123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157 |
- # 获取数据
- from LIB.BACKEND import DBManager
- import os
- import pandas as pd
- import numpy as np
- import datetime
- # import matplotlib.pyplot as plt
- #参数初始化
- Capacity = 53.6
- PackFullChrgVolt=69.99
- CellFullChrgVolt=3.6
- CellVoltNums=20
- CellTempNums=4
- FullChrgSoc=98
- PeakSoc=57
- #获取数据时间段
- def cal_deltsoc(sn, end_time, start_time):
- end_time = end_time
- strat_time = start_time
- SNnum=str(sn)
- sn = sn
- st = strat_time
- et = end_time
- dbManager = DBManager.DBManager()
- df_data = dbManager.get_data(sn=sn, start_time=st, end_time=et, data_groups=['bms'])
- df_bms = df_data['bms']
- # 计算电芯Soc差
- packcrnt = df_bms['总电流[A]']
- packvolt = df_bms['总电压[V]']
- SOC = df_bms['SOC[%]']
- SOH = df_bms['SOH[%]']
- bmsstat = (df_bms['充电状态']).astype(int)
- time = pd.to_datetime(df_bms['时间戳'], format='%Y-%m-%d %H:%M:%S')
- # 筛选充电数据
- ChgStart = []
- ChgEnd = []
- for i in range(3, len(time) - 3):
- if i==3 and bmsstat[i]==2 and bmsstat[i+1]==2 and bmsstat[i+2]==2:
- ChgStart.append(i)
- elif bmsstat[i-2]!=2 and bmsstat[i-1]!=2 and bmsstat[i]==2:
- ChgStart.append(i)
- elif bmsstat[i-1]==2 and bmsstat[i]!=2 and bmsstat[i+1]!=2:
- ChgEnd.append(i)
- elif i == (len(time) - 4) and bmsstat[len(bmsstat)-1] == 2 and bmsstat[len(bmsstat)-2] == 2:
- ChgEnd.append(len(time)-1)
- # 筛选充电起始Soc<46%,电芯温度>15℃的数据
- if ChgStart:
- ChgStartValid = []
- ChgEndValid = []
- for i in range(min(len(ChgStart),len(ChgEnd))):
- # 获取最小温度值
- celltemp = []
- for j in range(1, CellTempNums + 1):
- s = str(j)
- temp = df_bms['单体温度' + s]
- celltemp.append(temp[ChgEnd[i]])
- if SOC[ChgStart[i]] < 46 and SOC[ChgEnd[i]]>80 and min(celltemp) > 10:
- if ((time[ChgEnd[i]]-time[ChgStart[i]]).total_seconds())/(ChgEnd[i]-ChgStart[i])<30:
- ChgStartValid.append(ChgStart[i])
- ChgEndValid.append(ChgEnd[i])
- # 计算充电每个单体到达DVDQ峰值的Ah差
- # 定义滑动平均滤波函数
- def np_move_avg(a, n, mode="same"):
- return (np.convolve(a, np.ones((n,)) / n, mode=mode))
- # 定义函数:切片Soc>50且Soc<80,并寻找峰值返回峰值点的时间
- def data_search(t, soc, cellvolt1, packcrnt1):
- cellvolt2 = np_move_avg(cellvolt1, 5, mode="same")
- Soc = 0
- Ah = 0
- Volt = [cellvolt2[0]]
- DV_Volt = []
- DQ_Ah = []
- DVDQ = []
- time1 = []
- soc1 = []
- soc2 = []
- for m in range(1, len(t)):
- Step = (t[m] - t[m - 1]).total_seconds()
- Soc = Soc - packcrnt1[m] * Step * 100 / (3600 * Capacity)
- Ah = Ah - packcrnt1[m] * Step / 3600
- if (cellvolt2[m] - Volt[-1]) > 0.9 and Ah>0:
- DQ_Ah.append(Ah)
- Volt.append(cellvolt2[m])
- DV_Volt.append(Volt[-1] - Volt[-2])
- DVDQ.append((DV_Volt[-1]) / DQ_Ah[-1])
- Ah = 0
- time1.append(t[m])
- soc1.append(Soc)
- soc2.append(soc[m])
- df_Data1 = pd.DataFrame({'Time': time1,
- 'SOC': soc2,
- 'DVDQ': DVDQ,
- 'AhSOC': soc1})
- df_Data1 = df_Data1[(df_Data1['SOC'] > 50) & (df_Data1['SOC'] < 75)]
- # 寻找峰值点,且峰值点个数>3
- if len(df_Data1['DVDQ'])>2:
- PeakIndex = df_Data1['DVDQ'].idxmax()
- df_Data2 = df_Data1[
- (df_Data1['SOC'] > (df_Data1['SOC'][PeakIndex] - 0.5)) & (df_Data1['SOC'] < (df_Data1['SOC'][PeakIndex] + 0.5))]
- if len(df_Data2) > 3:
- return df_Data1['AhSOC'][PeakIndex]
- else:
- df_Data1 = df_Data1.drop([PeakIndex])
- PeakIndex = df_Data1['DVDQ'].idxmax()
- return df_Data1['AhSOC'][PeakIndex]
- # 计算最大最小Soc差
- if ChgStartValid:
- DetaSoc2 = []
- DetaSoc=[]
- DetaSoc_SN=[]
- DetaSoc_time=[]
- for i in range(len(ChgStartValid)):
- DetaSoc1 = []
- for j in range(1, CellVoltNums + 1):
- s = str(j)
- cellvolt = df_bms['单体电压' + s]
- cellvolt = list(cellvolt[ChgStartValid[i]:ChgEndValid[i]])
- Time = list(time[ChgStartValid[i]:ChgEndValid[i]])
- Packcrnt = list(packcrnt[ChgStartValid[i]:ChgEndValid[i]])
- SOC1 = list(SOC[ChgStartValid[i]:ChgEndValid[i]])
- a = data_search(Time, SOC1, cellvolt, Packcrnt)
- if a:
- DetaSoc1.append(a) # 计算到达峰值点的累计Soc
- if DetaSoc1:
- DetaSoc2.append(max(DetaSoc1) - min(DetaSoc1))
- DetaSocMean = np.mean(DetaSoc2)
- DetaSoc.append(DetaSocMean)
- DetaSoc_SN.append(SNnum)
- DetaSoc_time.append(time[ChgStartValid[-1]])
- result_DetaSoc={'time':DetaSoc_time,
- 'SN号':DetaSoc_SN,
- 'Soc差':DetaSoc}
- Result_DetaSoc=pd.DataFrame(result_DetaSoc)
- return Result_DetaSoc
- return pd.DataFrame()
|