diff --git a/docs/api_userguide.md b/docs/api_userguide.md
index 5931044..9684ebe 100644
--- a/docs/api_userguide.md
+++ b/docs/api_userguide.md
@@ -250,8 +250,10 @@ detect_obj.rate_predict(data)
 ![data_info](images/python_api_rate_predict.png)
 
 ### 三、LIB库
-Metis工程目录下time_series_detector/lib目录为时间序列异常检测学件，可以在python代码或C代码中调用
-libdetect.so在CentOs7.2下编译，目前仅支持在CentOs7.2或更高版本Centos使用
+Metis工程目录下time_series_detector/lib为学件动态库目录，库文件可以在代码中加载调用
+
+libdetect.so目前支持在CentOs7.2+系统环境下使用
+
 
 
 #### Python代码中调用:
@@ -277,7 +279,7 @@ handle = metis_lib.load_model("./xgb_default_model")
 from ctypes import *
 
 class ValueData(Structure):
-    _fields_ = [('value_a', POINTER(c_int)), ('value_b', POINTER(c_int)), ('value_c', POINTER(c_int)),
+    _fields_ = [('data_a', POINTER(c_int)), ('data_b', POINTER(c_int)), ('data_c', POINTER(c_int)),
                 ('len_a', c_int), ('len_b', c_int), ('len_c', c_int)]
 
 # test data
@@ -309,9 +311,9 @@ print result, prob
 	
 ```
 typedef struct {
-    int* value_a;
-    int* value_b;
-    int* value_c;
+    int* data_a;
+    int* data_b;
+    int* data_c;
     int len_a;
     int len_b;
     int len_c;
@@ -322,7 +324,7 @@ typedef struct {
 | 名称  | 类型 |必填| 默认值 | 说明 |
 | --- | --- | --- |---- | --- |
 | handle|  int| 是| 无|模型句柄，由load_model返回|
-| ValueData|  struct| 是| 无|待检测数据|
+| data_value|  ValueData| 是| 无|待检测数据|
 
 
 
@@ -361,7 +363,7 @@ metis_lib = so("./libdetect.so")
 from ctypes import *
 
 class RateData(Structure):
-_fields_ = [('value_a', POINTER(c_double)), ('value_b', POINTER(c_double)), ('value_c', POINTER(c_double)),
+_fields_ = [('data_a', POINTER(c_double)), ('data_b', POINTER(c_double)), ('data_c', POINTER(c_double)),
             ('len_a', c_int), ('len_b', c_int), ('len_c', c_int)]
 
 # test data
@@ -393,9 +395,9 @@ print result, prob
 	
 ```
 typedef struct {
-    double* value_a;
-    double* value_b;
-    double* value_c;
+    double* data_a;
+    double* data_b;
+    double* data_c;
     int len_a;
     int len_b;
     int len_c;
@@ -405,7 +407,7 @@ typedef struct {
 
 | 名称  | 类型 |必填| 默认值 | 说明 |
 | --- | --- | --- |---- | --- |
-| ValueData|  struct| 是| 无|待检测数据|
+| data_value|  RateData| 是| 无|待检测数据|
 
 
 
@@ -424,9 +426,9 @@ typedef struct {
 | result | c_int | 检测结果是否异常。0:异常；1:正常 |
 | prob | c_float | 概率值，值越小，判定为异常的置信度越高 |
 
-####C代码中调用:
+#### C代码中调用:
 
-在C中调用检测函数，需要include头文件detect.h，在编译时链接libdetect.so文件。
+在C中调用检测函数，需要include头文件detect.h，在编译时链接libdetect.so文件
 ##### 1、量值检测
 * 功能说明：根据参考数据检测最近一个数据点是否异常
 
@@ -451,9 +453,9 @@ typedef struct {
 	
 ```
 typedef struct {
-    int* value_a;
-    int* value_b;
-    int* value_c;
+    int* data_a;
+    int* data_b;
+    int* data_c;
     int len_a;
     int len_b;
     int len_c;
@@ -464,7 +466,7 @@ typedef struct {
 | 名称  | 类型 |必填| 默认值 | 说明 |
 | --- | --- | --- |---- | --- |
 | handle|  int| 是| 无|模型句柄，由load_model返回|
-| ValueData|  struct| 是| 无|待检测数据|
+| value_data|  ValueData| 是| 无|待检测数据|
 
 
 
@@ -501,9 +503,9 @@ printf ("ret=%d result =%d prob = %f \n", ret, sample_result, prob);
 	
 ```
 typedef struct {
-    double* value_a;
-    double* value_b;
-    double* value_c;
+    double* data_a;
+    double* data_b;
+    double* data_c;
     int len_a;
     int len_b;
     int len_c;
@@ -513,7 +515,7 @@ typedef struct {
 
 | 名称  | 类型 |必填| 默认值 | 说明 |
 | --- | --- | --- |---- | --- |
-| ValueData|  struct| 是| 无|待检测数据|
+| rate_data|  RateData| 是| 无|待检测数据|
 
