model:
with open(f"{output_dir}/rf.pkl", "wb") as f:
pickle.dump(clf, f)
# Calculate f1 and save classification report
calcualte_classification_report(
y_train, y_train_pred, y_test, y_test_pred, le, save=save, output_dir=output_dir
)
return
代码
文本
运行 RF 模型
代码
文本
[6]
# if len(sys.argv) < 3:
# filter = 1
# cross_val = None
# # "extended" # "simple"
# else:
# filter = int(sys.argv[1])
# print(f"Filter: {filter}")
# cross_val = str(sys.argv[2])
# print(f"Cross validation: {cross_val}")
filter = 1
cross_val = None
# Create new folder with results, name is datetime
now = datetime.datetime.now()
now_str = now.strftime("%Y-%m-%d_%H-%M-%S")
if filter:
train_data_f = "/bohr/eisecm-0e5z/v1/AutoECM/data/interpolated/train_data_filtered.csv"
test_data_f = "/bohr/eisecm-0e5z/v1/AutoECM/data/interpolated/test_data_filtered.csv"
output_dir = f"/bohr/eisecm-0e5z/v1/AutoECM/results/clf_filtered/rf/{now_str}"
else:
train_data_f = "/bohr/eisecm-0e5z/v1/AutoECM/data/interpolated/train_data.csv"
test_data_f = "/bohr/eisecm-0e5z/v1/AutoECM/data/interpolated/test_data.csv"
output_dir = f"/bohr/eisecm-0e5z/v1/AutoECM/results/clf/rf/{now_str}"
# os.mkdir(output_dir)
baseline_model(
train_data_f,
test_data_f,
output_dir,
compare_models=False,
cross_val=cross_val,
filtered=filter,
save=False,
save_model=False,
)
print("Done")
[0.42032517 0.41422605 0.39457023 0.40045123 0.40953975]
F1 score train: 1.000
F1 score test: 0.399
precision recall f1-score support
L-R-RCPE 0.3399 0.3444 0.3421 151
L-R-RCPE-RCPE 0.2131 0.2021 0.2074 193
L-R-RCPE-RCPE-RCPE 0.2609 0.1842 0.2159 228
RC-G-G 0.6557 0.8734 0.7491 229
RC-RC-RCPE-RCPE 0.5833 0.7061 0.6389 228
RCPE-RCPE 0.2731 0.2731 0.2731 216
RCPE-RCPE-RCPE 0.2431 0.1947 0.2162 226
RCPE-RCPE-RCPE-RCPE 0.2648 0.2601 0.2624 223
Rs_Ws 0.6866 0.6866 0.6866 67
accuracy 0.3981 1761
macro avg 0.3912 0.4139 0.3991 1761
weighted avg 0.3714 0.3981 0.3808 1761
Done
代码
文本
CNN 模型
加载工具包
代码
文本
[7]
%%capture
# CNN Baseline model
# Largely inspired by Kaggle MNIST approach:
# https://www.kaggle.com/code/elcaiseri/mnist-simple-cnn-keras-accuracy-0-99-top-1/notebook
# The architecture is only slightly adapted to account for bigger images.
# Many lines of code are directly copeid form the above mentioned example.
import pandas as pd
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Rescaling # convolution layers
from keras.layers import Dense, Flatten # core layers
import tensorflow as tf
from tensorflow import keras
from keras.layers import BatchNormalization
from sklearn.metrics import f1_score, classification_report
代码
文本
[8]
# Set seeds for reproducibility
# Some randomness might still be present depending on the used libraries/hardware/GPU
# https://machinelearningmastery.com/reproducible-results-neural-networks-keras/
# tf.random.set_seed(20)
# np.random.seed(20)
SEED = 20
SHUFFLE_TRAIN_VAL = True
VAL_SPLIT = 0.1
代码
文本
构建 CNN 架构
代码
文本
[9]
def cnn_architecture(
input_shape: Tuple[int, int], nb_classes: int, adaptive_based_on_val: bool = True
) -> keras.Model:
"""CNN architecture
Parameters
----------
input_shape: Tuple
Shape of input data
nb_classes: int
Number of classes
adaptive_based_on_val_loss: bool
If True, the model os compiled with an optimizer and loss that is monitored for
early stopping. If False, the model is compiled with a another optimizer.
