C#实现的bp神经网络并应用于综合评价
来源：互联网 发布：东方财富网 mac 编辑：程序博客网 时间：2024/06/05 18:32
由于课程设计选的题目是基于神经网络的综合评价，利用暑假时间用C#实现的bp神经网络。其中用到的_Matrix类是C#实现的矩阵类http://blog.csdn.net/lanqiuchaoren/article/details/37738665。此bp神经网络包含1个隐藏层，其中输入层，隐藏层，输出层个数都可以根据需要更改。
具体bp神经网络代码如下
BP类：
using Matrix_Mul;using Excel = Microsoft.Office.Interop.Excel;using System;using System.Collections.Generic;using System.IO;using System.Linq;using System.Reflection;using System.Text;using System.Threading.Tasks;using Microsoft.Office.Interop.Excel;namespace BPNETSerial{ public class BP { /// <summary> /// 判断是否训练过网络 /// </summary> Boolean IsTrained; /// <summary> /// 用于矩阵的相关计算 /// </summary> _Matrix_Calc matrix_Calc; /// <summary> /// 输入层节点数 /// </summary> int innum; /// <summary> /// 测试数据组数 /// </summary> int train_num; /// <summary> /// 训练数据组数 /// </summary> int test_num; public int Test_num { get { return test_num; } set { test_num = value; } } /// <summary> /// 测试数据维度; /// </summary> int sampdim; /// <summary> /// 隐藏层节点数 /// </summary> int midnum; /// <summary> /// 输出层节点数 /// </summary> int outnum; /// <summary> /// 迭代次数 /// </summary> int iteration; /// <summary> /// 输入层与隐藏层间的权值 /// </summary> _Matrix w1; /// <summary> /// 输入层与隐藏层间的阀值 /// </summary> _Matrix b1; /// <summary> /// 输出层与隐藏层间的权值 /// </summary> _Matrix w2; /// <summary> /// 输出层与隐藏层间的阀值 /// </summary> _Matrix b2; /// <summary> /// 保存w1的值 /// </summary> _Matrix w1_1; /// <summary> /// 保存w2的值 /// </summary> _Matrix w2_1; /// <summary> /// 用于综合评价的矩阵（基于bp神经网络测试结果） /// </summary> _Matrix comprehesiveEvaluationMatrix; /// <summary> /// 综合评价结果输出矩阵（基于bp神经网络测试结果） /// </summary> _Matrix comprehensiveEvaluationResultMatrix; /// <summary> /// 保存b1的值 /// </summary> _Matrix b1_1; /// <summary> /// 保存b2的值 /// </summary> _Matrix b2_1; /// <summary> /// 学习率 /// </summary> double xite; /// <summary> /// 误差 /// </summary> double error; public double[] comprehensiveEvaluation; double accu_average; /// <summary> /// 误差率 /// </summary> double[] accuracy; /// <summary> /// 训练输入数据 /// </summary> _Matrix input_train; public _Matrix Input_train { get { return input_train; } set { this.input_train = value; } } /// <summary> /// 训练输出数据 /// </summary> _Matrix output_train; public _Matrix Output_train { get { return output_train; } set { this.output_train = value; } } /// <summary> /// 归一化后的训练输入数据 /// </summary> _Matrix input_train_Norm; /// <summary> /// 归一化后的训练输出数据 /// </summary> _Matrix output_train_Norm; /// <summary> /// 测试输入数据 /// </summary> _Matrix input_test; public _Matrix Input_test { get { return input_test; } set { this.input_test = value; } } /// <summary> /// 预期输出数据(归一化前) /// </summary> public _Matrix fore_test; /// <summary> /// 预期输出数据(归一化后） /// </summary> public _Matrix fore; /// <summary> /// 测试输出数据 /// </summary> _Matrix output_test; public _Matrix Output_test { get { return output_test; } set { this.output_test = value; } } /// <summary> /// 误差矩阵 /// </summary> _Matrix error_test; /// <summary> /// 归一化后的测试输入数据 /// </summary> _Matrix input_test_Norm; /// <summary> /// 归一化后的测试输出数据 /// </summary> _Matrix output_test_Norm; /// <summary> /// 构造函数 /// </summary> /// <param name="innum"></param> /// <param name="midnum"></param> /// <param