您好,登錄后才能下訂單哦!
這篇文章將為大家詳細講解有關使用Python 怎么實現曲線擬合和求積分,文章內容質量較高,因此小編分享給大家做個參考,希望大家閱讀完這篇文章后對相關知識有一定的了解。
首先要安裝Anaconda python庫,然后來運用這些數學工具。
###最小二乘法試驗### import numpy as np import pymysql from scipy.optimize import leastsq from scipy import integrate ###繪圖,看擬合效果### import matplotlib.pyplot as plt from sympy import * path='E:\PythonProgram\oildata.txt' lieh0 =[] #初始第一列,油管高度 liev1 =[] #初始第二列,油槍記錄的體積 h_median =[] # 高度相鄰中位數 h_subtract =[] #相鄰高度作差 v_subtract =[] #相鄰體積作差 select_h_subtr =[] #篩選的高度作差 ######## select_v_subtr =[] #篩選的體積作差 VdtH=[] #篩選的V 和 t 的 倒數。 def loadData(Spath,lie0,lie1): with open(Spath,'r') as f0: for i in f0: tmp=i.split() lie0.append(float(tmp[0])) lie1.append(float(tmp[2])) print ("source lie0",len(lie0)) def connectMySQL(): db = pymysql.connect(host='10.**.**.**', user='root', passwd='***', db="zabbix", charset="utf8") # 校罐 cur = db.cursor() try: # 查詢 cur.execute("SELECT * FROM station_snapshot limit 10 ") for row in cur.fetchall(): # print(row) id = row[0] snapshot_id = row[1] DateTime = row[13] attr1V = row[5] attr2H = row[6] print("id=%d ,snapshot_id=%s,DateTime=%s,attr1V =%s, attr2H=%s ", (id, snapshot_id, DateTime, attr1V, attr2H)) except: print("Error: unable to fecth data of station_stock") try: cur.execute("SELECT * FROM can_stock limit 5"); for row in cur.fetchall(): # print(row) stockid = row[0] stationid = row[1] DateTime = row[4] Volume = row[5] Height = row[8] print("stockid=%d ,stationid=%s,DateTime=%s,Volume =%f, Height=%f ", (stockid, stationid, DateTime, Volume, Height)) except: print("Error: unable to fecth data of can_snapshot") cur.close() db.close() def formatData(h_med,h_subtr,v_subtr): lh0 = lieh0[:] del lh0[0] print("lh0 size(): ",len(lh0)) lh2 =lieh0[:] del lh2[len(lh2)-1] print("lh2 size() : ",len(lh2)) lv0 =liev1[:] del lv0[0] #print (liev1) print ("Souce liev1 size() : ",len(liev1)) print ("lv1 size() :",len(lv0)) """ lv1 =liev1[:] del lv1[len(lv1)-1] print("lv1 size(): ",len(lv1)) """ h_med[:] =[(x+y)/2 for x,y in zip(lh0,lh2)] ###采樣點(Xi,Yi)### print("h_med size() : ", len(h_med)) h_subtr[:] = [(y-x) for x,y in zip(lh0,lh2)] print("h_subtr size() : ", len(h_subtr)) # v_subtr[:] = [(y-x) for x,y in zip(lv0,lv1)] v_subtr[:] = lv0 print("v_subtr size() : ", len(v_subtr)) def removeBadPoint(h_med,h_sub,v_sub): for val in h_sub: position=h_sub.index(val) if 0.01 > val > -0.01: del h_sub[position] del h_med[position] del v_sub[position] v_dt_h_ay = [(y/x) for x, y in zip(h_sub, v_sub)] return v_dt_h_ay def selectRightPoint(h_med,h_subtr,v_dt_h_ay): for val in v_dt_h_ay: pos = v_dt_h_ay.index(val) if val > 20 : del v_dt_h_ay[pos] del h_med[pos] del h_subtr[pos] for val in v_dt_h_ay: ptr = v_dt_h_ay.index(val) if val < 14: del v_dt_h_ay[ptr] del h_med[ptr] del h_subtr[ptr] def writeFile(h_mp, v_dt_h): s='\n'.join(str(num)[1:-1] for num in