用神经网络学习Fe原子光谱并反向求导计算权重

最新推荐文章于 2023-07-16 15:08:31 发布

黑榆

最新推荐文章于 2023-07-16 15:08:31 发布

阅读量659

点赞数

分类专栏：用神经网络模拟原子文章标签：原子发射光谱概率幅神经网络学习原子光谱

本文链接：https://blog.csdn.net/georgesale/article/details/80238432

版权

用神经网络模拟原子专栏收录该内容

358 篇文章 5 订阅

订阅专栏

首先构造一个1*26*26的神经网络

根据铁的发射光谱

强度	*	波长10^-10m	*	归一化
1000	P	2483.2708	y0	0.566498
600	P	2488.1426	y1	0.56761
500	P	2490.6443	y2	0.56818
400	P	2522.8494	y3	0.575527
400	P	2719.0273	y4	0.62028
300	P	2788.1047	y5	0.636039
400	P	3440.606	y6	0.784891
600	P	3581.1931	y7	0.816963
600	P	3719.9348	y8	0.848613
700	P	3734.8638	y9	0.852019
600	P	3737.1316	y10	0.852536
600	P	3745.5613	y11	0.854459
400	P	3749.4854	y12	0.855355
300	P	3758.2329	y13	0.85735
500	P	3820.4253	y14	0.871538
500	P	3859.9114	y15	0.880546
300	P	4045.8125	y16	0.922955
200	P	4383.5449	y17	1

让x0=sigmoid（26），让y0的目标函数等于0.566498（ 2483.2708 / 4383.5449），并依次让y17的目标函数等于1. y18-y25=0，让学习率等于0.1，采用随机梯度下降法迭代。

这个网络是可以收敛的比如其中的一组数据

输出	*	目标函数	迭代次数	换算输出	真实值	*	误差
0.565266	0	0.566498315	339764	2477.867	2483.271	y0	0.002176
0.567686	1	0.567609699	339764	2488.478	2488.143	y1	0.000135
0.568894	2	0.568180401	339764	2493.772	2490.644	y2	0.001256
0.57364	3	0.575527218	339764	2514.577	2522.849	y3	0.003279
0.61905	4	0.620280472	339764	2713.635	2719.027	y4	0.001983
0.636826	5	0.636038814	339764	2791.556	2788.105	y5	0.001238
0.786496	6	0.784891242	339764	3447.643	3440.606	y6	0.002045
0.817404	7	0.816962796	339764	3583.126	3581.193	y7	0.00054
0.850477	8	0.848613368	339764	3728.104	3719.935	y8	0.002196
0.852941	9	0.852019059	339764	3738.904	3734.864	y9	0.001082
0.852904	10	0.852536403	339764	3738.743	3737.132	y10	0.000431
0.855403	11	0.854459435	339764	3749.696	3745.561	y11	0.001104
0.856492	12	0.855354624	339764	3754.471	3749.485	y12	0.00133
0.857735	13	0.857350155	339764	3759.921	3758.233	y13	0.000449
0.873373	14	0.871537851	339764	3828.469	3820.425	y14	0.002106
0.882092	15	0.880545652	339764	3866.688	3859.911	y15	0.001756
0.924514	16	0.922954502	339764	4052.649	4045.813	y16	0.00169
0.999991	17	1	339764	4383.504	4383.545	y17	9.26E-06

在网络迭代了339764次以后得到这组数据，y0-y17的输出值与目标函数的误差最大小于0.4%，误差平均0.14%。比如第一组数据目标函数是0.566498315，网络输出是0.565266，将这两个值都*4383.545，得到输出值2477.867，真实测量的值是2483.271，误差是0.2176%。

并且可以很容易的导出当网络收敛时的权重值。按照平方映射的原理，神经网络的权重与输出是一一对应的关系，也就是对一个输出与其对应的权重的解是唯一的。

也就是说函数

f（26，w1，w2）=[y0，y17]

的解数组w1，w2是唯一的。

W1[1*26]第一层权重

W2[26*26]第二层权重

铁在激发态的发射光谱是唯一的，与之对应的概率幅应该也是唯一的。现在解Fe的薛定谔方程是不可能的，但解一个1*26*26的神经网络是可能的，如果把权重理解成是概率幅这个1*26*26的假想原子的概率幅就是数组w1。