Returns
-------
model: keras model
CNN model
"""
model = Sequential()
model.add(
Rescaling(1.0 / 127.5, offset=-1, input_shape=(input_shape[0], input_shape[1], 1))
)
model.add(Conv2D(filters=64, kernel_size=(6, 6), activation="relu", strides=(2, 2)))
model.add(Conv2D(filters=64, kernel_size=(3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization())
model.add(Conv2D(filters=128, kernel_size=(3, 3), activation="relu"))
model.add(Conv2D(filters=128, kernel_size=(3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization())
model.add(Conv2D(filters=256, kernel_size=(3, 3), activation="relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(BatchNormalization())
model.add(Flatten())
model.add(Dense(512, activation="relu"))
model.add(Dense(nb_classes, activation="softmax"))
if adaptive_based_on_val:
model.compile(
loss="categorical_crossentropy", optimizer="adam", metrics=["accuracy"]
)
else:
optimizer_adamax = tf.keras.optimizers.Adamax(
learning_rate=0.001,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-07,
weight_decay=None,
clipnorm=None,
clipvalue=None,
global_clipnorm=None,
use_ema=False,
ema_momentum=0.99,
ema_overwrite_frequency=None,
jit_compile=True,
)
model.compile(
loss="categorical_crossentropy",
optimizer=optimizer_adamax,
metrics=["accuracy"],
)
model.summary()
return model
代码
文本
[10]
def train_model(
model: keras.Model, train_ds, val_ds, adaptive_based_on_val: bool
) -> Tuple[keras.callbacks.History, keras.Model]:
if adaptive_based_on_val:
es = keras.callbacks.EarlyStopping(
monitor="val_accuracy", # metrics to monitor
patience=60, # how many epochs before stop
verbose=1,
mode="max", # we need the maximum accuracy.
restore_best_weights=True, #
)
rp = keras.callbacks.ReduceLROnPlateau(
monitor="val_accuracy",
factor=0.2,
patience=3,
verbose=1,
mode="max",
min_lr=0.00001,
)
h = model.fit(train_ds, validation_data=val_ds, epochs=200, callbacks=[rp, es])
# Do 5 epochs with a very low learning rate
# Can be omitted if the model is already well converged
# model.compile(
# loss="categorical_crossentropy",
# optimizer=keras.optimizers.Adam(learning_rate=1e-5),
# metrics=["accuracy"],
# )
# h = model.fit(val_ds, epochs=5)
else:
h = model.fit(train_ds, epochs=150)
return h, model
代码
文本
定义执行函数
代码
文本
[14]
def main(
input_shape: Tuple[int, int],
nb_classes: int = 9,
input_dir: str = "",
output_dir: str = "",
adaptive_based_on_val: bool = True,
) -> None:
if adaptive_based_on_val:
train_ds = tf.keras.utils.image_dataset_from_directory(
f"{input_dir}/train",
validation_split=VAL_SPLIT,
subset="training",
seed=SEED,
image_size=input_shape,
label_mode="categorical",
shuffle=SHUFFLE_TRAIN_VAL,
color_mode="grayscale",
)
val_ds = tf.keras.utils.image_dataset_from_directory(
f"{input_dir}/train",
validation_split=VAL_SPLIT,
subset="validation",
seed=SEED,
image_size=input_shape,
batch_size=64,
label_mode="categorical",
shuffle=SHUFFLE_TRAIN_VAL,
color_mode="grayscale",
)
else:
train_ds = tf.keras.utils.image_dataset_from_directory(
f"{input_dir}/train",
seed=SEED,
image_size=input_shape,
label_mode="categorical",
shuffle=True,
batch_size=1000,
color_mode="grayscale",
)
val_ds = np.nan
test_ds = tf.keras.utils.image_dataset_from_directory(
f"{input_dir}/test",
validation_split=None,
seed=SEED,
image_size=input_shape,
batch_size=2000,
label_mode="categorical",
shuffle=False,
color_mode="grayscale",
)
y_train = []
for image_batch, labels_batch in train_ds:
y_train.append(labels_batch)
y_train = np.concatenate(y_train, axis=0)
for image_batch, labels_batch in test_ds:
y_test = np.array(labels_batch)
# Build model
model = cnn_architecture(
input_shape, nb_classes, adaptive_based_on_val=adaptive_based_on_val
