python实现多种风机数据清洗，绘制功率曲线，包括kmeans，DBSCAN，KernelDensity等方法
#分区间进行dbscan聚类 def dbscan_filter(df,eps=0.6,min_samples=1.5): x = df.wind_speed y = df.power ws_min = min(x)-0.1 ws_max = max(x) bin_num = int((ws_max-ws_min)/0.5) 1 # 风速以0.5为间隔 bins = [i*0.5 ws_min for i in range(bin_num 1)] s = pd.cut(x,bins,labels=list(range(bin_num))) # 划分区间 df['bin'] = s df_group = df.groupby('bin', sort=False) # 分组 norm_index = [] abnorm_index = [] for _, data in df_group: data_filter = data.copy() if len(data_filter) == 1: norm_index = data_filter.index.tolist() elif len(data_filter)>=2: ws = data['wind_speed'].tolist() pw = data['power'].tolist() cnt_raw = len(ws) raw = np.array([[ws[i], pw[i]] for i in range(0, cnt_raw)]) db = DBSCAN(eps=eps, min_samples=min_samples).fit(raw)#eps为半径，min_samples为最小的样本数 labels = db.labels_ data_ = data.copy() data_['cluster_db'] = labels # 在数据集最后一列加上经过DBSCAN聚类后的结果，-1为临界点或离群点 norm_index = data_.loc[data_.cluster_db!=-1].index.tolist() abnorm_index = data_.loc[data_.cluster_db==-1].index.tolist() ws_n = df.loc[norm_index,'wind_speed'] pw_n = df.loc[norm_index,'power'] ws_abn = df.loc[abnorm_index,'wind_speed'] pw_abn = df.loc[abnorm_index,'power'] return ws_n,pw_n,ws_abn,pw_abn
评论