 
 * 返回参数：
diff --git a/time_series_detector/common/tsd_common.py b/time_series_detector/common/tsd_common.py
index c8a5064..41ce338 100644
--- a/time_series_detector/common/tsd_common.py
+++ b/time_series_detector/common/tsd_common.py
@@ -78,3 +78,19 @@ def normalize_time_series(split_time_series):
         normalized_data_a
     ]
     return normalized_split_time_series
+
+
+def normalize_time_series_by_max_min(split_time_series):
+    """
+    Normalize the split_time_series by max_min_normalization.
+
+    :param split_time_series: [[data_c_left], [data_c_right], [data_b_left], [data_b_right], [data_a]]
+    :return: max_min_normalized time_series
+    """
+    time_series = split_time_series[0] + split_time_series[1][1:] + split_time_series[2] + split_time_series[3][1:] + split_time_series[4]
+    max_value = np.max(time_series)
+    min_value = np.min(time_series)
+    if max_value - min_value > 0:
+        normalized_time_series = list((np.array(time_series) - min_value) / float(max_value - min_value))
+
+    return normalized_time_series
diff --git a/time_series_detector/feature/classification_features.py b/time_series_detector/feature/classification_features.py
index 6d53dab..4df69bb 100644
--- a/time_series_detector/feature/classification_features.py
+++ b/time_series_detector/feature/classification_features.py
@@ -10,6 +10,8 @@ Unless required by applicable law or agreed to in writing, software distributed
 
 import numpy as np
 import tsfresh.feature_extraction.feature_calculators as ts_feature_calculators
+from time_series_detector.common.tsd_common import DEFAULT_WINDOW, split_time_series
+from statistical_features import time_series_mean, time_series_variance, time_series_standard_deviation, time_series_median
 
 
 def time_series_autocorrelation(x):
@@ -35,6 +37,8 @@ def time_series_autocorrelation(x):
     :return type: float
     """
     lag = int((len(x) - 3) / 5)
+    if np.sqrt(np.var(x)) < 1e-10:
+        return 0
     return ts_feature_calculators.autocorrelation(x, lag)
 
 
@@ -47,6 +51,8 @@ def time_series_coefficient_of_variation(x):
     :return: the value of this feature
     :return type: float
     """
+    if np.sqrt(np.var(x)) < 1e-10:
+        return 0
     return np.mean(x) / np.sqrt(np.var(x))
 
 
@@ -74,15 +80,132 @@ def time_series_binned_entropy(x):
         result.append(ts_feature_calculators.binned_entropy(x, value))
     return result
 
+
+def time_series_value_distribution(x):
+    """
+    Given buckets, calculate the percentage of elements in the whole time series
+    in different buckets
+
+    :param x: normalized time series
+    :type x: pandas.Series
+    :return: the values of this feature
+    :return type: list
+    """
+    thresholds = [0, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.99, 1.0, 1.0]
+    return list(np.histogram(x, bins=thresholds)[0] / float(len(x)))
+
+
+def time_series_daily_parts_value_distribution(x):
+    """
+    Given buckets, calculate the percentage of elements in three subsequences
+    of the whole time series in different buckets
+
+    :param x: normalized time series
+    :type x: pandas.Series
+    :return: the values of this feature
+    :return type: list
+    """
+    thresholds = [0, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.99, 1.0, 1.0]
+    split_value_list = split_time_series(x, DEFAULT_WINDOW)
+    data_c = split_value_list[0] + split_value_list[1][1:]
+    data_b = split_value_list[2] + split_value_list[3][1:]
+    data_a = split_value_list[4]
+    count_c = list(np.histogram(data_c, bins=thresholds)[0])
+    count_b = list(np.histogram(data_b, bins=thresholds)[0])
+    count_a = list(np.histogram(data_a, bins=thresholds)[0])
+    return list(np.array(count_c) / float(len(data_c))) + list(np.array(count_b) / float(len(data_b))) + list(np.array(count_a) / float(len(data_a)))
+
+
+def time_series_daily_parts_value_distribution_with_threshold(x):
+    """
+    Split the whole time series into three parts: c, b, a.
+    Given a threshold = 0.01, return the percentage of elements of time series
+    which are less than threshold
+
+    :param x: normalized time series
+    :type x: pandas.Series
+    :return: 6 values of this feature
+    :return type: list
+    """
+    threshold = 0.01
+    split_value_list = split_time_series(x, DEFAULT_WINDOW)
+    data_c = split_value_list[0] + split_value_list[1][1:]
+    data_b = split_value_list[2] + split_value_list[3][1:]
+    data_a = split_value_list[4]
+
+    # the number of elements in time series which is less than threshold:
+    nparray_data_c_threshold = np.array(data_c)
+    nparray_data_c_threshold[nparray_data_c_threshold < threshold] = -1
+    nparray_data_b_threshold = np.array(data_b)
+    nparray_data_b_threshold[nparray_data_b_threshold < threshold] = -1
+    nparray_data_a_threshold = np.array(data_a)
+    nparray_data_a_threshold[nparray_data_a_threshold < threshold] = -1
+
+    # the total number of elements in time series which is less than threshold:
+    nparray_threshold_count = (nparray_data_c_threshold == -1).sum() + (nparray_data_b_threshold == -1).sum() + (nparray_data_a_threshold == -1).sum()
+
+    if nparray_threshold_count == 0:
+        features = [0, 0, 0]
+    else:
+        features = [
+            (nparray_data_c_threshold == -1).sum() / float(nparray_threshold_count),
+            (nparray_data_b_threshold == -1).sum() / float(nparray_threshold_count),
+            (nparray_data_a_threshold == -1).sum() / float(nparray_threshold_count)
+        ]
+
+    features.extend([
+                    (nparray_data_c_threshold == -1).sum() / float(len(data_c)),
+                    (nparray_data_b_threshold == -1).sum() / float(len(data_b)),
+                    (nparray_data_a_threshold == -1).sum() / float(len(data_a))
+                    ])
+    return features
+
+
+def time_series_window_parts_value_distribution_with_threshold(x):
+    """
+    Split the whole time series into five parts.
+    Given a threshold = 0.01, return the percentage of elements of time series
+    which are less than threshold
+
+    :param x: normalized time series
+    :type x: pandas.Series
+    :return: 5 values of this feature
+    :return type: list
+    """
+    threshold = 0.01
+    split_value_list = split_time_series(x, DEFAULT_WINDOW)
+
+    count_list = []
+    for value_list in split_value_list:
+        nparray_threshold = np.array(value_list)
+        nparray_threshold[nparray_threshold < threshold] = -1
+        count_list.append((nparray_threshold == -1).sum())
+
+    if sum(count_list) == 0:
+        features = [0, 0, 0, 0, 0]
+    else:
+        features = list(np.array(count_list) / float((DEFAULT_WINDOW + 1)))
+
+    return features
+
+
 # add yourself classification features here...
 