)
# Train model
h, model = train_model(
model, train_ds, val_ds, adaptive_based_on_val=adaptive_based_on_val
)
# Predict class probabilities as 2 => [0.1, 0, 0.9, 0, 0, 0, 0, 0, 0, 0]
y_test_pred = model.predict(test_ds)
Y_test_pred = np.argmax(y_test_pred, 1) # Decode Predicted labels
# Predict class probabilities as 2 => [0.1, 0, 0.9, 0, 0, 0, 0, 0, 0, 0]
y_train_pred = model.predict(train_ds)
Y_train_pred = np.argmax(y_train_pred, 1) # Decode Predicted labels
Y_test = np.argmax(y_test, 1) # Decode Predicted labels
Y_train = np.argmax(y_train, 1) # Decode Predicted labels
f1_score(Y_test, Y_test_pred, average="macro")
f1_score(Y_train, Y_train_pred, average="macro")
with open("/bohr/eisecm-0e5z/v1/AutoECM/data/le_name_mapping.json", "r") as f:
mapping = json.load(f)
le = LabelEncoder()
mapping["classes"] = [mapping[str(int(i))] for i in range(9)]
le.classes_ = np.array(mapping["classes"])
plot_cm(Y_test, Y_test_pred, le, save=False, figname=f"{output_dir}/cm_test")
plot_cm(Y_train, Y_train, le, save=False, figname=f"{output_dir}/cm_train")
proportion_correct = f1_score(Y_test, Y_test_pred, average="macro")
print("Test Accuracy: {}".format(proportion_correct))
# Save model and weights
# model.save(f"{output_dir}/cnn_model.h5")
# Calculate f1 and save classification report
calcualte_classification_report(
Y_train, Y_train_pred, Y_test, Y_test_pred, le, save=False, output_dir=output_dir
)
plt.close("all")
return
代码
文本
定义多状态运行函数
代码
文本
[15]
def calculate_stats_multiple_run(
input_dir: str, output_dir: str, nb_runs: int, input_shape: Tuple[int, int]
) -> pd.DataFrame:
"""Calculate stats for multiple runs of the CNN model."""
y_pred = []
class_reports = []
test_ds = tf.keras.utils.image_dataset_from_directory(
f"{input_dir}/test",
validation_split=None,
seed=SEED,
image_size=input_shape,
batch_size=2000,
label_mode="categorical",
shuffle=False,
color_mode="grayscale",
)
for image_batch, labels_batch in test_ds:
y_test = np.array(labels_batch)
Y_test = np.argmax(y_test, 1)
for i in range(nb_runs):
# read y_pred from file
output_dir_ = f"{output_dir}/{i}/"
y_pred_i = np.loadtxt(f"{output_dir_}/pred_test.txt")
y_pred.append(y_pred_i)
report_dict = classification_report(Y_test, y_pred_i, output_dict=True)
class_reports.append(pd.DataFrame(report_dict))
y_pred = np.array(y_pred)
cnn_stats = pd.DataFrame(
columns=[
"macro avg mean",
"macro avg std",
"weighted avg mean",
"weighted avg std",
],
index=["recall", "f1-score"],
)
recalls_mavg = [class_reports[i].loc["recall", "macro avg"] for i in range(nb_runs)]
recalls_wavg = [
class_reports[i].loc["recall", "weighted avg"] for i in range(nb_runs)
]
f1s_mavg = [class_reports[i].loc["f1-score", "macro avg"] for i in range(nb_runs)]
f1s_wavg = [class_reports[i].loc["f1-score", "weighted avg"] for i in range(nb_runs)]
cnn_stats.loc["recall", "macro avg mean"] = np.mean(recalls_mavg)
cnn_stats.loc["recall", "macro avg std"] = np.std(recalls_mavg)
cnn_stats.loc["f1-score", "macro avg mean"] = np.mean(f1s_mavg)
cnn_stats.loc["f1-score", "macro avg std"] = np.std(f1s_mavg)
cnn_stats.loc["recall", "weighted avg mean"] = np.mean(recalls_wavg)
cnn_stats.loc["recall", "weighted avg std"] = np.std(recalls_wavg)
cnn_stats.loc["f1-score", "weighted avg mean"] = np.mean(f1s_wavg)
cnn_stats.loc["f1-score", "weighted avg std"] = np.std(f1s_wavg)
return cnn_stats
代码
文本
执行模型运行
代码
文本
[16]
# print the tensor flow version
print(tf.__version__)
# if len(sys.argv) < 2:
# filter = 1
# runs = 10
# else:
# filter = int(sys.argv[1])
# runs = 1
# if len(sys.argv) > 2:
# runs = int(sys.argv[2])
filter = 1
runs = 1
print(f"Filter: {filter}")
print(f"Repeating the training {runs} times.")
