Matlab实现粒子群优化神经网络算法

以下是Matlab实现粒子群优化神经网络算法的示例代码：

% 设置神经网络参数
inputSize = 3;
hiddenSize = 5;
outputSize = 1;

% 初始化粒子群参数
numParticles = 10;
maxIterations = 100;
c1 = 1.5;
c2 = 1.5;
w = 0.8;

% 生成粒子群
particlePositions = rand(numParticles, hiddenSize * (inputSize + outputSize) + outputSize);
particleVelocities = zeros(numParticles, size(particlePositions, 2));
particleBestPositions = particlePositions;
particleBestErrors = inf(numParticles, 1);
globalBestPosition = particlePositions(1, :);
globalBestError = inf;

% 设置训练数据
X = [0 0 1; 0 1 1; 1 0 1; 1 1 1];
Y = [0; 1; 1; 0];

% 训练神经网络
for i = 1:maxIterations
for j = 1:numParticles
% 将粒子的位置转换为神经网络的权重和偏置
particleWeights1 = reshape(particlePositions(j, 1:hiddenSize * inputSize), inputSize, hiddenSize);
particleBias1 = reshape(particlePositions(j, hiddenSize * inputSize + 1:hiddenSize * (inputSize + 1)), 1, hiddenSize);
particleWeights2 = reshape(particlePositions(j, hiddenSize * (inputSize + 1) + 1:end - outputSize), hiddenSize, outputSize);
particleBias2 = particlePositions(j, end - outputSize + 1:end);

% 计算粒子的预测结果和误差
[Ypred, ~] = feedforward(particleWeights1, particleBias1, particleWeights2, particleBias2, X);
error = loss(Ypred, Y);

% 更新粒子的最佳位置和最小误差
if error < particleBestErrors(j)
particleBestPositions(j, :) = particlePositions(j, :);
particleBestErrors(j) = error;
end

% 更新全局最佳位置和最小误差
if error < globalBestError
globalBestPosition = particlePositions(j, :);
globalBestError = error;
end

% 更新粒子的速度和位置
particleVelocities(j, :) = w * particleVelocities(j, :) ...
+ c1 * rand(1, size(particlePositions, 2)) .* (particleBestPositions(j, :) - particlePositions(j, :)) ...
+ c2 * rand(1, size(particlePositions, 2)) .* (globalBestPosition - particlePositions(j, :));
particlePositions(j, :) = particlePositions(j, :) + particleVelocities(j, :);
end
end

% 将全局最佳位置转换为神经网络的权重和偏置
globalBestWeights1 = reshape(globalBestPosition(1:hiddenSize * inputSize), inputSize, hiddenSize);
globalBestBias1 = reshape(globalBestPosition(hiddenSize * inputSize + 1:hiddenSize * (inputSize + 1)), 1, hiddenSize);
globalBestWeights2 = reshape(globalBestPosition(hiddenSize * (inputSize + 1) + 1:end - outputSize), hiddenSize, outputSize);
globalBestBias2 = globalBestPosition(end - outputSize + 1:end);

% 计算全局最佳预测结果和误差
[Ypred, ~] = feedforward(globalBestWeights1, globalBestBias1, globalBestWeights2, globalBestBias2, X);
error = loss(Ypred, Y);

% 输出结果
disp(['Global best error: ' num2str(error)]);
disp(['Global best weights and biases: ' num2str(globalBestPosition)]);

% 定义神经网络正向传播函数
function [Ypred, Z] = feedforward(weights1, bias1, weights2, bias2, X)
Z = tanh(X * weights1 + bias1);
Ypred = Z * weights2 + bias2;
end

% 定义损失函数
function error = loss(Ypred, Y)
error = sum((Ypred - Y) .^ 2) / size(Y, 1);
end