如何在Matlab中计算两个正态分布的交点？（用于检测概率计算）

Got it, let's tackle this problem step by step. Since you're working on calculating false alarm and detection probabilities for a detector, finding the intersection of two normal PDFs is key—and you already tried symbolic computation per Ander Biguri's suggestion, so let's build on that and cover other reliable methods in MATLAB.

1. Refined Symbolic Computation #

Since you already started with symbolic methods, here's a more structured approach that focuses on real-valued solutions and lets you plug in your actual parameters easily:

% Define symbolic variables for parameters and the variable x
syms x mu0 sigma0 mu1 sigma1 real;

% Use MATLAB's built-in symbolic normpdf to define the two distributions
f0 = normpdf(x, mu0, sigma0);
f1 = normpdf(x, mu1, sigma1);

% Set up the equality equation and solve for real x values
eq = f0 == f1;
solutions = solve(eq, x, 'Real', true);

% Substitute your actual H0/H1 parameters (replace with your values)
param_map = {mu0, sigma0, mu1, sigma1};
actual_params = {0, 1, 2, 1.5}; 
numeric_solutions = subs(solutions, param_map, actual_params);

% Convert symbolic results to numeric values for further calculations
numeric_solutions = double(numeric_solutions);

Depending on your parameters, you'll get 0, 1, or 2 real solutions (or infinitely many if the two distributions are identical).

2. Numerical Root-Finding with fzero #

If symbolic computation feels slow or runs into edge cases with complex parameters, numerical methods are a robust fallback. We'll use fzero to find roots of the difference between the two PDFs:

% Define your H0 and H1 parameters
mu0 = 0; sigma0 = 1;
mu1 = 2; sigma1 = 1.5;

% Create a function for the difference between the two PDFs
pdf_diff = @(x) normpdf(x, mu0, sigma0) - normpdf(x, mu1, sigma1);

% First, plot the PDFs to guess initial points (critical for fzero to converge!)
x_range = linspace(min(mu0, mu1)-3*max(sigma0, sigma1), max(mu0, mu1)+3*max(sigma0, sigma1), 1000);
plot(x_range, normpdf(x_range, mu0, sigma0), x_range, normpdf(x_range, mu1, sigma1));
legend('H0 PDF', 'H1 PDF');

% Pick initial guesses based on the plot
% For the intersection between the two means
guess1 = (mu0 + mu1)/2;
intersection1 = fzero(pdf_diff, guess1);

% For a second intersection (if it exists, e.g., left of mu0)
guess2 = mu0 - 3*sigma0;
intersection2 = fzero(pdf_diff, guess2);

% Collect valid intersections (remove NaNs if no second root exists)
intersections = [intersection1, intersection2];
intersections = intersections(~isnan(intersections));

The plot step is essential—fzero relies on a good initial guess to find the correct root.

3. Analytical Formula (Fastest Option) #

You can derive a closed-form solution by taking the logarithm of both PDFs and simplifying the equality. This avoids symbolic/numerical overhead entirely. Here's a reusable MATLAB function implementing this:

function intersections = norm_pdf_intersections(mu0, sigma0, mu1, sigma1)
    % Calculate coefficients of the quadratic equation derived from f0(x) = f1(x)
    a = 1/(2*sigma0^2) - 1/(2*sigma1^2);
    b = mu1/(sigma1^2) - mu0/(sigma0^2);
    c = (mu0^2)/(2*sigma0^2) - (mu1^2)/(2*sigma1^2) + log(sigma0/sigma1);
    
    intersections = [];
    
    if abs(a) < eps % Case: variances are equal
        if abs(mu0 - mu1) < eps % Distributions are identical
            warning('Distributions are identical—infinitely many intersections');
        else
            % Single intersection at the midpoint of the two means
            intersections = (mu0 + mu1)/2;
        end
    else
        discriminant = b^2 - 4*a*c;
        if discriminant < 0 % No real intersections
            return;
        elseif discriminant == 0 % One real intersection
            intersections = -b/(2*a);
        else % Two real intersections
            sqrt_d = sqrt(discriminant);
            x1 = (-b + sqrt_d)/(2*a);
            x2 = (-b - sqrt_d)/(2*a);
            intersections = sort([x1, x2]); % Sort for consistency
        end
    end
end

Use it like this:

mu0 = 0; sigma0 = 1;
mu1 = 2; sigma1 = 1.5;
ints = norm_pdf_intersections(mu0, sigma0, mu1, sigma1);

Bonus: Calculate False Alarm and Detection Probabilities #

Once you have your threshold (the relevant intersection point), compute the probabilities directly with normcdf:

threshold = ints(1); % Adjust the index if you need the other intersection
false_alarm_prob = 1 - normcdf(threshold, mu0, sigma0); % H0 right tail probability
detection_prob = 1 - normcdf(threshold, mu1, sigma1); % H1 right tail probability

内容的提问来源于stack exchange，提问作者Pep