now = datetime.datetime.now()
now_str = now.strftime("%Y-%m-%d_%H-%M-%S")
if filter:
input_dir = "/bohr/eisecm-0e5z/v1/AutoECM/data/images_filtered"
output_dir = f"/bohr/eisecm-0e5z/v1/AutoECM/results/clf_filtered/cnn/{now_str}"
else:
input_dir = "/bohr/eisecm-0e5z/v1/AutoECM/data/images"
output_dir = f"/bohr/eisecm-0e5z/v1/AutoECM/results/clf/cnn/{now_str}"
# os.mkdir(output_dir)
# if False, then uses ada_delta with decay, else uses stopping and decay based on val_acc
adaptive_based_on_val = True
# If true, the model will be trained nb_runs times and the results will be saved in a folder
# This is useful to assess randomness in the model which can come from GPU usage
input_shape = (58, 58)
nb_classes = 9
# Create new folder with results, name is datetime
if runs == 1:
main(input_shape, nb_classes, input_dir, output_dir, adaptive_based_on_val)
else:
for i in range(runs):
print(f"Run {i+1} of {runs}")
output_dir_ = f"{output_dir}/{i}"
os.mkdir(output_dir_)
main(input_shape, nb_classes, input_dir, output_dir_, adaptive_based_on_val)
cnn_stats = calculate_stats_multiple_run(input_dir, output_dir, runs, input_shape)
print(cnn_stats)
# Save cnn_stats
cnn_stats.to_csv(f"{output_dir}/cnn_stats.txt", sep="\t")
print("Done")
2.11.0
Filter: 1
Repeating the training 1 times.
Found 7040 files belonging to 9 classes.
Using 6336 files for training.
Found 7040 files belonging to 9 classes.
Using 704 files for validation.
Found 1761 files belonging to 9 classes.
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
rescaling_1 (Rescaling) (None, 58, 58, 1) 0
conv2d_5 (Conv2D) (None, 27, 27, 64) 2368
conv2d_6 (Conv2D) (None, 25, 25, 64) 36928
max_pooling2d_3 (MaxPooling (None, 12, 12, 64) 0
2D)
batch_normalization_3 (Batc (None, 12, 12, 64) 256
hNormalization)
conv2d_7 (Conv2D) (None, 10, 10, 128) 73856
conv2d_8 (Conv2D) (None, 8, 8, 128) 147584
max_pooling2d_4 (MaxPooling (None, 4, 4, 128) 0
2D)
batch_normalization_4 (Batc (None, 4, 4, 128) 512
hNormalization)
conv2d_9 (Conv2D) (None, 2, 2, 256) 295168
max_pooling2d_5 (MaxPooling (None, 1, 1, 256) 0
2D)
batch_normalization_5 (Batc (None, 1, 1, 256) 1024
hNormalization)
flatten_1 (Flatten) (None, 256) 0
dense_2 (Dense) (None, 512) 131584
dense_3 (Dense) (None, 9) 4617
=================================================================
Total params: 693,897
Trainable params: 693,001
Non-trainable params: 896
_________________________________________________________________
Epoch 1/200
198/198 [==============================] - 5s 9ms/step - loss: 2.1449 - accuracy: 0.2061 - val_loss: 2.2997 - val_accuracy: 0.1364 - lr: 0.0010
Epoch 2/200
198/198 [==============================] - 2s 7ms/step - loss: 2.0010 - accuracy: 0.2367 - val_loss: 2.1737 - val_accuracy: 0.1562 - lr: 0.0010
Epoch 3/200