h_mp) v='\n'.join(str(vdt)[1:-1] for vdt in v_dt_h) open(r'h_2.txt','w').write(s) open(r'v_dt_h.txt','w').write(v) print("write h_median: ",len(h_mp)) # print("V_dt also is (y-x) : ",v_dt_h,end="\n") print("Write V_dt_h : ",len(v_dt_h)) # file=open('data.txt','w') # file.write(str(h_mp)); # file.close def integralCalculate(coeff,xspace): vCalcute =[] x = Symbol('x') a, b, c, d = coeff[0] y = a * x ** 3 + b * x ** 2 + c * x + d i=0 while (i< len(xspace)-1) : m = integrate(y, (x, xspace[i], xspace[i+1])) vCalcute.append(abs(m)) i=i+1 print("求導結果:",vCalcute) print("求導長度 len(VCalcute): ",len(vCalcute)) return vCalcute ###需要擬合的函數func及誤差error### def func(p,x): a,b,c,d=p return a*x**3+b*x**2+c*x+d #指明待擬合的函數的形狀,設定的擬合函數。 def error(p,x,y): return func(p,x)-y #x、y都是列表,故返回值也是個列表 def leastSquareFitting(h_mp,v_dt_hl): p0=[1,2,6,10] #a,b,c 的假設初始值,隨著迭代會越來越小 #print(error(p0,h_mp,v_dt_h,"cishu")) #目標是讓error 不斷減小 #s="Test the number of iteration" #試驗最小二乘法函數leastsq得調用幾次error函數才能找到使得均方誤差之和最小的a~c Para=leastsq(error,p0,args=(h_mp,v_dt_hl)) #把error函數中除了p以外的參數打包到args中 a,b,c,d=Para[0] #leastsq 返回的第一個值是a,b,c 的求解結果,leastsq返回類型相當于c++ 中的tuple print(" a=",a," b=",b," c=",c," d=",d) plt.figure(figsize=(8,6)) plt.scatter(h_mp,v_dt_hl,color="red",label="Sample Point",linewidth=3) #畫樣本點 x=np.linspace(200,2200,1000) y=a*x**3+b*x**2+c*x+d integralCalculate(Para,h_subtract) plt.plot(x,y,color="orange",label="Fitting Curve",linewidth=2) #畫擬合曲線 # plt.plot(h_mp, v_dt_hl,color="blue", label='Origin Line',linewidth=1) #畫連接線 plt.legend() plt.show() def freeParameterFitting(h_mp,v_dt_hl): z1 = np.polyfit(h_mp, v_dt_hl, 6) # 第一個擬合,自由度為6 # 生成多項式對象 p1 = np.poly1d(z1) print("Z1:") print(z1) print("P1:") print(p1) print("\n") x = np.linspace(400, 1700, 1000) plt.plot(h_mp, v_dt_hl, color="blue", label='Origin Line', linewidth=1) # 畫連接線 plt.plot(x, p1(x), 'gv--', color="black", label='Poly Fitting Line(deg=6)', linewidth=1) plt.legend() plt.show() def main(): loadData(path, lieh0, liev1) connectMySQL() # 讀取oildata數據庫 formatData(h_median, h_subtract, v_subtract) # 去除被除數為0對應的點,并得到v 和 h 求導 值的列表 VdtH[:] = removeBadPoint(h_median, h_subtract, v_subtract) print("h_median1:", len(h_median)) print("VdtH1 : ", len(VdtH)) # 賽選數據,去除異常點 selectRightPoint(h_median, h_subtract, VdtH) print("h_median2:", len(h_median)) print("h_subtract: ", len(h_subtract)) print("VdtH2 : ", len(VdtH)) h_mp = np.array(h_median) v_dt_h = np.array(VdtH) writeFile(h_mp, v_dt_h) # 最小二乘法作圖 leastSquareFitting(h_mp, v_dt_h) # 多項式自由參數法作圖 freeParameterFitting(h_mp, v_dt_h) if __name__ == '__main__': main()
Python是一種跨平臺的、具有解釋性、編譯性、互動性和面向對象的腳本語言,其最初的設計是用于編寫自動化腳本,隨著版本的不斷更新和新功能的添加,常用于用于開發獨立的項目和大型項目。
關于使用Python 怎么實現曲線擬合和求積分就分享到這里了,希望以上內容可以對大家有一定的幫助,可以學到更多知識。如果覺得文章不錯,可以把它分享出去讓更多的人看到。
免責聲明:本站發布的內容(圖片、視頻和文字)以原創、轉載和分享為主,文章觀點不代表本網站立場,如果涉及侵權請聯系站長郵箱:is@yisu.com進行舉報,并提供相關證據,一經查實,將立刻刪除涉嫌侵權內容。