w1	w2
-289231	13440.09
-184276	5746.641
-150364	-3681.14
-315329	-13240.5
-308674	-13222.6
-187156	61710.62
-232312	25977.4
-149144	1457.193
-331982	-4799.87
-9.96015	39776.41
-308328	7400.994
-316422	23767.48
-309505	32589.59
-89258.6	46552.49
-99351.1	38326.61
-162172	126244.2
-31395.4	72932.22
-239448	58161.51
-325371	-129198
-389070	-94858.2
-175706	-104918
-206652	-113942
-188505	-138800
-280748	-123848
-207741	-126320
-238336	-147196
	34878.72
	29346.68
	-17743.7
	14090.25
	-1928.21
	74673.72
	38424.44
	21401.24
	8041.499
	47601.46
	-29088.4
	-28228.9
	4495.003
	6256.264
	3489.706
	57591.33
	105921.2
	67575.13
	-132246
	-94526.6
	-113108
	-117161
	-133005
	-128110
	-124757
	-138847
	12553.36
	8697.032
	-8757.7
	1114.923
	1649.876
	24914.72
	19867.4
	16915.61
	16274.64
	25312.21
	28995.74
	41878.13
	31661.42
	42582.78
	33261.61
	74445.68
	22340.78
	24146.83
	-68853.2
	-41682.3
	-60919.5
	-61177
	-61013.6
	-67413.6
	-64074.2
	-61696.8
	17144.49
	11957.47
	-9903.28
	-14209.4
	-19048.3
	64128.1
	25608.63
	-1193.35
	-6075.69
	38875.13
	-233.775
	24321.83
	24884.13
	40816.1
	30174.92
	131885.9
	85056.81
	67387.23
	-163971
	-107928
	-142048
	-145082
	-155942
	-158832
	-153632
	-160109
	7402.726
	5382.465
	-7468.57
	-9765.08
	-3514.59
	58885.49
	26058.19
	4135.339
	8605.426
	36730.19
	31609.26
	59129.38
	50163.73
	69423.76
	52633.26
	149443.2
	57875.75
	52487.78
	-144531
	-90199.5
	-126735
	-127869
	-131648
	-140167
	-134523
	-133676
	19218.97
	14972.73
	-13098.4
	-5562.89
	-11048
	30910.07
	24608.65
	20978
	14304.34
	35232.98
	19860.11
	35075.26
	25597.94
	38761.39
	29535.95
	85661.85
	48124.46
	43343.23
	-107806
	-68081.1
	-95498.9
	-95807.3
	-97629.3
	-105486
	-99882.4
	-99158.9
	48203.37
	29100.31
	-14832.4
	-6878.44
	-26974.9
	46153.77
	48171.35
	73608.98
	36427.92
	80652.17
	-7061.46
	-41091
	17810.44
	17218.54
	21821.69
	25492.75
	118774.4
	81588.86
	-108996
	-101524
	-82996.6
	-96469.2
	-139470
	-104168
	-110862
	-154688
	17423.11
	35814.79
	-6881.36
	7360.52
	18821.53
	7765.998
	83040.39
	76254.57
	69664.26
	101626.1
	97091.59
	86469.19
	95904.32
	98033.71
	102040.1
	65192.8
	45933.59
	157684.1
	-115257
	-158816
	-92536.3
	-114219
	-175995
	-125579
	-129516
	-208219
	1545.673
	8400.948
	-10857.3
	952.6439
	10206.8
	61997.5
	25241.1
	3808.554
	18224.17
	30874.29
	37400.02
	69051.72
	54584.66
	74190.62
	52055.38
	145045.3
	53471.71
	47635.22
	-150416
	-81975.5
	-140083
	-133434
	-117997
	-147221
	-133887
	-114198
	4649.505
	4702.019
	4587.002
	5212.054
	8364.657
	9651.707
	22400.04
	26041.01
	30124.08
	30561.65
	30705.45
	30952.87
	31337.41
	31637.91
	33698.22
	35174.17
	44386.12
	226564.8
	-228219
	-225025
	-227148
	-231361
	-233101
	-239739
	-227776
	-239525
	52035.31
	32856.45
	-19991.1
	18818.39
	814.3729
	101355
	55014.28
	51523.93
	31064.8
	73632.97
	-30028.2
	-52511.8
	23145.56
	15889.04
	16278.21
	54739.8
	137544.3
	83833.1
	-146056
	-111760
	-121714
	-129022
	-156236
	-140310
	-140177
	-166040
	9213.466
	4361.307
	-3524.05
	6123.005
	16600.3
	85422.57
	33350.49
	354.9062
	8633.272
	40596.96
	33615.44
	56354.05
	63238.54
	80619.22
	63428.38
	162250.3
	62843.14
	51074.04
	-132280
	-89761.2
	-110442
	-116605
	-132719
	-126978
	-127334
	-138640
	7004.421
	9723.781
	-11834.4
	4467.904
	12590.27
	64087.16
	28024.76
	6912.146
	20186.44
	33466.08
	42195.79
	73485.8
	58558.33
	78665.17
	56534.08
	149946.6
