当前位置：首页 > news >正文

基于MATLAB实现的Elman神经网络用于电力负载预测

news 2025/11/12 6:28:04

1. 数据准备

% 假设已经加载了历史电力负载数据和相关影响因子
% 数据文件路径
dataFile = 'path_to_load_data.mat'; % 替换为实际数据文件路径% 加载数据
data = load(dataFile);
loadData = data.loadData; % 历史电力负载数据
tempData = data.tempData; % 平均温度数据
weatherData = data.weatherData; % 天气类型数据
dateTypeData = data.dateTypeData; % 日期类型数据

2. 数据预处理

% 数据归一化
loadDataNorm = (loadData - min(loadData)) / (max(loadData) - min(loadData));
tempDataNorm = (tempData - min(tempData)) / (max(tempData) - min(tempData));% 构建训练数据集
% 假设使用前90%的数据作为训练集，后10%作为测试集
trainRatio = 0.9;
trainSize = round(size(loadDataNorm, 1) * trainRatio);
trainData = loadDataNorm(1:trainSize, :);
trainTemp = tempDataNorm(1:trainSize, :);
trainWeather = weatherData(1:trainSize, :);
trainDateType = dateTypeData(1:trainSize, :);testData = loadDataNorm(trainSize+1:end, :);
testTemp = tempDataNorm(trainSize+1:end, :);
testWeather = weatherData(trainSize+1:end, :);
testDateType = dateTypeData(trainSize+1:end, :);

3. 构建Elman神经网络

% 定义Elman神经网络结构
inputSize = size(trainData, 2); % 输入层大小
hiddenSize = 10; % 隐藏层大小
outputSize = size(trainData, 2); % 输出层大小% 初始化权重
w1 = randn(hiddenSize, inputSize + hiddenSize); % 输入层到隐藏层的权重
w2 = randn(outputSize, hiddenSize); % 隐藏层到输出层的权重
b1 = randn(hiddenSize, 1); % 隐藏层偏置
b2 = randn(outputSize, 1); % 输出层偏置% 训练Elman神经网络
numEpochs = 100; % 训练轮数
learningRate = 0.01; % 学习率for epoch = 1:numEpochsfor i = 1:size(trainData, 1)% 获取当前样本x = trainData(i, :);temp = trainTemp(i, :);weather = trainWeather(i, :);dateType = trainDateType(i, :);% 构建输入向量inputVector = [x; temp; weather; dateType];% 前向传播hiddenInput = w1 * inputVector + b1;hiddenOutput = 1 ./ (1 + exp(-hiddenInput)); % Sigmoid激活函数outputInput = w2 * hiddenOutput + b2;output = outputInput; % 线性激活函数% 计算误差error = trainData(i, :) - output;% 反向传播deltaOutput = error;deltaHidden = (1 - hiddenOutput) .* hiddenOutput .* (w2' * deltaOutput);% 更新权重w2 = w2 + learningRate * deltaOutput * hiddenOutput';b2 = b2 + learningRate * deltaOutput;w1 = w1 + learningRate * deltaHidden * inputVector';b1 = b1 + learningRate * deltaHidden;end
end

4. 负载预测

% 使用训练好的Elman神经网络进行预测
predictions = zeros(size(testData, 1), size(testData, 2));
for i = 1:size(testData, 1)% 获取当前样本x = testData(i, :);temp = testTemp(i, :);weather = testWeather(i, :);dateType = testDateType(i, :);% 构建输入向量inputVector = [x; temp; weather; dateType];% 前向传播hiddenInput = w1 * inputVector + b1;hiddenOutput = 1 ./ (1 + exp(-hiddenInput)); % Sigmoid激活函数outputInput = w2 * hiddenOutput + b2;output = outputInput; % 线性激活函数% 保存预测结果predictions(i, :) = output;
end

5. 评估预测结果

% 计算预测误差
error = testData - predictions;
mseError = mean(error .^ 2);% 显示预测结果
disp('预测误差：');
disp(mseError);% 绘制预测结果与真实数据的对比图
figure;
plot(1:size(testData, 1), testData, 'b', 'DisplayName', '真实负载');
hold on;
plot(1:size(predictions, 1), predictions, 'r--', 'DisplayName', '预测负载');
legend('show');
xlabel('时间');
ylabel('负载');
title('电力负载预测结果');