Eine Grafische Auswertung von Daten ist nicht immer Einfach.
Ich habe leider nichts gefunden mit dem ich einfach aus einer mySQL-DB eine Grafik erstellen kann. Daher habe ich dann aus mehreren Seite im Internet eine Python-Script zusammenkopiert. Die erstellten Diagramme können Sie jedoch sehen lassen: Heizung
Der Link verweist auf eine Kopie der aktuellen Daten. Das Beagle-Board selbst hat keinen direkten Internetzugang.
Nachfolgend das Python-Script:
# Get DS18B20 Temperatur-Data from mySQL and use mathplot # 4.4.2015 import sys import os import time import math import datetime import MySQLdb as mdb import numpy import matplotlib matplotlib.use('agg') matplotlib.rcParams['figure.dpi'] = 200 matplotlib.rcParams['figure.figsize'] = [13.65, 10.24] matplotlib.rcParams['lines.linewidth'] = 0.5 matplotlib.rcParams['axes.color_cycle'] = ['r', 'g', 'b', 'c', 'm', 'k'] matplotlib.rcParams['axes.labelsize'] = 'large' matplotlib.rcParams['font.size'] = 8 matplotlib.rcParams['grid.linestyle']='-' import matplotlib.pyplot as plt # Check if ther is a commandline-parameter if (len(sys.argv)) == 2 and int(sys.argv[1]) > 0: TimeValue = sys.argv[1] else: TimeValue = "24" #print TimeValue DBconn = mdb.connect('localhost', 'root', 'XXXXXXXXXX', 'measure') cursor = DBconn.cursor(mdb.cursors.DictCursor) sql1 = "select M.SensorID as id, S.name as name" sql1 += " from measure M inner join sensor S on ( S.id = M.SensorID )" sql1 += " where M.ts > (DATE_SUB(now(), INTERVAL " + TimeValue + " HOUR)) group by M.SensorID" cursor.execute(sql1) Sensors = cursor.fetchall() Sens = len ( Sensors ) +1 CurSens = 1 cursor = DBconn.cursor() sql1 = "select unix_timestamp(M.ts) as ts, M.value / 1000 as value " sql1 += "from measure M " sql1 += "where M.ts > (DATE_SUB(now(), INTERVAL " + TimeValue + " HOUR)) and M.SensorID = " tempmax = 0 tempmin = 0 Title = "Temperaturen unserer Heizung: " + TimeValue + " Stunden" for Sensor in Sensors: #print Sensor['id'], Sensor['name'], #Title += Sensor['name'] + ", " sql2 = sql1 + str(Sensor['id']) + " order by TS desc" cursor.execute(sql2) Raw = cursor.fetchall() Raw = numpy.fromiter(Raw, count=-1, dtype=[('', numpy.float)]*2) Raw = Raw.view(numpy.float).reshape(-1, 2) for value in Raw: if tempmax < value[1]: tempmax = value[1] if tempmin > value[1]: tempmin = value[1] (samples,ports)=Raw.shape #print 'Samples: {}, DataPoints: {}'.format(samples,ports) plotme=numpy.zeros((samples,ports-1)) # make an array the same shape minus the epoch numbers for y in range(ports-1): for x in range(samples-1): seconds = Raw[x+1,0]-Raw[x,0] plotme[x,y] = Raw[x,y+1] plotme[samples-1,y] = None dts = map(datetime.datetime.fromtimestamp, Raw[:,0]) lines = plt.plot(dts, plotme, label=Sensor['name']) #plt.text(dts[0], plotme[0], Sensor['name'], fontsize=8) position = ((samples / ( Sens )) * ( CurSens )) # if the calculation of the position does not work??!! if position > samples: position = samples / 2 # print samples, position plt.annotate(Sensor['name'], xy=(dts[position], plotme[position]), xycoords='data', xytext=(15, 15), textcoords='offset points', arrowprops=dict(arrowstyle="->", connectionstyle="angle,angleA=0,angleB=90,rad=10"), ) CurSens += 1 DBconn.close() #print plotme # set up the plot details we want plt.grid(True) plt.ylabel('Temp $^\circ$C') tempmax = int(( tempmax + 8 ) / 5 ) * 5 if tempmin < 0 : tempmin = int(( tempmin - 3 ) / 5) * 5 print "Max: " + str(tempmax) print "Min: " + str(tempmin) plt.axis(ymax=(tempmax) ,ymin=(tempmin) ) plt.xlabel(time.asctime()) plt.title(Title) plt.hold(True) # format Time and Date plt.gca().xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%d.%m. %H:%M')) plt.gca().xaxis.set_major_locator(matplotlib.dates.AutoDateLocator()) #lines = plt.plot(dts,plotme) plt.gcf().autofmt_xdate() FileName = "/tmp/" + TimeValue + "hour.png" plt.savefig(FileName) Destination = '/var/www/' os.system("cp " + FileName + " {}".format(Destination))