	51125.88
	46744.12
	-148125
	-81476.3
	-137035
	-131369
	-117866
	-144864
	-132701
	-114774
	35900.91
	40081.74
	-17945.5
	-10149.9
	-24179.7
	24370.28
	55907.63
	82169.72
	51242.74
	87669.74
	20863.86
	7657.828
	25117.61
	36531.57
	31552.98
	47092.71
	107150.2
	92600
	-148226
	-115600
	-127121
	-133001
	-155327
	-146506
	-141339
	-165403
	12859.11
	37619.66
	-18245.8
	-10332.2
	-12099.6
	15377.91
	48067.23
	62273.55
	42682.47
	63883.65
	43234.9
	54323.33
	30925.88
	50642.69
	41085.14
	65248.93
	45510.22
	52641.71
	-96908.2
	-76134.7
	-77593.8
	-88135.9
	-109058
	-98664.7
	-96894
	-117113
	22825.76
	17134.89
	-13886.6
	-22728.6
	-32244.4
	-1524.18
	29358.27
	44342.43
	20261.71
	48645.27
	29943.52
	38850.8
	20175.39
	35168.2
	31759.42
	71177.56
	38746.89
	44928.76
	-87195.1
	-68419.5
	-69021.9
	-77775.8
	-98989.4
	-85361.4
	-87276.1
	-106615
	41125.82
	51626.06
	-33500.2
	-24125.6
	-43478.8
	-10845.3
	44229.09
	84828.62
	51480.22
	74952.73
	52296.76
	62026.75
	20098.91
	33963.26
	47856.54
	25742.37
	76574.29
	119045.4
	-123962
	-129419
	-101045
	-115064
	-155374
	-126276
	-129794
	-171609
	-85902.5
	-25158.2
	76440.13
	-52602.8
	29253.15
	-4454.2
	19304.26
	75014.19
	86767.77
	112646.7
	145740.3
	78813.91
	140008.8
	187674.8
	148589.3
	95599.59
	150220.8
	140633.6
	-117499
	-139229
	-97341.1
	-105208
	-160190
	-111607
	-116344
	-186222
	-93090.6
	-31680.8
	87593.66
	-49976.1
	41725.92
	8538.181
	21905.49
	75180.55
	90450.28
	120976.6
	164210.9
	88170.44
	161841.1
	214025.4
	171666.6
	112675.6
	158302.3
	146657.9
	-108050
	-144825
	-83440.5
	-96536.1
	-166811
	-101009
	-112042
	-198277
	5696.335
	10474.02
	-7181.84
	9643.025
	17282.09
	101495.4
	32138.04
	-13039.9
	496.1402
	36699.52
	14200.66
	45406.29
	51737.87
	70141.7
	49928.36
	168010.8
	87931.66
	64589.25
	-172864
	-108588
	-151783
	-152759
	-157852
	-167228
	-160383
	-160317
	24226.68
	17457.24
	-17615.8
	-31166.5
	-40528.2
	14857.36
	32210.46
	39893.35
	18333.73
	54678.06
	32032.34
	52880.43
	29017.32
	49460.85
	40920.75
	116589.4
	58704.66
	61916.61
	-143471
	-98066.6
	-121935
	-127583
	-141956
	-140274
	-136802
	-147552
	15114.04
	8729.593
	-6724.16
	-22544.7
	-25961.1
	32414.23
	27477.76
	21721.96
	5179.487
	45878.7
	15316.83
	26068.76
	26350.52
	40224.01
	35477.78
	101227.3
	60941.43
	54264.52
	-108763
	-85892.3
	-84943.5
	-96170.3
	-125452
	-104569
	-109000
	-135235
	10147.83
	7712.585
	-4616.55
	-14169.6
	-15336.7
	37159.41
	23214.36
	14387.2
	4406.055
	37435.1
	8118.148
	15608.35
	22927.4
	33363.03
	28232.04
	82892.98
	58838.11
	47852.96
	-95864
	-73928.6
	-76701
	-84756.1
	-106943
	-92211.2
	-94576.5
	-114449
	11294.53
	9045.65
	-4870.6
	-8602.18
	-11525.4
	63683.67
	23311.12
	-5010.18
	-10545.8
	33939.74
	-10249.8
	5310.051
	18277.84
	28995.02
	22365.24
	106064.2
	80878.04
	58604.43
	-128281
	-93029.9
	-105602
	-113176
	-135164
	-123151
	-124041
	-142447
	11991.31
	14197.69
	-11015.7
	6317.796
	4040.614
	52657
	22172.58
	4776.338
	7775.326
	26648.3
	4546.519
	20721.85
	22886.13
	32736.44
	21631.08
	83241.72
	59419.27
	43390.68
	-113153
	-67802.3
	-102827
	-100355
	-96549.6
	-110433
	-102475
	-96114.5
	12242.91
	7614.779
	-4394.28
	-3674.84
	-2296.06
	55946.06
	26251.93
	8657.254
	5245.345
	37446.02
	13198.73
	23937.86
	34419.45
	46025.8
	37246.43
	103800.2
	62617.83
	48950.08
	-106807
	-77880.6
	-87704
	-94293.2
	-113027
	-102598
	-103806
	-119646