name="outnum"></param> /// <param name="num"></param> /// <param name="sampDim"></param> /// <param name="input_train"></param> /// <param name="output_train"></param> /// <param name="xite"></param> public BP(int innum, int midnum, int outnum, int train_num, int sampDim, int iteration, double xite) { this.innum = innum; this.midnum = midnum; this.outnum = outnum; this.iteration = iteration; matrix_Calc = new _Matrix_Calc(); this.train_num = train_num; this.sampdim = sampDim; this.xite = xite; this.input_train = new _Matrix(train_num, sampDim); input_train.init_matrix(); this.output_train = new _Matrix(train_num, outnum); output_train.init_matrix(); //初始化w1,w2,b1,b2; w1 = InitWB(midnum, innum); w2 = InitWB(midnum, outnum); b1 = InitWB(midnum, 1); b2 = InitWB(outnum, 1); w1_1 = new _Matrix(w1); b1_1 = new _Matrix(b1); w2_1 = new _Matrix(w2); b2_1 = new _Matrix(b2); } /// <summary> /// 应用与BP训练时的计算，矩阵的每一个值乘上学习率 /// </summary> /// <param name="data"></param> /// <param name="xite"></param> /// <returns></returns> public _Matrix AddStudyRate(_Matrix data, double xite) { for (int i = 0; i < data.m; i++) { for (int j = 0; j < data.n; j++) { data.write(i, j, data.read(i, j) * xite); } } return data; } /// <summary> /// 归一化 /// </summary> /// <param name="data"></param> /// <returns></returns> public _Matrix Normalize(_Matrix data) //@_@ to do test { _Matrix dat = new _Matrix(data); for (int i = 0; i < dat.m; i++) { double min = 100000.0; double max = -100000.0; for (int j = 0; j < dat.n; j++) { double s = dat.read(i, j); if (s > max) { max = s; } else if (s < min) { min = s; } } for (int j = 0; j < dat.n; j++) { double s = dat.read(i, j); s = (s - min) / (max - min); dat.write(i, j, s); } } return dat; } /// <summary> /// 初始化w1,w2,b1,b2 /// </summary> /// <param name="m"></param> /// <param name="n"></param> /// <returns></returns> public _Matrix InitWB(int m, int n) { _Matrix mat = new _Matrix(m, n); mat.init_matrix(); Random rand = new Random(); for (int i = 0; i < m; i++) { for (int j = 0; j < n; j++) { double s; s = (rand.NextDouble() - 0.5) * 2; mat.write(i, j, s); } } return mat; } /// <summary> /// 获取矩阵的某一行 /// </summary> /// <param name="data"></param> /// <param name="kk"></param> /// <returns></returns> public _Matrix GetRow(_Matrix data, int kk) { _Matrix p = new _Matrix(1, data.n); p.init_matrix(); for (int i = 0; i < data.n; i++) { p.write(0, i, data.read(kk, i)); } return p; } public _Matrix GetColumn(_Matrix data, int kk) { _Matrix p = new _Matrix(data.m, 1); p.init_matrix(); for (int i = 0; i < data.m; i++) { p.write(i, 0, data.read(i, kk)); } return p; } public double sumsqr(_Matrix data) { double s = 0.0; for (int i = 0; i < data.m; i++) { for (int j = 0; j < data.n; j++) { s += data.read(i, j) * data.read(i, j); } } return s; } /// <summary> /// 训练网络 /// </summary> /// <param name="input_train">输入矩阵（可以是未归一化的）</param> /// <param name="output_train">期望的输出矩阵（可以是未归一化的）</param> public void trainBP(_Matrix input_train, _Matrix output_train) { this.input_train = input_train; this.output_train = output_train; this.input_train_Norm = Normalize(input_train); this.output_train_Norm = Normalize(output_train); for (int ii = 