错误的 pgfplots 线性回归计算

Question

我尝试绘制以下数据：

\pgfplotstableread[col sep=semicolon]{
record;value
2004;48
2009;34
2010;32
2016;22
2020;21
}\valuetotest

经过

\pgfkeys{
/pgf/number format/.cd,
set decimal separator={,{\!}},
set thousands separator={}}
\begin{tikzpicture}[]
\begin{axis}[
grid=both, %none
major grid style={thick},
xtick distance={5},
ytick distance={10},
minor x tick num=4,
minor y tick num=4,
width=13.5cm, height=13cm,
axis x line=middle, 
axis y line=middle, 
samples=100,
ymin=-0.2, ymax=60.5,
xmin=2000, xmax=2025,
domain=2000:2025,
xlabel={Years},
ylabel={$\mu g/m^3$}
]
\addplot[only marks,mark=square*,blue] table[x=record,y=value] {\valuetotest};
    \addlegendentry{values}
\addplot[no marks,blue,dashed,] table[header=false,x=record,y={create col/linear regression={y=value}}] {\valuetotest};
    \addlegendentry{pgfplots computed lin regr}
\addplot[smooth,orange,domain=2000:2025]{\pgfplotstableregressiona*x+\pgfplotstableregressionb};
    \addlegendentry{a =\pgfmathprintnumber[]{\pgfplotstableregressiona}; b=\pgfmathprintnumber[]{\pgfplotstableregressionb}}
\addplot[smooth,blue,domain=2000:2025]{-1.674744898*x+3400.651786};
    \addlegendentry{manual lin regr}
\end{axis}

\end{tikzpicture}

但如图所示，pgfplots 似乎计算出了错误的线性回归值（我用 excel 计算了“手动值”）。

提前感谢