198/198 [==============================] - 2s 7ms/step - loss: 1.8845 - accuracy: 0.2677 - val_loss: 1.8519 - val_accuracy: 0.2912 - lr: 0.0010
Epoch 4/200
198/198 [==============================] - 1s 7ms/step - loss: 1.8112 - accuracy: 0.2936 - val_loss: 1.7922 - val_accuracy: 0.3054 - lr: 0.0010
Epoch 5/200
198/198 [==============================] - 1s 7ms/step - loss: 1.7492 - accuracy: 0.3086 - val_loss: 1.7895 - val_accuracy: 0.3026 - lr: 0.0010
Epoch 6/200
198/198 [==============================] - 1s 7ms/step - loss: 1.6989 - accuracy: 0.3278 - val_loss: 1.7652 - val_accuracy: 0.3111 - lr: 0.0010
Epoch 7/200
198/198 [==============================] - 1s 7ms/step - loss: 1.6562 - accuracy: 0.3445 - val_loss: 1.7682 - val_accuracy: 0.3310 - lr: 0.0010
Epoch 8/200
198/198 [==============================] - 1s 7ms/step - loss: 1.6211 - accuracy: 0.3568 - val_loss: 1.7493 - val_accuracy: 0.3324 - lr: 0.0010
Epoch 9/200
198/198 [==============================] - 1s 7ms/step - loss: 1.5755 - accuracy: 0.3804 - val_loss: 1.8102 - val_accuracy: 0.3224 - lr: 0.0010
Epoch 10/200
198/198 [==============================] - 1s 7ms/step - loss: 1.5216 - accuracy: 0.3979 - val_loss: 2.0015 - val_accuracy: 0.2670 - lr: 0.0010
Epoch 11/200
193/198 [============================>.] - ETA: 0s - loss: 1.4858 - accuracy: 0.4137
Epoch 11: ReduceLROnPlateau reducing learning rate to 0.00020000000949949026.
198/198 [==============================] - 1s 7ms/step - loss: 1.4875 - accuracy: 0.4121 - val_loss: 1.8725 - val_accuracy: 0.3111 - lr: 0.0010
Epoch 12/200
198/198 [==============================] - 1s 7ms/step - loss: 1.3166 - accuracy: 0.4923 - val_loss: 2.2674 - val_accuracy: 0.2244 - lr: 2.0000e-04
Epoch 13/200
198/198 [==============================] - 1s 7ms/step - loss: 1.2452 - accuracy: 0.5180 - val_loss: 1.9390 - val_accuracy: 0.3210 - lr: 2.0000e-04
Epoch 14/200
193/198 [============================>.] - ETA: 0s - loss: 1.1811 - accuracy: 0.5492
Epoch 14: ReduceLROnPlateau reducing learning rate to 4.0000001899898055e-05.
198/198 [==============================] - 2s 7ms/step - loss: 1.1784 - accuracy: 0.5507 - val_loss: 1.8511 - val_accuracy: 0.3310 - lr: 2.0000e-04
Epoch 15/200
198/198 [==============================] - 2s 7ms/step - loss: 1.1248 - accuracy: 0.5735 - val_loss: 1.7421 - val_accuracy: 0.3580 - lr: 4.0000e-05
Epoch 16/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0990 - accuracy: 0.5859 - val_loss: 1.7609 - val_accuracy: 0.3551 - lr: 4.0000e-05
Epoch 17/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0782 - accuracy: 0.5939 - val_loss: 1.7784 - val_accuracy: 0.3480 - lr: 4.0000e-05
Epoch 18/200
192/198 [============================>.] - ETA: 0s - loss: 1.0690 - accuracy: 0.5972
Epoch 18: ReduceLROnPlateau reducing learning rate to 1e-05.