0; ii < iteration; ii++) { for (int i = 0; i < train_num; i++) { var x = GetColumn(input_train_Norm, i); _Matrix I = new _Matrix(1, midnum); I.init_matrix(); _Matrix lout = new _Matrix(1, midnum); lout.init_matrix(); for (int j = 0; j < midnum; j++) { _Matrix t = new _Matrix(1, 1); t.init_matrix(); _Matrix aaa = GetColumn(input_train_Norm, i); _Matrix tt = matrix_Calc.transposs(aaa); _Matrix ttt = matrix_Calc.transposs(GetRow(w1, j)); t = matrix_Calc.multiplys(tt, ttt); I.write(0, j, t.read(0, 0) + b1.read(j, 0)); double s = 1 / (1 + Math.Exp(-I.read(0, j))); lout.write(0, j, s); } _Matrix yn; _Matrix y = matrix_Calc.transposs(w2); _Matrix yy = matrix_Calc.transposs(lout); _Matrix yyy = matrix_Calc.multiplys(y, yy); yn = matrix_Calc.adds(yyy, b2); _Matrix e; e = GetColumn(output_train_Norm, i); e = matrix_Calc.subtracts(e, yn); _Matrix dw2; dw2 = matrix_Calc.multiplys(e, lout); _Matrix db2 = matrix_Calc.transposs(e); _Matrix dw1 = new _Matrix(innum, midnum); dw1.init_matrix(); _Matrix db1 = new _Matrix(1, midnum); db1.init_matrix(); double[] FI = new double[midnum]; for (int j = 0; j < midnum; j++) { double S = 1 / (1 + Math.Exp(-I.read(0, j))); FI[j] = S; } for (int k = 0; k < innum; k++) { for (int j = 0; j < midnum; j++) { double s = 0.0; for (int tt = 0; tt < outnum; tt++) { s += e.arr[tt] * w2.read(j, tt); } dw1.write(k, j, FI[j] * x.read(k, 0) * s); db1.write(j, 1, FI[j] * s); } } _Matrix sw1 = matrix_Calc.transposs(dw1); _Matrix sb1 = matrix_Calc.transposs(db1); _Matrix sw2 = matrix_Calc.transposs(dw2); _Matrix sb2 = matrix_Calc.transposs(db2); w1 = matrix_Calc.adds(w1_1, AddStudyRate(sw1, xite)); _Matrix aaaa = AddStudyRate(sb1, xite); b1 = matrix_Calc.adds(b1_1, aaaa); w2 = matrix_Calc.adds(w2_1, AddStudyRate(sw2, xite)); b2 = matrix_Calc.adds(b2_1, AddStudyRate(sb2, xite)); w1_1 = new _Matrix(w1); b1_1 = new _Matrix(b1); w2_1 = new _Matrix(w2); b2_1 = new _Matrix(b2); } } } /// <summary> /// 将测试数据代入进行测试 /// </summary> /// <param name="input_test">测试组的输入数据</param> /// <param name="output_test">测试组的预期输出数据</param> /// <param name="test_num">测试组的组数</param> public void testBP(_Matrix input_test, _Matrix output_test, int test_num) { this.input_test = input_test; this.output_test = output_test; this.input_test_Norm = Normalize(input_test); this.output_test_Norm = Normalize(output_test); fore_test = new _Matrix(output_test.m, output_test.n); fore_test.init_matrix(); error_test = new _Matrix(output_test.m, output_test.n); error_test.init_matrix(); this.test_num = test_num; for (int i = 0; i < test_num; i++) { double[] I = new double[midnum]; _Matrix lout = new _Matrix(1, midnum); lout.init_matrix(); for (int j = 0; j < midnum; j++) { _Matrix s = GetColumn(input_test_Norm, i); s = matrix_Calc.transposs(s); _Matrix ss = GetRow(w1, j); ss = matrix_Calc.transposs(ss); _Matrix sss = matrix_Calc.multiplys(s, ss); I[j] = sss.arr[0] + b1.read(j, 0); lout.write(0, j, 1 / (1 + Math.Exp(-I[j]))); } _Matrix t = matrix_Calc.transposs(w2); _Matrix tt = matrix_Calc.transposs(lout); _Matrix ttt = matrix_Calc.adds(matrix_Calc.multiplys(t, tt), b2); for (int j = 0; j < fore_test.m; j++) { fore_test.write(j, i, ttt.read(j, 0)); } } error_test = matrix_Calc.subtracts(fore_test, output_test_Norm); error = sumsqr(error_test); Console.WriteLine(error); } /// <summary> /// 获得综合评价矩阵 /// </summary> /// <param name="num">综合评价矩阵的维度</param> public void GetComprehensiveEvaluationMatrix(int num) { if (output_test_Norm.arr==null) { return; } output_test_Norm = matrix_Calc.transposs(output_test_Norm); fore_test = matrix_Calc.transposs(fore_test); comprehesiveEvaluationMatrix = new _Matrix(output_test_Norm.m,2*num); comprehesiveEvaluationMatrix.init_matrix(); for (int i = 0; i < comprehesiveEvaluationMatrix.m; i++) { comprehesiveEvaluationMatrix.write(i,0,output_test_Norm.read(i,0)); double s = 0.0; s = output_test_Norm.read(i,1)+output_test_Norm.read(i,2)+output_test_Norm.read(i,3)+output_test_Norm.read(i,4)+output_test_Norm.read(i,5); s = s / 5; comprehesiveEvaluationMatrix.write(i, 1, s); comprehesiveEvaluationMatrix.write(i, 2, output_test_Norm.read(i, 6)); comprehesiveEvaluationMatrix.write(i, 3, output_test_Norm.read(i, 7)); comprehesiveEvaluationMatrix.write(i, 4, fore_test.read(i, 0)); s = fore_test.read(i, 1) + fore_test.read(i, 2) + fore_test.read(i, 3) + fore_test.read(i, 4) + fore_test.read(i, 5); s = s / 5; comprehesiveEvaluationMatrix.write(i, 5, s); comprehesiveEvaluationMatrix.write(i, 6, fore_test.read(i, 6)); comprehesiveEvaluationMatrix.write(i, 7, fore_test.read(i, 7)); } } /// <summary> /// 进行综合评价,获得综合评价后的结果矩阵 /// </summary> /// <param name="ComEval">各维度权值</param> /// <param name="data">评价矩阵</param> public void ComEvaluation(double [] ComEval) { comprehensiveEvaluationResultMatrix = new _Matrix(comprehesiveEvaluationMatrix.m,3); comprehensiveEvaluationResultMatrix.init_matrix(); _Matrix data = new _Matrix(comprehesiveEvaluationMatrix); for (int i = 0; i < data.m; i++) { double s = 0.0; for (int j = 0; j < data.n/2; j++) { s += ComEval[j] * data.read(i,j); } comprehensiveEvaluationResultMatrix.write(i,0,i+1); comprehensiveEvaluationResultMatrix.write(i,1,s); s = 0.0; for (int j = data.n/2; j < data.n; j++) { s += ComEval[j-data.n/2] * data.read(i, j); } comprehensiveEvaluationResultMatrix.write(i, 2, s); } } /// <summary> /// ComEvaResult矩阵写入EXCEL /// </summary> public void ComEvaResult_Excel() { if (comprehensiveEvaluationResultMatrix.arr==null) { return; } var excelApp = new Microsoft.Office.Interop.Excel.Application(); Workbooks workbooks = excelApp.Workbooks; Workbook workBook = workbooks.Add(Type.Missing); Worksheet workSheet = (Worksheet)workBook.Worksheets[1];//取得sheet1 for (int i = 1; i <=comprehensiveEvaluationResultMatrix.m; i++) { for (int j = 1; j <=comprehensiveEvaluationResultMatrix.n; j++) { workSheet.Cells[i, j] = comprehensiveEvaluationResultMatrix.read(i-1,j-1); } } workBook.SaveAs(@"d:\comEvaResult.xlsx", Type.Missing, Type.Missing, Type.Missing, Type.Missing, Type.Missing, Excel.XlSaveAsAccessMode.xlNoChange, Type.Missing, Type.Missing, Type.Missing, Type.Missing, Type.Missing); workbooks.Close(); } /// <summary> /// 把求得的W,B,w,b /// </summary> public void WB_Excel() { var excelApp = new Microsoft.Office.Interop.Excel.Application(); Workbooks workbooks = excelApp.Workbooks; Workbook workBook = workbooks.Add(Type.Missing); Worksheet workSheet = (Worksheet)workBook.Worksheets[1];//取得sheet1 workSheet.Name = "w1"; for (int i = 1; i <= this.w1.m; i++) { for (int j = 1; j <= this.w1.n; j++) { workSheet.Cells[i, j] = this.w1.read(i - 1, j - 1); } } workBook.Worksheets.Add(Type.Missing,Type.Missing,Type.Missing,Type.Missing); workSheet = (Worksheet)workBook.Worksheets[1]; workSheet.Name = "b1"; for (int i = 1; i <= this.b1.m; i++) { for (int j = 1; j <=b1.n; j++) { workSheet.Cells[i, j] = this.b1.read(i-1,j-1); } } workBook.Worksheets.Add(Type.Missing, Type.Missing, Type.Missing, Type.Missing); workSheet = (Worksheet)workBook.Worksheets[1]; workSheet.Name = "w2"; for (int i = 1; i <= this.w2.m; i++) { for (int j = 1; j <= w2.n; j++) { workSheet.Cells[i, j] = this.w2.read(i - 1, j - 1); } } workBook.Worksheets.Add(Type.Missing, Type.Missing, Type.Missing, Type.Missing); workSheet = (Worksheet)workBook.Worksheets[1]; workSheet.Name = "b2"; for (int i = 1; i <= this.b2.m; i++) { for (int j = 1; j <= b2.n; j++) { workSheet.Cells[i, j] = this.b2.read(i - 1, j - 1); } } workBook.SaveAs(@"d:\saveWB.xlsx", Type.Missing, Type.Missing, Type.Missing, Type.Missing, Type.Missing, Excel.XlSaveAsAccessMode.xlNoChange, Type.Missing, Type.Missing, Type.Missing, Type.Missing, Type.Missing); workbooks.Close(); } /// <summary> /// 输出结果写入Excel /// </summary> public void output_Excel() { var excelApp = new Microsoft.Office.Interop.Excel.Application(); Workbooks workbooks = excelApp.Workbooks; Workbook workBook = workbooks.Add(Type.Missing); Worksheet workSheet = (Worksheet)workBook.Worksheets[1];//取得sheet1 for (int j = 1; j < 9; j++) workSheet.Cells[1, j] = accuracy[j - 1]; workSheet.Cells[2, 1] = accu_average; workBook.SaveAs(@"d:\result.xlsx", Type.Missing, Type.Missing, Type.Missing, Type.Missing, Type.Missing, Excel.XlSaveAsAccessMode.xlNoChange, Type.Missing, Type.Missing, Type.Missing, Type.Missing, Type.Missing); workbooks.Close(); } /// <summary> /// 反归一化获得输出结果 /// </summary> public void ConvNorm() { fore = matrix_Calc.transposs(this.fore_test); _Matrix output = matrix_Calc.transposs(this.Output_train); for (int i = 0; i < output.n; i++) { double max = -100000.0; double min = 100000.0; for (int j = 0; j < output.m; j++) { if (max < output.read(j, i)) { max = output.read(j, i); } else if (min > output.read(j, i)) { min = output.read(j, i); } } for (int j = 0; j < fore.m; j++) { double s = (max - min) * fore.read(j, i) + min; fore.write(j, i, s); } } } public void CalcAccuracy() { accuracy = new double[outnum]; _Matrix output = matrix_Calc.transposs(output_train); accu_average = 0.0; for (int i = 0; i < outnum; i++) { double accu = 0.0; for (int j = 0; j < test_num; j++) { accu += Math.Abs(fore.read(j, i) - output.read(j, i)) / output.read(j, i); accu_average += Math.Abs(fore.read(j, i) - output.read(j, i)) / output.read(j, i); } accuracy[i] = accu / 180; accu_average = accu_average / (outnum) / (test_num); } } }}主程序：
