C#Mathnet最小二乘法拟合自定义指数曲线

mengyingguxing

已于 2023-11-01 14:33:04 修改

阅读量2k

点赞数 1

文章标签：最小二乘法算法人工智能

于 2023-11-01 14:30:10 首次发布

本文链接：https://blog.csdn.net/mengyingguxing/article/details/134160134

版权

mathnet自带了Fit.Exponential来拟合指数曲线，但是拟合的效果跟Excel拟合差不多，往往这种程度的达不到项目的要求，需要使用Levenberg-Marquardt算法（LevenbergMarquardtSolver）或非线性最小二乘算法（NonlinearLeastSquares）来获得更好的拟合结果。

 // 定义我们的函数
        // f(x; A1, B1,x0)=A1*e^(B1(x-x0))
        private Vector<double> MyModel(Vector<double> p, Vector<double> x)
        {
            var y = CreateVector.Dense<double>(x.Count);
            var A1 = p[0];
            var B1 = p[1];
            var x0 = p[2];
            for (int i = 0; i < x.Count; i++)
            {
                y[i] = A1 * Math.Exp((B1 * (x[i] - x0)));
            }
            return y;
        }
        //derivatives:求导数
        //       df/A1 = e^(B1*x-x0)
        //       df/B1 = A1*e^(B1*x-x0)*(x-x0)
        //       df/x0 = A1*e^(B1*x-x0)*(-B1)
        private Matrix<double> MyPrime(Vector<double> p, Vector<double> x)
        {
            var prime = Matrix<double>.Build.Dense(x.Count, p.Count);
            var A1 = p[0];
            var B1 = p[1];
            var x0 = p[2];
            for (int i = 0; i < x.Count; i++)
            {
                prime[i, 0] = Math.Exp(B1 * (x[i] - x0));
                prime[i, 1] = A1 * Math.Exp(B1 * (x[i] - x0)) * (x[i] - x0);
                prime[i, 2] = A1 * Math.Exp(B1 * (x[i] - x0)) * (-B1);
            }
            return prime;
        }


        private Vector<double> MyStart = new DenseVector(new double[] { 100, 0.01, 1 });

        private void btnFit_Click(object sender, EventArgs e)
        {
            var xData = data.Keys.ToArray();
            var yData = data.Values.ToArray();
            var result1 = Fit.Exponential(xData, yData);
            double goodnessOfFit1 = GoodnessOfFit.RSquared(xData.Select(x => result1.A * Math.Exp(result1.R * x)), yData);

            var obj = ObjectiveFunction.NonlinearModel(MyModel, MyPrime, new DenseVector(xData), new DenseVector(yData));
            var solver = new LevenbergMarquardtMinimizer();
            var result2 = solver.FindMinimum(obj, MyStart);
            double goodnessOfFit2 = GoodnessOfFit.RSquared(xData.Select(x => result2.MinimizingPoint[0] * Math.Exp((result2.MinimizingPoint[1] * (x - result2.MinimizingPoint[2])))), yData);

            // 输出结果

            Console.WriteLine("goodnessOfFit1:" + goodnessOfFit1);
            Console.WriteLine("goodnessOfFit2:" + goodnessOfFit2);


        }

自定义的拟合度相对较高