198/198 [==============================] - 1s 7ms/step - loss: 1.0648 - accuracy: 0.5993 - val_loss: 1.7926 - val_accuracy: 0.3494 - lr: 4.0000e-05
Epoch 19/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0411 - accuracy: 0.6128 - val_loss: 1.7753 - val_accuracy: 0.3537 - lr: 1.0000e-05
Epoch 20/200
198/198 [==============================] - 2s 7ms/step - loss: 1.0294 - accuracy: 0.6184 - val_loss: 1.7799 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 21/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0230 - accuracy: 0.6203 - val_loss: 1.7848 - val_accuracy: 0.3580 - lr: 1.0000e-05
Epoch 22/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0194 - accuracy: 0.6228 - val_loss: 1.7880 - val_accuracy: 0.3509 - lr: 1.0000e-05
Epoch 23/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0126 - accuracy: 0.6181 - val_loss: 1.7933 - val_accuracy: 0.3509 - lr: 1.0000e-05
Epoch 24/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0123 - accuracy: 0.6269 - val_loss: 1.7965 - val_accuracy: 0.3494 - lr: 1.0000e-05
Epoch 25/200
198/198 [==============================] - 1s 7ms/step - loss: 1.0128 - accuracy: 0.6217 - val_loss: 1.8002 - val_accuracy: 0.3551 - lr: 1.0000e-05
Epoch 26/200
198/198 [==============================] - 2s 7ms/step - loss: 1.0011 - accuracy: 0.6312 - val_loss: 1.8028 - val_accuracy: 0.3551 - lr: 1.0000e-05
Epoch 27/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9958 - accuracy: 0.6266 - val_loss: 1.8041 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 28/200
198/198 [==============================] - 2s 7ms/step - loss: 0.9896 - accuracy: 0.6365 - val_loss: 1.8085 - val_accuracy: 0.3537 - lr: 1.0000e-05
Epoch 29/200
198/198 [==============================] - 2s 7ms/step - loss: 0.9915 - accuracy: 0.6386 - val_loss: 1.8143 - val_accuracy: 0.3608 - lr: 1.0000e-05
Epoch 30/200
198/198 [==============================] - 2s 7ms/step - loss: 0.9815 - accuracy: 0.6370 - val_loss: 1.8165 - val_accuracy: 0.3622 - lr: 1.0000e-05
Epoch 31/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9812 - accuracy: 0.6400 - val_loss: 1.8207 - val_accuracy: 0.3636 - lr: 1.0000e-05
Epoch 32/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9765 - accuracy: 0.6373 - val_loss: 1.8205 - val_accuracy: 0.3636 - lr: 1.0000e-05
Epoch 33/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9730 - accuracy: 0.6411 - val_loss: 1.8257 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 34/200
198/198 [==============================] - 2s 7ms/step - loss: 0.9683 - accuracy: 0.6447 - val_loss: 1.8304 - val_accuracy: 0.3651 - lr: 1.0000e-05
Epoch 35/200
198/198 [==============================] - 2s 7ms/step - loss: 0.9600 - accuracy: 0.6509 - val_loss: 1.8354 - val_accuracy: 0.3622 - lr: 1.0000e-05
Epoch 36/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9566 - accuracy: 0.6471 - val_loss: 1.8387 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 37/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9504 - accuracy: 0.6525 - val_loss: 1.8416 - val_accuracy: 0.3551 - lr: 1.0000e-05
Epoch 38/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9483 - accuracy: 0.6533 - val_loss: 1.8458 - val_accuracy: 0.3608 - lr: 1.0000e-05
Epoch 39/200
198/198 [==============================] - 2s 7ms/step - loss: 0.9404 - accuracy: 0.6588 - val_loss: 1.8507 - val_accuracy: 0.3565 - lr: 1.0000e-05
Epoch 40/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9408 - accuracy: 0.6577 - val_loss: 1.8532 - val_accuracy: 0.3565 - lr: 1.0000e-05
Epoch 41/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9328 - accuracy: 0.6588 - val_loss: 1.8585 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 42/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9278 - accuracy: 0.6610 - val_loss: 1.8623 - val_accuracy: 0.3608 - lr: 1.0000e-05
Epoch 43/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9253 - accuracy: 0.6594 - val_loss: 1.8663 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 44/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9193 - accuracy: 0.6637 - val_loss: 1.8664 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 45/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9127 - accuracy: 0.6664 - val_loss: 1.8735 - val_accuracy: 0.3594 - lr: 1.0000e-05