using NPOI.HSSF.UserModel;using System;using System.Collections.Generic;using System.IO;using System.Linq;using System.Text;using System.Threading.Tasks;using Matrix_Mul;using Excel=Microsoft.Office.Interop.Excel;using System.Reflection;namespace BPNETSerial{ class Program { static void Main(string[] args) { //初始化bp神经网络 BP bp = new BP(9,15,8,1620,9,300,0.2); //创建一个mat来便于对_Matrix类进行计算 _Matrix_Calc mat = new _Matrix_Calc(); using (FileStream stream = new FileStream(@"train.xls", FileMode.Open, FileAccess.Read)) { HSSFWorkbook Workbook = new HSSFWorkbook(stream); var Sheet = Workbook.GetSheetAt(0); int j = 0; for (int i = 1; i < 1621; i++) { var row = Sheet.GetRow(i); for (int k = 0; k < row.Cells.Count; k++) { bp.Input_train.arr[j++] = row.GetCell(k).NumericCellValue; } } } using (FileStream stream = new FileStream(@"train.xls", FileMode.Open, FileAccess.Read)) { HSSFWorkbook Workbook = new HSSFWorkbook(stream); var Sheet = Workbook.GetSheetAt(1); int j = 0; for (int i = 1; i < 1621; i++) { var row = Sheet.GetRow(i); for (int k = 0; k < row.Cells.Count; k++) { bp.Output_train.arr[j++] = row.GetCell(k).NumericCellValue; } } } using (FileStream stream = new FileStream(@"train.xls", FileMode.Open, FileAccess.Read)) { bp.Test_num = 180; bp.Input_test = new _Matrix(180,9); bp.Input_test.init_matrix(); HSSFWorkbook Workbook = new HSSFWorkbook(stream); var Sheet = Workbook.GetSheetAt(2); int j = 0; for (int i = 1; i < 181; i++) { var row = Sheet.GetRow(i); for (int k = 0; k < row.Cells.Count; k++) { bp.Input_test.arr[j++] = row.GetCell(k).NumericCellValue; } } } using (FileStream stream = new FileStream(@"train.xls", FileMode.Open, FileAccess.Read)) { bp.Output_test = new _Matrix(180, 8); bp.Output_test.init_matrix(); HSSFWorkbook Workbook = new HSSFWorkbook(stream); var Sheet = Workbook.GetSheetAt(3); int j = 0; for (int i = 1; i < 181; i++) { var row = Sheet.GetRow(i); for (int k = 0; k < row.Cells.Count; k++) { bp.Output_test.arr[j++] = row.GetCell(k).NumericCellValue; } } } bp.Input_train = mat.transposs(bp.Input_train); _Matrix Output_train = new _Matrix(bp.Output_train); bp.Output_train = mat.transposs(bp.Output_train); bp.Input_test = mat.transposs(bp.Input_test); bp.Output_test = mat.transposs(bp.Output_test); bp.trainBP(bp.Input_train,bp.Output_train); bp.testBP(bp.Input_test,bp.Output_test,180); //以下的代码均是为了测试网络准确度和实现综合评价所写 //bp.ConvNorm(); //bp.CalcAccuracy(); //bp.GetComprehensiveEvaluationMatrix(4); //bp.comprehensiveEvaluation = new double[4] {0.3309,0.2201,0.2696,0.1793 }; //bp.ComEvaluation(bp.comprehensiveEvaluation); //bp.ComEvaResult_Excel(); //bp.WB_Excel(); //bp.output_Excel(); } }}其中数据来自来自Minifab(Minifab是针对Intel公司的半导体生产线提炼出的用于研究的调度仿真模型，在机器数和加工步数都较少的情况下，能够反映半导体生产线的一些本质问题，满足研究的需要。)
训练输入数据是1620组9维数据，输出是1620组8维数据；测试输入数据为180组9维数据，输出是180组8维数据；训练后总误差为3.4%项目源代码：http://download.csdn.net/detail/lanqiuchaoren/7628945
0 0
- C#实现的bp神经网络并应用于综合评价
- BP神经网络的实现
- BP神经网络的实现
- 基于模糊神经网络的综合评价
- BP神经网络的Java实现
- bp神经网络的java实现
- BP神经网络的C++实现
- 神经网络BP的简单实现
- BP神经网络的Python实现
- BP神经网络的简单实现
- 简单的BP神经网络实现
- BP神经网络的matlab实现
- BP神经网络的Python实现
- bp神经网络的python实现
- BP神经网络的MATLAB实现
- BP神经网络的实现详解
- BP神经网络的matlab实现
- BP神经网络及其在教学质量评价中 的应用
- OpenVPN多处理之-多队列TUN多线程
- poj3979
- 跳过权限表启动mysql不用密码
- android App Widgets
- Unity3D 声音格式及导入
- C#实现的bp神经网络并应用于综合评价
- No protocol specified
- OpenUI5 - SAP开源中的移动大战略
- 编程之美hiho一下求最长回文串的长度
- 基于IOS的FTP详解(五)删除文件或者目录
- R语言设置 坐标轴刻度
- ubuntu 安装五笔拼音输入法
- 为什么说IP和UDP是不可靠的,而TCP是可靠的
- POJ 3461 Oulipo KMP算法题解