 
 def get_classification_features(x):
     classification_features = [
+        time_series_mean(x),
+        time_series_variance(x),
+        time_series_standard_deviation(x),
+        time_series_median(x),
         time_series_autocorrelation(x),
         time_series_coefficient_of_variation(x)
     ]
+    classification_features.extend(time_series_value_distribution(x))
+    classification_features.extend(time_series_daily_parts_value_distribution(x))
+    classification_features.extend(time_series_daily_parts_value_distribution_with_threshold(x))
+    classification_features.extend(time_series_window_parts_value_distribution_with_threshold(x))
     classification_features.extend(time_series_binned_entropy(x))
-    # append yourself classification features here...
+    # add yourself classification features here...
 
     return classification_features
diff --git a/time_series_detector/feature/feature_service.py b/time_series_detector/feature/feature_service.py
index 19379cc..5e8e237 100644
--- a/time_series_detector/feature/feature_service.py
+++ b/time_series_detector/feature/feature_service.py
@@ -34,9 +34,10 @@ def extract_features(time_series, window):
     split_time_series = tsd_common.split_time_series(time_series, window)
     # nomalize time_series
     normalized_split_time_series = tsd_common.normalize_time_series(split_time_series)
+    max_min_normalized_time_series = tsd_common.normalize_time_series_by_max_min(split_time_series)
     s_features = statistical_features.get_statistical_features(normalized_split_time_series[4])
     f_features = fitting_features.get_fitting_features(normalized_split_time_series)
-    c_features = classification_features.get_classification_features(normalized_split_time_series[0] + normalized_split_time_series[1][1:] + normalized_split_time_series[2] + normalized_split_time_series[3][1:] + normalized_split_time_series[4])
+    c_features = classification_features.get_classification_features(max_min_normalized_time_series)
     # combine features with types
     features = s_features + f_features + c_features
     return features
diff --git a/time_series_detector/feature/fitting_features.py b/time_series_detector/feature/fitting_features.py
index 9a146b6..d78702e 100644
--- a/time_series_detector/feature/fitting_features.py
+++ b/time_series_detector/feature/fitting_features.py
@@ -49,7 +49,7 @@ def time_series_weighted_moving_average(x):
     for w in range(1, min(50, DEFAULT_WINDOW), 5):
         w = min(len(x), w)  # avoid the case len(value_list) < w
         coefficient = np.array(range(1, w + 1))
-        temp_list.append((np.dot(coefficient, x[-w:])) / (w * (w + 1) / 2))
+        temp_list.append((np.dot(coefficient, x[-w:])) / float(w * (w + 1) / 2))
     return list(np.array(temp_list) - x[-1])
 
 
@@ -210,6 +210,11 @@ def time_series_periodic_features(data_c_left, data_c_right, data_b_left, data_b
             periodic_features.append(-1)
         else:
             periodic_features.append(1)
+
+    step = DEFAULT_WINDOW / 6
+    for w in range(1, DEFAULT_WINDOW, step):
+        periodic_features.append(min(max(data_a[w - 1:w + step]) - data_a[-1], 0))
+        periodic_features.append(max(min(data_a[w - 1:w + step]) - data_a[-1], 0))
     return periodic_features
 
 # add yourself fitting features here...
diff --git a/time_series_detector/model/xgb_default_model b/time_series_detector/model/xgb_default_model
index 362a7d9..7547402 100644
Binary files a/time_series_detector/model/xgb_default_model and b/time_series_detector/model/xgb_default_model differ