Epoch 46/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9110 - accuracy: 0.6690 - val_loss: 1.8764 - val_accuracy: 0.3565 - lr: 1.0000e-05
Epoch 47/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9090 - accuracy: 0.6675 - val_loss: 1.8820 - val_accuracy: 0.3551 - lr: 1.0000e-05
Epoch 48/200
198/198 [==============================] - 1s 7ms/step - loss: 0.9090 - accuracy: 0.6730 - val_loss: 1.8864 - val_accuracy: 0.3494 - lr: 1.0000e-05
Epoch 49/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8978 - accuracy: 0.6771 - val_loss: 1.8935 - val_accuracy: 0.3523 - lr: 1.0000e-05
Epoch 50/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8965 - accuracy: 0.6753 - val_loss: 1.8995 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 51/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8913 - accuracy: 0.6757 - val_loss: 1.9026 - val_accuracy: 0.3565 - lr: 1.0000e-05
Epoch 52/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8865 - accuracy: 0.6791 - val_loss: 1.9083 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 53/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8780 - accuracy: 0.6834 - val_loss: 1.9111 - val_accuracy: 0.3480 - lr: 1.0000e-05
Epoch 54/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8734 - accuracy: 0.6843 - val_loss: 1.9166 - val_accuracy: 0.3480 - lr: 1.0000e-05
Epoch 55/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8750 - accuracy: 0.6875 - val_loss: 1.9207 - val_accuracy: 0.3509 - lr: 1.0000e-05
Epoch 56/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8720 - accuracy: 0.6880 - val_loss: 1.9271 - val_accuracy: 0.3494 - lr: 1.0000e-05
Epoch 57/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8637 - accuracy: 0.6872 - val_loss: 1.9311 - val_accuracy: 0.3523 - lr: 1.0000e-05
Epoch 58/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8599 - accuracy: 0.6869 - val_loss: 1.9357 - val_accuracy: 0.3537 - lr: 1.0000e-05
Epoch 59/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8602 - accuracy: 0.6919 - val_loss: 1.9388 - val_accuracy: 0.3537 - lr: 1.0000e-05
Epoch 60/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8539 - accuracy: 0.6866 - val_loss: 1.9399 - val_accuracy: 0.3523 - lr: 1.0000e-05
Epoch 61/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8505 - accuracy: 0.6940 - val_loss: 1.9484 - val_accuracy: 0.3480 - lr: 1.0000e-05
Epoch 62/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8446 - accuracy: 0.6963 - val_loss: 1.9534 - val_accuracy: 0.3480 - lr: 1.0000e-05
Epoch 63/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8392 - accuracy: 0.7009 - val_loss: 1.9591 - val_accuracy: 0.3480 - lr: 1.0000e-05
Epoch 64/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8407 - accuracy: 0.7025 - val_loss: 1.9635 - val_accuracy: 0.3466 - lr: 1.0000e-05
Epoch 65/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8317 - accuracy: 0.7038 - val_loss: 1.9682 - val_accuracy: 0.3494 - lr: 1.0000e-05
Epoch 66/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8245 - accuracy: 0.7098 - val_loss: 1.9708 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 67/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8246 - accuracy: 0.7077 - val_loss: 1.9767 - val_accuracy: 0.3494 - lr: 1.0000e-05
Epoch 68/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8164 - accuracy: 0.7121 - val_loss: 1.9843 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 69/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8128 - accuracy: 0.7128 - val_loss: 1.9858 - val_accuracy: 0.3466 - lr: 1.0000e-05
Epoch 70/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8135 - accuracy: 0.7094 - val_loss: 1.9941 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 71/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8060 - accuracy: 0.7153 - val_loss: 1.9960 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 72/200
198/198 [==============================] - 1s 7ms/step - loss: 0.8025 - accuracy: 0.7145 - val_loss: 2.0003 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 73/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7939 - accuracy: 0.7173 - val_loss: 2.0047 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 74/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7944 - accuracy: 0.7164 - val_loss: 2.0107 - val_accuracy: 0.3409 - lr: 1.0000e-05
