遗传算法优化神经网络matlab代码

以下是一个遗传算法优化神经网络的 MATLAB 代码示例：

% 首先，定义神经网络模型结构和初始权重
input_size = 10;
hidden_size = 20;
output_size = 5;
W1 = randn(input_size, hidden_size);
W2 = randn(hidden_size, output_size);

% 然后，定义适应度函数（即神经网络的性能评估函数）
function fitness = nn_fitness(W1, W2, X, Y)
    % 基于给定权重计算神经网络输出
    Z = X * W1;
    H = tanh(Z);
    Y_hat = H * W2;

    % 计算误差
    error = Y_hat - Y;
    squared_error = error .^ 2;
    fitness = -mean(squared_error(:)); % 适应度越高，误差越小
end

% 定义遗传算法参数
population_size = 100;
mutation_rate = 0.01;
num_generations = 50;

% 生成初始种群
population = cell(population_size, 1);
for i = 1:population_size
    population{i}.W1 = randn(input_size, hidden_size);
    population{i}.W2 = randn(hidden_size, output_size);
end

% 迭代遗传算法
for generation = 1:num_generations
    % 计算每个个体的适应度
    fitness_scores = zeros(population_size, 1);
    for i = 1:population_size
        fitness_scores(i) = nn_fitness(population{i}.W1, population{i}.W2, X, Y);
    end

    % 选择并复制适应度高的个体
    sorted_indices = sortrows([(1:population_size)', fitness_scores], -2);
    elite_population = population(sorted_indices(1:ceil(population_size/2)), :);
    new_population = cell(population_size, 1);
    for i = 1:ceil(population_size/2)
        new_population{i} = elite_population{i};
    end

    % 交叉和变异生成新的个体
    for i = ceil(population_size/2)+1:population_size
        parent1 = elite_population{randi(length(elite_population))};
        parent2 = elite_population{randi(length(elite_population))};
        child = struct();
        for field = fieldnames(parent1)'
            field = field{1};
            if rand() < 0.5
                child.(field) = parent1.(field);
            else
                child.(field) = parent2.(field);
            end
            if rand() < mutation_rate
                child.(field) = child.(field) + randn(size(child.(field))) * 0.1;
            end
        end
        new_population{i} = child;
    end

    % 更新种群
    population = new_population;
end

% 选择最佳个体
fitness_scores = zeros(population_size, 1);
for i = 1:population_size
    fitness_scores(i) = nn_fitness(population{i}.W1, population{i}.W2, X, Y);
end
[best_fitness, best_index] = max(fitness_scores);
best_individual = population{best_index};

% 使用最佳个体的权重计算神经网络输出
Z = X * best_individual.W1;
H = tanh(Z);
Y_hat = H * best_individual.W2;

这个代码示例使用遗传算法优化了一个拥有一个隐含层的前馈神经网络，以最小化它在给定数据集上的均方误差。其中，`X` 和 `Y` 分别是输入和输出数据，`population_size` 表示种群大小，`mutation_rate` 表示变异率，`num_generations` 表示迭代代数。每代遗传算法中，首先计算每个个体的适应度，然后选择并复制适应度高的个体，接着使用交叉和变异生成新的个体，最后更新种群。最终，选择最佳个体，并使用它的权重计算神经网络输出。