Epoch 75/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7919 - accuracy: 0.7183 - val_loss: 2.0178 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 76/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7853 - accuracy: 0.7244 - val_loss: 2.0250 - val_accuracy: 0.3395 - lr: 1.0000e-05
Epoch 77/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7788 - accuracy: 0.7259 - val_loss: 2.0272 - val_accuracy: 0.3423 - lr: 1.0000e-05
Epoch 78/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7839 - accuracy: 0.7229 - val_loss: 2.0356 - val_accuracy: 0.3409 - lr: 1.0000e-05
Epoch 79/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7741 - accuracy: 0.7287 - val_loss: 2.0390 - val_accuracy: 0.3423 - lr: 1.0000e-05
Epoch 80/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7696 - accuracy: 0.7276 - val_loss: 2.0442 - val_accuracy: 0.3423 - lr: 1.0000e-05
Epoch 81/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7638 - accuracy: 0.7304 - val_loss: 2.0492 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 82/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7576 - accuracy: 0.7334 - val_loss: 2.0502 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 83/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7557 - accuracy: 0.7393 - val_loss: 2.0601 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 84/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7557 - accuracy: 0.7366 - val_loss: 2.0622 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 85/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7437 - accuracy: 0.7413 - val_loss: 2.0712 - val_accuracy: 0.3409 - lr: 1.0000e-05
Epoch 86/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7436 - accuracy: 0.7369 - val_loss: 2.0714 - val_accuracy: 0.3494 - lr: 1.0000e-05
Epoch 87/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7346 - accuracy: 0.7465 - val_loss: 2.0790 - val_accuracy: 0.3438 - lr: 1.0000e-05
Epoch 88/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7347 - accuracy: 0.7473 - val_loss: 2.0897 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 89/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7295 - accuracy: 0.7442 - val_loss: 2.0915 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 90/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7300 - accuracy: 0.7451 - val_loss: 2.0987 - val_accuracy: 0.3423 - lr: 1.0000e-05
Epoch 91/200
198/198 [==============================] - 2s 8ms/step - loss: 0.7197 - accuracy: 0.7511 - val_loss: 2.1026 - val_accuracy: 0.3423 - lr: 1.0000e-05
Epoch 92/200
198/198 [==============================] - 1s 7ms/step - loss: 0.7160 - accuracy: 0.7516 - val_loss: 2.1100 - val_accuracy: 0.3409 - lr: 1.0000e-05
Epoch 93/200
198/198 [==============================] - 2s 8ms/step - loss: 0.7131 - accuracy: 0.7551 - val_loss: 2.1166 - val_accuracy: 0.3409 - lr: 1.0000e-05
Epoch 94/200
194/198 [============================>.] - ETA: 0s - loss: 0.7071 - accuracy: 0.7542Restoring model weights from the end of the best epoch: 34.
198/198 [==============================] - 1s 7ms/step - loss: 0.7074 - accuracy: 0.7538 - val_loss: 2.1234 - val_accuracy: 0.3452 - lr: 1.0000e-05
Epoch 94: early stopping
1/1 [==============================] - 0s 358ms/step
198/198 [==============================] - 1s 4ms/step
Test Accuracy: 0.36838940113856067
F1 score train: 0.107
F1 score test: 0.368
precision recall f1-score support
L-R-RCPE 0.3598 0.3907 0.3746 151
L-R-RCPE-RCPE 0.2000 0.2435 0.2196 193
L-R-RCPE-RCPE-RCPE 0.1887 0.1316 0.1550 228
RC-G-G 0.6543 0.6943 0.6737 229
RC-RC-RCPE-RCPE 0.4868 0.5658 0.5233 228
RCPE-RCPE 0.2767 0.2639 0.2701 216
RCPE-RCPE-RCPE 0.1596 0.1327 0.1449 226
RCPE-RCPE-RCPE-RCPE 0.2028 0.1928 0.1977 223
Rs_Ws 0.6629 0.8806 0.7564 67
accuracy 0.3481 1761
macro avg 0.3546 0.3884 0.3684 1761
weighted avg 0.3306 0.3481 0.3372 1761
Done
代码
文本
点个赞吧
本文被以下合集收录
Battery Model
Samuel
更新于 2024-09-09
2 篇0 人关注
推荐阅读
公开
电化学阻抗谱EIS预测 —— 使用四个特征频率下的阻抗预测全频段EISJiaweiMiao
发布于 2024-04-14
2 赞5 转存文件
公开
AI+电芯 | 基于LSTM模型和脉冲电信号的电芯EIS预测JiaweiMiao
发布于 2023-10-22
5 赞40 转存文件